Formats

Use this reference to read agent6's journal, CLI JSON, manifest and HTTP payloads. Field names come from the serializers, dataclasses and models. Meanings come from their docstrings and descriptions. JSON means a dictionary builder has no narrower declared field type. Python tuples serialize as JSON arrays, None as null, and paths as strings. Nested dictionary keys use dots and array items use [].

Support levels

Data contracts defines the support levels.

Formats are not stable until 1.0.

• Durable but private: snapshots preserve run history; transcripts retain diagnostic history.
• Disposable: caches and live coordination files.
• Config TOML is public; Configuration owns its generated setting reference.
• Machine definitions are public; see Agent state machines.

Journal

logs.jsonl contains one JSON object per line. A torn last line is dropped the next time the journal opens for appending. The envelope below surrounds each event's payload. The journal records describe current writers; historical reads tolerate unknown events. Optional event fields may be absent or null. Integer values also satisfy a record's float field, but booleans do not.

Review targets retain snapshot_commit as the cloneable commit for the exact head_tree. snapshot_ref names its session-owned hidden ref; hidden_paths and paths keep their coverage. It defaults empty on historical records, whose unanchored trees may be unavailable. The ref also retains the base tree and is removed with its session, not by chain pruning.

logs.jsonl event envelope

Support: public. Append one record under its envelope (ENVELOPE_KEYS), then notify the listeners.

Source: sessions.event_log.EventSink.emit.

Field	Type	Meaning
ts	JSON	The write time in UTC ISO 8601 with microseconds.
type	JSON	The record's TYPE.
event_id	JSON	A UUID hex identifier on durable events; absent on streaming deltas.

approval.answer

Support: public. The operator answered an approval request.

Field	Type	Meaning
id	str	Approval prompt identifier.
approved	bool	Whether the operator approved the request.
source	str	Where the request or result came from.

approval.prompt

Support: public. A command is waiting for the operator's approval.

Field	Type	Meaning
id	str	Approval prompt identifier.
prompt	str	The words presented to the operator.
standing	bool	Whether approval can cover later calls in this scope.
scope	str, optional	What a standing answer covers: command or mcp.; null for none.
call_id	int, optional	The tool call correlated with this event.

btw.answered

Support: public. A side question received an answer.

Field	Type	Meaning
btw_id	str	Side-question identifier.
block	str	The rendered question and answer block.

btw.opened

Support: public. The operator opened a side question.

Field	Type	Meaning
btw_id	str	Side-question identifier.
question	str	The question presented to the operator.

budget.turn

Support: public. A turn's usage heartbeat before its model call.

Field	Type	Meaning
turn	int	The execution's turn number.
input_tokens	int	Input tokens consumed so far.
output_tokens	int	Output token count.
cache_read_tokens	int	Cached input tokens consumed so far.
cost_usd	float	Estimated spend in USD.

budget.update

Support: public. The tracker's accounting since this process began, its caps and the account's reading.

Field	Type	Meaning
settled_children	list, optional	Child session IDs included in this execution's cumulative spend.
usd_total	float	Cumulative USD spend across provider/model routes, including purchased credits.
input_total	int, optional	Cumulative input tokens.
output_total	int, optional	Cumulative output tokens.
cache_read_total	int, optional	Cumulative cached input tokens.
cache_write_total	int, optional	Cumulative cache-write tokens.
usd_partial	bool, optional	Whether unpriced usage makes spend a lower bound.
usd_cap	float, optional	Spend cap in USD; -1 means unlimited.
usd_running_children	float, optional	USD the children still running have spent so far; outside usd_total until each settles, inside the cap.
tokens_unmetered	int, optional	Tokens with no known price.
tokens_running_children	int, optional	Unpriced tokens of the children still running; outside tokens_unmetered.
tokens_fallback_cap	int, optional	Token cap used for unpriced usage.
plan_observed_points	float, optional	The largest observed plan movement in points across providers and windows; account-wide, so includes other consumers.
plan_cap	float, optional	Subscription points cap.
account	object, optional	The latest subscription account reading; null before a plan call runs.
account.provider	str	The provider entry whose account was read.
account.used_percent	float	The binding window's used percent.
account.resets_at	float	When the binding window resets, as epoch seconds; 0 when unknown.
account.windows	list	Every window the backend reported, the primary first.
account.windows[].name	str	The backend's label: primary, secondary, or a per-model family.
account.windows[].used_percent	float	Percent of the included allowance used.
account.windows[].window_minutes	int	The window's length; 0 when the backend gave none.
account.windows[].resets_at	float	When the window resets, as epoch seconds.
account.credits_has	bool	Whether the account holds purchased credits.
account.credits_unlimited	bool	Whether the credits are unlimited.
account.credits_balance	str	The credit balance as the backend sent it; empty when none.

checkpoint.committed

Support: public. A turn's checkpoint commit landed.

Field	Type	Meaning
turn	int	The execution's turn number.
sha	str	The commit identifier, possibly abbreviated.
subject	str	The landed commit's subject.
task_id	str, optional	Task identifier.

checkpoint.failed

Support: public. A checkpoint commit could not land.

Field	Type	Meaning
turn	int	The execution's turn number.
error	str	The failure diagnostic.
worktree_status	str	Git status when the commit failed.
commit_subject	str	The attempted commit subject.

checkpoint.left_out

Support: public. A checkpoint left ignored or untracked edits outside its commit.

Field	Type	Meaning
turn	int	The execution's turn number.
ignored	list	Ignored edits left outside the checkpoint.
untracked	list	Untracked edits left outside the checkpoint.

command.backgrounded

Support: public. A command was handed back as a background job.

Field	Type	Meaning
background_id	str	Background job identifier.
handle	str	Launcher-instance-scoped job handle.
elapsed_s	float	Elapsed command time in seconds at handback.

compact.deduped

Support: public. Repeated tool results were removed from context.

Field	Type	Meaning
count	int	Number of affected entries or notices.
calls	list	The affected tool call identities.

compact.dropped

Support: public. Tool results were elided from context.

Field	Type	Meaning
count	int	Number of affected entries or notices.
calls	list	The affected tool call identities.

compact.gist.call

Support: public. The model was asked to summarize files for context.

Field	Type	Meaning
files	int	Number of files to summarize.

compact.gist.done

Support: public. File summaries were created or demoted during compaction.

Field	Type	Meaning
gisted	int	Number of file summaries created.
demoted	int	Number of summaries replaced by smaller placeholders.
paths	list	Affected repository-relative paths.
demoted_paths	list	Paths whose summaries were demoted.

compact.gist.failed

Support: public. The file-summary call failed.

Field	Type	Meaning
error	str	The failure diagnostic.

compact.refused

Support: public. There was too little history to summarize.

Field	Type	Meaning
reason	str	The reason for the outcome.

compact.requested

Support: public. The operator requested a context restart.

Field	Type	Meaning
focus	str	The requested summary or review focus.

compact.restored

Support: public. A resumed context restored its elision and summary counts.

Field	Type	Meaning
elided	int	Number of restored elision markers.
gists	int	Number of restored file summaries.

compact.summarise.call

Support: public. The model was asked to summarize the conversation.

Field	Type	Meaning
messages	int	Number of conversation messages.

compact.summarise.done

Support: public. A summary replaced the conversation history.

Field	Type	Meaning
summary_chars	int	Length of the summary in characters.
summary	str	The result or outcome in words.
kept_messages	int	Number of recent messages kept after the summary.

compact.summarise.failed

Support: public. The conversation-summary call failed.

Field	Type	Meaning
error	str	The failure diagnostic.

compact.thinking_dropped

Support: public. Reasoning text was removed from context.

Field	Type	Meaning
messages	int	Number of conversation messages.
chars	int	Text length in characters.

decision.recorded

Support: public. An operator ruling was recorded in decisions.md.

Field	Type	Meaning
question	str	The question presented to the operator.
answer	str	The operator's answer.

decision.unrecorded

Support: public. An operator ruling could not be recorded or was missing at finish.

Field	Type	Meaning
error	str, optional	The failure diagnostic.
missing	int, optional	Number of operator rulings missing at finish.

diff.updated

Support: public. A chain commit landed with its patch.

Field	Type	Meaning
sha	str	The commit identifier, possibly abbreviated.
patch	str	The commit's patch, capped at 8000 bytes.

environment.acquired

Support: public. The run holds its container: as the owner whose start made it appear, or as a borrower.

Field	Type	Meaning
execution_id	str, optional	Execution identifier.
lifecycle_ref	str	The lease key, the one the target's lifecycle_ref names.
role	str	owner or borrower.
reason	str	Why the run holds that role, in the operator's words.
started	bool	Whether this run's start command made the container appear.
start_s	float, optional	How long the start command took in seconds; null when none ran.

environment.starting

Support: public. The run's [container].start command is running; a container follows, or the run refuses.

Field	Type	Meaning
execution_id	str, optional	Execution identifier.
command	list	The start command's argv.

execution.environment

Support: public. An environment lifecycle observation, independent of job settlement.

Field	Type	Meaning
execution_id	str, optional	Execution identifier.
target_id	str, optional	Recorded target reference.
outcome	str	left_running, stopped, unknown or not_applicable.
reason	str	The observation's reason, not stop authority.

execution.jobs

Support: public. The command owner observed job settlement or could not confirm it.

Field	Type	Meaning
execution_id	str, optional	Execution identifier.
target_id	str, optional	Recorded target reference.
outcome	str	settled or unknown.
reason	str	Why settlement is unknown, empty when confirmed.

execution.target

Support: public. Immutable target facts recorded before workspace commands.

Field	Type	Meaning
execution_id	str, optional	Execution identifier.
target_id	str	Host-issued evidence reference, not launcher identity.
target	object	ExecutionTargetStamp facts; null fields stay unobserved.

finish.refused.contract

Support: public. The finish payload did not satisfy the machine's result contract.

Field	Type	Meaning
turn	int	The execution's turn number.
problems	list	Violations of the machine's finish contract.

finish.refused.metric

Support: public. A metric run's early finish was returned to the model.

Field	Type	Meaning
turn	int	The execution's turn number.
nudges_used	int	Number of notices issued for this condition.
budget_remaining	float, optional	Remaining budget fraction; null without a finite cap.
trigger	str, optional	The action that triggered the gate or panel.

finish.refused.review

Support: public. The review panel rejected the declared finish.

Field	Type	Meaning
turn	int	The execution's turn number.
ending	str	The declared finish being reviewed.

finish.refused.tasks

Support: public. A finish with open subtasks was returned to the model.

Field	Type	Meaning
turn	int	The execution's turn number.
nudges_used	int	Number of notices issued for this condition.
trigger	str	The action that triggered the gate or panel.

finish.refused.verify

Support: public. A finish without green verification was returned to the model.

Field	Type	Meaning
turn	int	The execution's turn number.
nudges_used	int	Number of notices issued for this condition.
trigger	str	The action that triggered the gate or panel.

graph.update

Support: public. The task graph and its current focus.

Field	Type	Meaning
nodes	object	Task nodes keyed by identifier.
cursor	str, optional	Current task identifier; null when none is selected.

jail.degraded

Support: public. The sandbox started weaker than requested or left surviving processes.

Field	Type	Meaning
detail	str	The sandbox degradation or surviving-process diagnostic.

jail.tool_unreachable

Support: public. A tool found on the host could not run inside the sandbox.

Field	Type	Meaning
binary	str	The unreachable executable's name.

loop.note

Support: public. A machine-creation progress note.

Field	Type	Meaning
text	str	The event's text.

loop.parked

Support: public. A standing run parked at a turn boundary.

Field	Type	Meaning
turn	int	The execution's turn number.
reason	str	The reason for the outcome.

loop.parked.resumed

Support: public. A parked run received work or an operator request.

Field	Type	Meaning
turn	int	The execution's turn number.

mcp.server_unavailable

Support: public. A configured MCP server could not start.

Field	Type	Meaning
server	str	Configured MCP server name.
error	str	The failure diagnostic.

media.refused

Support: public. An attachment was refused by its tool or model route.

Field	Type	Meaning
provider_tool_call_id	str	The provider's tool-use identifier.
reason	str	The reason for the outcome.
route	str, optional	The model route that refused the attachment.
text	str, optional	The event's text.
attachment	object, optional	Attachment metadata without its binary content.

metric.auto_call

Support: public. An automatic metric sample started.

Field	Type	Meaning
turn	int	The execution's turn number.
sha	str	The commit identifier, possibly abbreviated.

metric.auto_failed

Support: public. An automatic metric sample failed.

Field	Type	Meaning
turn	int	The execution's turn number.
error	str	The failure diagnostic.

metric.ceiling_reached

Support: public. The metric ceiling ended the run.

Field	Type	Meaning
turn	int	The execution's turn number.

metric.end

Support: public. The metric command finished.

Field	Type	Meaning
command	list	Command argv.
exit_code	int, optional	Command exit code; null if no exit was observed.
duration_s	float	Command duration in seconds.
stdout_tail	str	Capped standard-output tail.
stderr_tail	str	Capped standard-error tail.
score	float, optional	Parsed metric score; null when unavailable.

metric.sample

Support: public. A metric sample was recorded.

Field	Type	Meaning
turn	int	The execution's turn number.
label	str	Metric sample label.
score	float, optional	Parsed metric score; null when unavailable.
exit_code	int, optional	Command exit code; null when unavailable.
sha	str	The commit identifier, possibly abbreviated.

metric.start

Support: public. The metric command started.

Field	Type	Meaning
command	list	Command argv.

model.attempt

Support: public. One request attempt of a model call ended; the ending attempt's model.result follows.

Field	Type	Meaning
execution_id	str, optional	The execution the call ran in; null for a call outside one.
panel_id	str, optional	Review panel identifier.
role	str	The calling model role.
model	str	Model identifier.
provider	str	Configured provider name.
model_call_id	str	Identifier pairing the attempt with its model call.
reviewer_index	int, optional	Zero-based reviewer index.
attempt	int	The attempt's number within the call, from 1.
request_id	str, optional	The HTTP attempt's id, as provider.request and the transcript name it.
ok	bool	Whether the attempt returned a response.
input_tokens	int, optional	Input tokens the attempt used.
output_tokens	int, optional	Output tokens the attempt used.
cache_read_tokens	int, optional	Cached input tokens the attempt used.
cache_write_tokens	int, optional	Cache-write tokens the attempt used.
cost_usd	float, optional	Provider-reported USD cost; zero means free, absent means unreported.
stop_reason	str, optional	The provider's response stop reason.
error	str, optional	The failure diagnostic.

model.call

Support: public. A model call started for a role.

Field	Type	Meaning
execution_id	str, optional	The execution the call ran in; null for a call outside one.
panel_id	str, optional	Review panel identifier.
role	str	The calling model role.
model	str	Model identifier.
provider	str	Configured provider name.
model_call_id	str	Identifier pairing the model call with its result.
reviewer_index	int, optional	Zero-based reviewer index.

model.reasoning_starvation

Support: public. Reasoning consumed a response without usable output.

Field	Type	Meaning
turn	int	The execution's turn number.
reasoning_chars	int	Reasoning output length in characters.
output_tokens	int	Output token count.
stop_reason	str, optional	The provider's response stop reason.

model.result

Support: public. A model call returned or failed, once per logical call, with its last attempt's usage.

Field	Type	Meaning
execution_id	str, optional	The execution the call ran in; null for a call outside one.
panel_id	str, optional	Review panel identifier.
role	str	The calling model role.
model	str	Model identifier.
provider	str	Configured provider name.
model_call_id	str	Identifier pairing the model call with its result.
reviewer_index	int, optional	Zero-based reviewer index.
ok	bool	Whether the operation succeeded.
text	str, optional	The event's text.
input_tokens	int, optional	Input tokens used by this model call.
output_tokens	int, optional	Output tokens used by this model call.
cache_read_tokens	int, optional	Cached input tokens used by this model call.
cache_write_tokens	int, optional	Cache-write tokens used by this model call.
cost_usd	float, optional	Provider-reported USD cost; zero means free, absent means unreported.
stop_reason	str, optional	The provider's response stop reason.
error	str, optional	The failure diagnostic.

model.text_delta

Support: public. A piece of assistant text streamed.

Field	Type	Meaning
role	str	The calling model role.
text	str	The event's text.
execution_id	str, optional	The execution the call ran in; null for a call outside one.
model_call_id	str, optional	The model call streaming the piece; null in a journal written before ids.
panel_id	str, optional	Review panel identifier.
reviewer_index	int, optional	The reviewer's roster position; null outside a reviewer.

model.thinking_delta

Support: public. A piece of assistant reasoning streamed.

Field	Type	Meaning
role	str	The calling model role.
text	str	The event's text.
execution_id	str, optional	The execution the call ran in; null for a call outside one.
model_call_id	str, optional	The model call streaming the piece; null in a journal written before ids.
panel_id	str, optional	Review panel identifier.
reviewer_index	int, optional	The reviewer's roster position; null outside a reviewer.

nudge.loop_guard

Support: public. Repeated identical tool calls triggered a warning.

Field	Type	Meaning
turn	int	The execution's turn number.
tool	str	The repeated tool's name.
streak	int	Number of consecutive matching turns or calls.

nudge.memory_flip

Support: public. Verification turned green and the model was asked to record the lesson.

Field	Type	Meaning
turn	int	The execution's turn number.

nudge.metric_plateau

Support: public. A metric plateau prompted another attempt.

Field	Type	Meaning
turn	int	The execution's turn number.
nudges_used	int	Number of notices issued for this condition.
budget_remaining	float, optional	Remaining budget fraction; null without a finite cap.

nudge.no_progress

Support: public. Repeated turns without progress prompted a warning.

Field	Type	Meaning
turn	int	The execution's turn number.
streak	int	Number of consecutive matching turns or calls.
level	int	The warning's escalation level.

nudge.plan_finish

Support: public. The model was asked to finish its plan within the remaining budget.

Field	Type	Meaning
turn	int	The execution's turn number.
budget_remaining	float, optional	Remaining budget fraction; null without a finite cap.

nudge.question

Support: public. A silent finish ended on a question and was returned to the model.

Field	Type	Meaning
turn	int	The execution's turn number.

nudge.run_budget

Support: public. The model was asked to finish within its remaining run budget.

Field	Type	Meaning
turn	int	The execution's turn number.
budget_remaining	float, optional	Remaining budget fraction; null without a finite cap.

nudge.silent_no_work

Support: public. A silent finish without work was returned to the model.

Field	Type	Meaning
turn	int	The execution's turn number.
nudges_used	int	Number of notices issued for this condition.

nudge.stagnation

Support: public. Elapsed time without edits or verification prompted a warning.

Field	Type	Meaning
turn	int	The execution's turn number.
elapsed_s	int	Elapsed seconds without edits or verification.

nudge.task_stuck

Support: public. Time on one task prompted a change of approach.

Field	Type	Meaning
task_id	str	Task identifier.
turns	int	Number of execution turns, or turns spent on one task.
nudges_used	int	Number of notices issued for this condition.

nudge.tool_error

Support: public. Repeated tool errors prompted a warning.

Field	Type	Meaning
turn	int	The execution's turn number.
streak	int	Number of consecutive matching turns or calls.
level	int	The warning's escalation level.

nudge.verify_broken

Support: public. An unusable verify command prompted a correction request.

Field	Type	Meaning
turn	int	The execution's turn number.

nudge.verify_settled

Support: public. Repeated green verification without edits prompted a finish request.

Field	Type	Meaning
turn	int	The execution's turn number.
idle	int	Consecutive settled turns.

nudge.went_quiet

Support: public. A response without tool calls prompted the model to continue.

Field	Type	Meaning
turn	int	The execution's turn number.
nudges_used	int	Number of notices issued for this condition.
nudges_max	int	Maximum notices allowed for this condition.
output_tokens	int	Output token count.

operator.presence

Support: public. The operator's observed attachment state changed.

Field	Type	Meaning
presence	str	The observed operator attachment and away state.

parallel.compared

Support: public. Parallel candidates were ranked.

Field	Type	Meaning
group	str	Parallel dispatch group identifier.
ranked_by	str	The method used to rank the candidates.
rationale	str	The ranking explanation.
ranking	list	Candidates in ranked order with their verify and cost readings.
judge_cost_usd	float	Ranking judge spend in USD.
judge_cost_partial	bool	Whether the judge spend is a lower bound.

parallel.dispatched

Support: public. Parallel children were dispatched.

Field	Type	Meaning
group	str	Parallel dispatch group identifier.
children	int	Child count for dispatch, or child results for completion.
tasks	list	Task texts sent to the children.
source	str, optional	Who dispatched: operator (a steer), model (the tool) or panel.

parallel.failed

Support: public. Parallel child dispatch or execution failed.

Field	Type	Meaning
group	str	Parallel dispatch group identifier.
children	list, optional	Child count for dispatch, or child results for completion.
error	str, optional	The failure diagnostic.

parallel.joined

Support: public. Parallel child results returned.

Field	Type	Meaning
group	str	Parallel dispatch group identifier.
children	list	Child count for dispatch, or child results for completion.
source	str, optional	Who dispatched: operator (a steer), model (the tool) or panel.

pin.added

Support: public. An operator instruction was pinned in context.

Field	Type	Meaning
text	str	The event's text.
chars	int	Text length in characters.
count	int	Total pinned instructions.

pin.refused

Support: public. An instruction exceeded the pinned-context limit.

Field	Type	Meaning
chars	int	Text length in characters.
limit	int	Pinned instruction character limit.

pin.restored

Support: public. A resumed or forked execution restored pinned instructions.

Field	Type	Meaning
pins	list	The full list of pinned instructions.
count	int	Total pinned instructions.

plan.read_failed

Support: public. The plan could not be read.

Field	Type	Meaning
path	str	The plan's file path.
error	str	The failure diagnostic.

plan.reread

Support: public. The plan was read back into context.

Field	Type	Meaning
path	str	The plan's file path.
bytes	int	UTF-8 text length in bytes.

plan.write_failed

Support: public. The plan could not be written.

Field	Type	Meaning
path	str	The plan's file path.
error	str	The failure diagnostic.

plan.written

Support: public. The plan was written to disk.

Field	Type	Meaning
path	str	The plan's file path.
bytes	int	UTF-8 text length in bytes.

provider.auth

Support: public. A provider credential refresh, exchange or recovery finished.

Field	Type	Meaning
provider	str	Configured provider name.
kind	str	The request or authentication operation kind.
outcome	str	The operation's outcome.

provider.empty_tool_call_retry

Support: public. A provider response with empty tool calls was retried.

Field	Type	Meaning
attempt	int	Retry attempt number.
stop_reason	str	The provider's response stop reason.

provider.fatal

Support: public. A non-retryable provider error stopped the call.

Field	Type	Meaning
status_code	int, optional	HTTP status code when available.
error	str	The failure diagnostic.

provider.request

Support: public. An HTTP attempt was made for a model call.

Field	Type	Meaning
execution_id	str, optional	The execution the call ran in; null for a call outside one.
role	str	The calling model role.
provider	str	Configured provider name.
model	str	Model identifier.
model_call_id	str	The logical model call that made this request.
panel_id	str, optional	Review panel identifier.
reviewer_index	int, optional	Zero-based reviewer index.
request_id	str	HTTP attempt identifier, also stored at the transcript's top level.
target	str	The immutable review target, or provider request class.

provider.retry

Support: public. A provider error prompted another attempt.

Field	Type	Meaning
attempt	int	Retry attempt number.
error	str	The failure diagnostic.

question.answer

Support: public. The operator answered questions or nobody could answer.

Field	Type	Meaning
id	str	Question prompt identifier.
answers	list	Answers aligned with the requested questions.
source	str	Where the request or result came from.
unseen	bool	Whether nobody could answer and no question remains queued.

question.delivered

Support: public. Queued answers were delivered to the model.

Field	Type	Meaning
ids	list	Identifiers of the delivered questions.

question.prompt

Support: public. Questions were presented to the operator.

Field	Type	Meaning
id	str	Question prompt identifier.
questions	list	The questions presented.
call_id	int, optional	The tool call correlated with this event.
wait	bool	Whether the call waits for answers.

review.finding.reopened

Support: public. An operator reopened a finding.

Field	Type	Meaning
finding_id	str	Finding identifier.
request_id	str	Request identifier used to correlate or deduplicate work.
source	str	Where the request or result came from.
version	int	Review-state version after the action.
reason	str	The reason for the outcome.

review.finding.resolved

Support: public. An operator recorded a finding's disposition.

Field	Type	Meaning
finding_id	str	Finding identifier.
request_id	str	Request identifier used to correlate or deduplicate work.
source	str	Where the request or result came from.
version	int	Review-state version after the action.
resolution	object	The finding's disposition and supporting reason.

review.fix.failed

Support: public. A finding's fix request could not be delivered.

Field	Type	Meaning
finding_id	str	Finding identifier.
request_id	str	Request identifier used to correlate or deduplicate work.
source	str	Where the request or result came from.
version	int	Review-state version after the action.
error	str	The failure diagnostic.

review.fix.requested

Support: public. An operator requested a fix for a finding.

Field	Type	Meaning
finding_id	str	Finding identifier.
request_id	str	Request identifier used to correlate or deduplicate work.
source	str	Where the request or result came from.
version	int	Review-state version after the action.
target	object	The immutable review target, or provider request class.
action	str	Fix delivery action: live-steer or resume.

review.fix.submitted

Support: public. A finding's fix request reached an execution.

Field	Type	Meaning
finding_id	str	Finding identifier.
request_id	str	Request identifier used to correlate or deduplicate work.
source	str	Where the request or result came from.
version	int	Review-state version after the action.
execution_id	str	Execution identifier.

review.panel

Support: public. A review panel completed and recorded its evidence and blocking decision.

Field	Type	Meaning
panel_id	str	Review panel identifier.
execution_id	str	Execution identifier.
turn	int	The execution's turn number.
trigger	str	The action that triggered the gate or panel.
target	object	The immutable review target, or provider request class.
findings	list	Grounded findings or panel observations.
injected_finding_ids	list	Finding identifiers supplied to the worker.
block_on	str	The configured panel blocking rule.
roster	list, optional	The reviewers that sat: index, model route, focus and temperature.
raw_blocked	bool	Whether the verdicts meet the blocking rule.
blocked	bool	Whether the armed gate blocks the finish.
disarmed	bool	Whether the review gate's rejection cap disarmed it.
n_block	int	Number of distinct blocking provider/model routes.
n_abstain	int	Number of abstaining reviewers.
completion	str	Whether the panel completed or was inconclusive.

review.rejection_cap_reached

Support: public. The review gate reached its rejection limit.

Field	Type	Meaning
turn	int	The execution's turn number.
rejections	int	Number of review rejections.

review.reused

Support: public. A complete panel over the same trees, paths, reviewers and rule stood in for a new one.

Field	Type	Meaning
panel_id	str	Review panel identifier.
execution_id	str	Execution identifier.
turn	int	The execution's turn number.
trigger	str	The action that triggered the gate or panel.
reused_panel_id	str	The complete panel whose verdict stood in.
target	object	The immutable review target both panels share.
findings	int	How many findings the reused panel recorded.

review.skipped

Support: public. A review panel could not run.

Field	Type	Meaning
panel_id	str	Review panel identifier.
execution_id	str	Execution identifier.
turn	int, optional	The execution's turn number.
trigger	str, optional	The action that triggered the gate or panel.
reason	str	The reason for the outcome.
error	str, optional	The failure diagnostic.
remaining	float, optional	Remaining budget fraction.

review.start

Support: public. A review panel began over an immutable target.

Field	Type	Meaning
panel_id	str	Review panel identifier.
execution_id	str	Execution identifier.
turn	int	The execution's turn number.
trigger	str	The action that triggered the gate or panel.
target	object	The immutable review target, or provider request class.
rule_sources	list	Base-tree rule files supplied to the panel.
rules_complete	bool	Whether every required rule was supplied.
rules_reason	str, optional	Why rule collection was incomplete; null when complete.
reviewers	list	The resolved reviewer configurations.

review.verdict

Support: public. A reviewer returned a grounded verdict or abstained.

Field	Type	Meaning
panel_id	str	Review panel identifier.
execution_id	str	Execution identifier.
turn	int, optional	The execution's turn number.
trigger	str, optional	The action that triggered the gate or panel.
reviewer_index	int	Zero-based reviewer index.
focus	str	The requested summary or review focus.
model_route	str	The reviewer's configured model route.
verdict	str	Reviewer verdict: pass, block or abstain.
findings	list	Grounded findings or panel observations.
summary	str	The result or outcome in words.
error	str, optional	The failure diagnostic.
session_id	str, optional	The reviewer's session, when assigned.

session.resume

Support: public. Emitted at the start of every resumed execution.

Field	Type	Meaning
session_id	str	Session identifier.
user_task	str	The displayed task text.
steer	str	The instruction supplied when resuming.
mode	str	The execution mode.
turn	int	The execution's turn number.
messages	int	Number of conversation messages.
execution_id	str	Execution identifier.

session.resume.interrupted

Support: public. Resume found tool calls that never returned and will not be rerun.

Field	Type	Meaning
turn	int	The execution's turn number.
tools	list	Interrupted tool calls that will not be rerun.

session.start

Support: public. Emitted once when the session is created.

Field	Type	Meaning
session_id	str, optional	Session identifier.
user_task	str	The displayed task text.
mode	str	The execution mode.
execution_id	str	Execution identifier.

session.stop

Support: public. The execution stopped, with its reason; the session stays resumable.

Field	Type	Meaning
execution_id	str	Execution identifier; empty when unknown.
reason	str	The reason for the outcome.
turns	int	Number of execution turns, or turns spent on one task.
summary	str, optional	The result or outcome in words.
verification	str, optional	Final verification state: passed, failed, unverified or not_applicable.
scoped	bool, optional	Whether verification ran only the selected nearby tests.
final_verify_id	str, optional	The final verification attempt identifier.
tree	str, optional	The final working-tree content identifier.
tool	str, optional	The repeated tool's name.
streak	int, optional	Number of consecutive matching turns or calls.

session.stranded

Support: public. A finished run left edits uncommitted and no commit landed.

Field	Type	Meaning

session.undone

Support: public. Undo forked the run before its last operator message.

Field	Type	Meaning
new_session_id	str	Identifier of the undo fork.
undone_text	str	The operator message taken back.

skills.warning

Support: public. A skill declaration produced a warning.

Field	Type	Meaning
warning	str	The skill diagnostic.

standing.resumed

Support: public. A standing goal resumed after absorbing a soft stop.

Field	Type	Meaning
reason	str	The reason for the outcome.
task_id	str	Task identifier.
turn	int	The execution's turn number.

standing.set

Support: public. The operator set a standing goal.

Field	Type	Meaning
task_id	str	Task identifier.
title	str	Task or standing-goal text.

steer.aborted

Support: public. The operator's steer aborted the run.

Field	Type	Meaning

steer.detached

Support: public. The operator's steer stopped the run for background resume.

Field	Type	Meaning

steer.exited

Support: public. The operator's steer ended the run and left the terminal.

Field	Type	Meaning

steer.feedback

Support: public. A steer directive returned feedback without invoking the model.

Field	Type	Meaning
text	str	The feedback for the operator.

steer.injected

Support: public. An operator instruction was injected into context.

Field	Type	Meaning
chars	int	Text length in characters.
text	str	The event's text.

steer.requested

Support: public. The loop reached an operator steer request.

Field	Type	Meaning
turn	int	The execution's turn number.

steer.signalled

Support: public. Ctrl-C requested an operator steer during the run.

Field	Type	Meaning
source	str	Where the request or result came from.

steer.skill

Support: public. An operator steer invoked a skill.

Field	Type	Meaning
name	str	Tool or skill name.
args	str	Tool argument preview, or the skill invocation's argument text.

steer.undo

Support: public. The operator's steer requested an undo fork.

Field	Type	Meaning

task.queued

Support: public. The operator queued a task.

Field	Type	Meaning
task_id	str	Task identifier.
title	str	Task or standing-goal text.

task.request_refused

Support: public. An operator task-graph request could not be applied.

Field	Type	Meaning
kind	str	The request or authentication operation kind.
text	str	The event's text.
error	str	The failure diagnostic.

task.retired

Support: public. The operator retired a task.

Field	Type	Meaning
task_id	str	Task identifier.
title	str	Task or standing-goal text.

task.surfaced

Support: public. A task became the current focus.

Field	Type	Meaning
task_id	str	Task identifier.

tool.call

Support: public. A tool was dispatched, including calls that a guard rejects.

Field	Type	Meaning
name	str	Tool or skill name.
args_preview	object	The arguments with long values clipped, or a skill's argument text.
call_id	int, optional	The tool call correlated with this event; null in a journal written before ids.
execution_id	str, optional	The dispatching execution; null in a journal written before ids.
model_call_id	str, optional	The model call whose response made the call; null for a harness-run tool.
panel_id	str, optional	Review panel identifier.
reviewer_index	int, optional	The reviewer's roster position; null outside a reviewer.
provider_tool_call_id	str, optional	The provider's tool-use id for the block; null for a harness-run tool.
turn	int, optional	The execution's turn number; null for a harness-run tool.

tool.result

Support: public. A dispatched tool returned its result or refusal.

Field	Type	Meaning
name	str	Tool or skill name.
ok	bool	Whether the handler returned a result; a command's failure is its exit_code.
summary	str	The result or outcome in words.
call_id	int, optional	The tool call correlated with this event; null in a journal written before ids.
execution_id	str, optional	The dispatching execution; null in a journal written before ids.
model_call_id	str, optional	The model call whose response made the call; null for a harness-run tool.
panel_id	str, optional	Review panel identifier.
reviewer_index	int, optional	The reviewer's roster position; null outside a reviewer.
provider_tool_call_id	str, optional	The provider's tool-use id for the block; null for a harness-run tool.
turn	int, optional	The execution's turn number; null for a harness-run tool.
exit_code	int, optional	A command's exit code; null while it runs in the background or when unknown.
duration_s	float, optional	A command's wall-clock seconds until the result.
timed_out	bool, optional	Whether the runner ended the command at its deadline.
exec_failed	bool, optional	Whether the command could not be started.
background_id	str, optional	The background job the command continues as, when handed back.
paths	list, optional	Affected repository-relative paths.
stdout_tail	str, optional	Capped standard-output tail.
stderr_tail	str, optional	Capped standard-error tail.
stdout_truncated	bool, optional	Whether stdout_tail is a cut of the output; null when its end is unknown.
stderr_truncated	bool, optional	Whether stderr_tail is a cut of the output; null when its end is unknown.
stopped_by_operator	bool, optional	Whether an immediate operator stop ended the command.
attachment	object, optional	Attachment metadata without its binary content.
head_tail	str, optional	A read preview or attachment summary.
lines_total	int, optional	Total lines in the file read.

verify.baseline

Support: public. The baseline verification attempt finished.

Field	Type	Meaning
ok	bool	Whether the operation succeeded.
turn	int	The execution's turn number.
verify_id	str, optional	Verification attempt identifier.

verify.end

Support: public. A verification attempt finished with evidence about the observed tree.

Field	Type	Meaning
verify_id	str, optional	Verification attempt identifier.
execution_id	str, optional	Execution identifier.
target_id	str, optional	Recorded execution-target reference.
command	list	Command argv.
tree_before	str, optional	Content identifier observed before verification.
tree_after	str, optional	Content identifier observed after verification.
scoped	bool	Whether verification ran only the selected nearby tests.
selected_paths	list	Paths selected for scoped verification.
outcome	str	The operation's outcome.
exit_code	int, optional	Command exit code; null if no exit was observed.
duration_s	float	Command duration in seconds.
timeout_s	float	Verification wall-clock limit in seconds.
stdout_tail	str	Capped standard-output tail.
stderr_tail	str	Capped standard-error tail.
stdout_truncated	bool, optional	Whether standard-output capture was truncated; null when unknown.
stderr_truncated	bool, optional	Whether standard-error capture was truncated; null when unknown.
baseline	bool	Whether the attempt measured the baseline tree.
reason	str	The reason for the outcome.

verify.inferred

Support: public. The verify command was inferred, restored, adopted or disabled.

Field	Type	Meaning
command	list	Resolved verify argv; empty when no command is adopted.
source	str	Where the request or result came from.
adopted_at	int, optional	Turn where the gate was adopted, restored or dropped.

verify.requested

Support: public. The harness requested verification.

Field	Type	Meaning
reason	str	Why the harness requested verification.
turn	int	The execution's turn number.

verify.scoped

Support: public. Verification was narrowed to tests near the changed files.

Field	Type	Meaning
paths	list	Affected repository-relative paths.
turn	int	The execution's turn number.

verify.start

Support: public. A verification attempt started.

Field	Type	Meaning
approval_required	bool, optional	Whether the attempt needs approval.
call_id	int, optional	The model tool call, absent for a harness gate.
verify_id	str, optional	Verification attempt identifier.
execution_id	str, optional	Execution identifier.
target_id	str, optional	Recorded execution-target reference.
tree_before	str, optional	Content identifier observed before verification.
command	list	Command argv.
scoped	bool	Whether verification ran only the selected nearby tests.
selected_paths	list	Paths selected for scoped verification.
timeout_s	float	Verification wall-clock limit in seconds.

verify.swapped

Support: public. Resume replaced the saved verify command.

Field	Type	Meaning
was	list	Saved verification argv.
now	list	Replacement verification argv.

verify.test_only_green

Support: public. Verification passed after changes confined to tests.

Field	Type	Meaning
turn	int	The execution's turn number.
paths	list	Affected repository-relative paths.

worktree.setup

Support: public. The worktree's setup command finished.

Field	Type	Meaning
execution_id	str, optional	Execution identifier.
target_id	str, optional	Recorded execution-target reference.
command	list	Command argv.
exit_code	int, optional	Command exit code; null if no exit was observed.
duration_s	float	Command duration in seconds.
exec_failed	bool	Whether the setup command could not be executed.
stdout_tail	str	Capped standard-output tail.
stderr_tail	str	Capped standard-error tail.

Provider transcripts

Support: durable but private.

transcripts/<timestamp>-<seq>.json holds one record per model attempt. The sequence is padded to six digits in the filename. The record is written with the redacted request before the send, with outcome: "incomplete", and rewritten in place when the attempt ends: completed, failed, interrupted (a steer) or aborted (a stop or a cancellation). A record stays incomplete while the call runs, and after a process ended before the attempt did; it does not say whether the provider received or billed the request. sessions transcript marks every attempt that did not complete. model_call_id joins a transcript to its logical call; request_id joins it to its provider.request event. Retries share the call id and have separate request ids. Older transcripts may omit these ids and outcome. Secret-bearing request and response headers are redacted, and their credential values of at least eight characters are scrubbed wherever they appear in the record. Attachment payloads are omitted and replaced with metadata, so these files cannot replay the attachment bytes. Each session keeps an objects/ store: one file per distinct JSON value, named by the SHA-256 of its canonical JSON (sort_keys=True, compact separators) after redaction and media omission. A transcript's request holds a reference, {"$object": "<sha256 hex>"}, in place of each message (a Responses input item) and of its system, instructions and tools values; a numbered checkpoint references its messages the same way. An object is written before any record that references it, the pre-send record included, so an unchanged retry writes nothing new. loop_state.json and turn_in_flight.json stay full: they are recovery state. Readers materialize references, accept older full records, and raise ObjectStoreError naming a missing object and its record; sessions transcript --json prints materialized records.

Artifact pruning

sessions prune --older-than DAYS removes transcripts/, checkpoints/ and objects/ together from finished sessions whose last execution ended more than DAYS ago, writing pruned.json first so an interrupted prune cannot reuse a sequence. The manifest, journal, receipt, recovery state and graph store stay. sessions transcript reports the prune, a fork at a removed checkpoint refuses with the same notice, and a resumed session continues its transcript sequence above the saved one. Nothing is deleted automatically and nothing is compressed.

Field	Type	Meaning
pruned_at	pydantic.types.AwareDatetime	The UTC time pruning began.
counts	dict[Literal['transcripts', 'checkpoints', 'objects'], int]	The file counts selected for removal in each directory.
last_transcript_seq	int	The highest removed sequence, including earlier prunes.

Attempt record

Source: providers.provider_types.TranscriptAttempt.write.

Field	Type	Meaning
ts	JSON	The UTC start timestamp in YYYYMMDDTHHMMSSffffffZ form.
seq	JSON	The run-global sequence, starting at 1 and continuing across resumes.
model_call_id	JSON	The logical call id from model.call; null without a call context.
request_id	JSON	The HTTP attempt id from provider.request; empty for a local process.
seat	JSON	The role or review focus label, not a unique seat identity.
outcome	JSON	incomplete, completed, failed, interrupted or aborted.
request	JSON	The request formed for the provider.
request.url	JSON	The URL dialled.
request.headers	JSON	The request headers, with secret-bearing values redacted.
request.body	JSON	The redacted body, with message and prompt references into objects/.
response	JSON	What the attempt received before it ended.
response.status	JSON	The HTTP status, or 0 when no response opened.
response.headers	JSON	The response headers, with secret-bearing values redacted.
response.body	JSON	The parsed response body or raw error text.
usage	JSON	The observed token counts and cost; null when the provider reported none.
error	JSON	The exception type and reason; empty when none was raised.
usage.input_tokens	int	Fresh input tokens billed.
usage.output_tokens	int	Completion tokens billed.
usage.cache_read_tokens	int	Prompt tokens served from the cache.
usage.cache_write_tokens	int	Prompt tokens written to the cache.
usage.cost_usd	float \| None	The reported cost; None means unreported and zero means a free call.

CLI JSON

sessions list --json emits summary_row objects with row_json additions. sessions show --json emits one _cmd_status object. ps --json emits an array of _Row records, including nested children.

sessions list --json row

Support: public. Return one listing row as JSON, the shape sessions list --json and /api/hub share.

Source: viewmodel.listing.summary_row.

Field	Type	Meaning
session_id	str	The session identifier.
mode	str	The session mode.
task	str	The operator's words, whole.
task_line	str	The task clipped to 200 characters, then a newline and continued: <steer>, clipped the same way, when a resume steer exists.
continued_task	str	The latest resume steer, whole; empty without a continuation.
status	str	The lifecycle status word.
status_detail	str	The detail explaining the status.
stop_reason	str	The last journaled end reason, or an empty string while running.
status_label	str	The rendered status with mode and unmerged marks.
status_level	str	The status presentation level.
last_activity	float	The last activity time in epoch seconds.
when	str	The rendered last activity time.
cost_usd	float	Cumulative spend in USD across executions.
usd_partial	bool	Whether unpriced usage makes spend a lower bound.
plan_observed_points	float	The plan movement observed across executions, in points.
plan_cap	float	The subscription points cap.
usage	str	The rendered usage cell: plan points, dollars, or both.
model	str	The resolved provider/model route.
model_from_flag	bool	Whether a model flag selected the route.
id_cell	str	The session identifier with winner and worktree marks.
unmerged	bool	Whether the run holds unmerged work.
verification	Literal['passed', 'failed', 'unverified', 'not_applicable'] \| None	The gate verdict, or null without a verdict.
winner	bool	Whether this child won its comparison.
number	int \| None	The child number, or null outside a group.
parent	str	The session that dispatched this child.
worktree	str \| None	The linked worktree path, or null in the operator's checkout.
branch	str \| None	The run's branch while it exists, or null.
open_findings	dict[str, int]	The open grounded findings by severity: block, warn, nit.
merge_note	str	The rendered open findings; "" with none open.
children	list[SummaryRow]	Nested listing rows for this parent's children.

sessions list --json group fields

Support: public. Return a listing row and its nested children as JSON.

Source: viewmodel.listing.row_json.

Field	Type	Meaning
when	str	The rendered group activity time.
group_last_activity	float	The group's latest activity time in epoch seconds.

sessions show --json

Support: public. Print a one-shot liveness and progress summary for a run.

Source: ui.cli.sessions_show._cmd_status.

Field	Type	Meaning
session_id	str	The session identifier.
mode	str \| None	The session mode.
task	str	The operator's words, whole.
continued_task	str	The latest resume steer, whole; empty without a continuation.
task_line	str	The task clipped to 200 characters, then a newline and continued: <steer>, clipped the same way, when a resume steer exists.
model	str	The resolved provider/model route, or ? when unknown.
models	dict[str, Any]	The recorded driver, reviewer, helper and reviewers, with effort and origin.
model_from_flag	bool	Whether a model flag selected the route.
preset	str \| None	The launch preset, or null when absent.
status	str	The lifecycle status word shared with listings.
status_label	str	The rendered status with mode and unmerged marks.
status_detail	str	The diagnostic or next action explaining the status.
end_summary	str \| None	The journaled closing sentence, or null when absent.
alive	bool	Whether the recorded worker identity is alive.
pid	int \| None	The recorded worker process identifier, or null when unreadable.
turn	int \| None	The latest turn number, or null before a turn.
last_event	str \| None	The latest event name, or null without a log.
last_event_age_s	float \| None	Seconds since the latest event, or null when unknown.
elapsed_s	float \| None	Seconds from the first start to the last event or current time.
stop_reason	str \| None	The journaled end reason, or null while unfinished.
baseline_ok	bool \| None	Whether baseline verification passed, or null when unrecorded.
verification	Literal['passed', 'failed', 'unverified', 'not_applicable'] \| None	The final gate word, or null when unrecorded.
budget	dict[str, Any]	The BudgetView with its rendered usd_text and plan_text, as attach --json carries it.
unattended_questions	int	Questions answered empty because nobody was attached.
setup_output	str \| None	Retained setup output, or null when absent.
environment_evidence	dict[str, Any]	Recorded targets, setup and lifecycle observations.
source_session_id	str \| None	The source session used by run --from, or null.
parent_session_id	str \| None	The fork parent, or null.
forked_from_turn	int \| None	The fork's source turn, or null.
forked_from_sha	str \| None	The fork's source commit, or null.
worktree	str \| None	The linked worktree path, or null.
placement	dict[str, Any] \| None	The child's rank and group id, or null outside a compared child.
comparison	dict[str, Any] \| None	The group record on its parent, or null for other sessions.
parallel	dict[str, Any] \| None	The ParallelLineage, or null outside a child.
children	list[SummaryRow]	The parent's summary_row objects.
run_branch	str \| None	The existing run branch, or null.
base_branch	str \| None	The base branch, or null when unknown.
merged_into	str \| None	The merge destination, or null when not merged.
pins	list[str]	The operator's active pinned instructions.

CLI text

Token + cost summary

The end-of-run summary groups usage by provider/model route. The journal's budget.update event carries the cumulative totals above.

Source: budget.BudgetTracker.format_summary.

Field	Type	Meaning
route	text	Each usage line starts with the provider/model route, such as chatgpt/gpt-6-astra.
in	text	Fresh input tokens for the route, or all routes on the TOTAL line.
out	text	Output tokens for the route, or all routes on the TOTAL line.
cache_r	text	Cache-read tokens for the route.
cache_w	text	Cache-write tokens for the route.
calls	text	Calls recorded for the route.
cost	text	USD spend, reported or estimated from that route's prices. The TOTAL includes purchased credits spent. = marks an exact total, ~ an estimate or lower bound, and + unpriced calls.
of	text	The USD cap when a route is USD-metered; unlimited means no cap.
unmetered	text	Tokens counted against the fallback cap, followed by that cap.
plan	text	One plan usage line per provider, with its binding window's used percent, observed window movement in points, optional cap, reset time and credit balance. Account-wide movement includes other consumers.
STATUS	text	The budget stop reason, when exhausted and requested by the caller.

Session snapshots

Support: public.

attach --json, /api/session/<id> and /api/draft/<name> read SessionState with the serializer's replacements and header fields below. The serializer removes task_label and renders log_tail as strings. The row additions belong to approvals and commits, not the snapshot root.

Field	Type	Meaning
session_id	JSON	The journaled session identifier with the directory name as fallback.
mode	JSON	The session mode with the manifest as fallback.
execution_id	str	The current execution's id, "" until its start is read.
tasks	tuple[TaskNodeView, ...]	The live task tree in DFS pre-order.
cursor_task_id	str \| None	The task the worker is on.
last_model_call	ModelCallView \| None	The last model call.
tool_calls	tuple[ToolCallView, ...]	The bounded tool-call history, most recent last.
last_verify	VerifyView \| None	The last verify gate run.
review	JSON	The shared review evidence with computed merge facts.
environment	EnvironmentView \| None	The container this execution holds, its start and its end; None outside one.
budget	JSON	The BudgetView with its rendered usd_text and plan_text.
approvals	tuple[ApprovalPrompt, ...]	Every approval asked this execution, answered or not.
questions	tuple[QuestionPrompt, ...]	Every question prompt asked this execution, answered or not.
log_tail	JSON	The retained log lines as plain strings rather than LogLine records.
log_count	int	The monotonic count of log lines ever; a live viewer diffs on it, since len(log_tail) freezes once the window saturates.
tool_count	int	The calls this execution dispatched, a finish tool apart, as the conversation's end line counts them; len(tool_calls) is a bounded window.
recent_diffs	tuple[DiffView, ...]	The bounded auto-commit diffs, for task filtering.
started	bool	An execution has begun; a parked or created run has none.
finished	bool	A session.stop was folded and no later execution began.
verification	Literal['passed', 'failed', 'unverified', 'not_applicable'] \| None	The final verification word, or None when unknown.
verify_scoped	bool	The judging gate ran scoped.
stop_reason	str	The session.stop reason word.
stranded	bool	This execution finished with its edits uncommitted and no commit landed.
end_summary	str	The end sentence the journal carries, "" when it carries none.
unattended_questions	int	Questions the harness answered empty because nobody was attached, in the current execution.
undone_to	str	The child session /undo forked, which surfaces follow.
undone_text	str	The message /undo took back, for the composer to refill.
finish_summary	str	The finish tool's summary, the agent's closing statement.
latest_diff	str	The patch of the most recent auto-commit.
commits	tuple[RunCommit, ...]	The run's per-turn commits, oldest first.
steer_requests	int	The monotonic count of mid-run steer requests; the TUI compares it with its own count to react once per press.
compact_elided	int	The elision markers in the current context.
compact_gists_live	int	The live gists in the current context; a demoted gist is back to a bare marker, and a tier-2 restart resets both.
pins	tuple[str, ...]	The operator's /pin instructions in force, most recent last.
last_event_at	float \| None	The last folded event's own ts, the idle anchor every "working… Ns" timer measures from, so replayed history reads as its true age.
context_used_percent	JSON	The last model call's context fill percentage, or null when unknown.
needs_new_work	JSON	Whether a resume requires a new task rather than continuing work.
open_approval	JSON	The current approval identifier, or null when none is open.
status	JSON	The lifecycle status word.
status_label	JSON	The rendered status with mode and unmerged marks when available.
task	JSON	The operator's words, whole.
continued_task	JSON	The latest resume steer, whole; empty without a continuation.
task_line	JSON	The task clipped to 200 characters, then a newline and continued: <steer>, clipped the same way, when a resume steer exists.
dead_state	JSON	The explanation for a worker that died without an end, or an empty string.
operator_blocked	JSON	Whether an unanswered blocking prompt is open.
live	JSON	Whether steer or stop can reach the session, or null without a directory.
policy	JSON	The rendered launch policy when a session directory is supplied.
ports	JSON	The worker's listening ports when a session directory is supplied.
shells	JSON	The background shell roster when a session directory is supplied.
plan_md	JSON	A plan session's deliverable when readable.
run_branch	JSON	The run's branch when it still exists.
base_branch	JSON	The recorded base branch when known.
worktree	JSON	The linked worktree path when recorded.
commits_ref	JSON	The ref holding the run's commits when available.
merged_into	JSON	The merge destination while the stamp still covers the run.
branch_line	JSON	The rendered branch and merge facts.
session_actions	JSON	Whether each session action applies to this session kind.
git_control	JSON	The launch git-control policy.
base_sha	JSON	The run chain's base commit.
unmerged	JSON	Whether the run holds unmerged work, when a repository is supplied.
children	JSON	The session's children by lineage, from the same listing as sessions show, for every session that has any.
dispatches	JSON	The session's dispatch records, each with its children's outcomes.
forked_from	JSON	The rendered fork lineage when recorded.
comparison	JSON	The group record from the parent, directly or through the child's group.
placement	JSON	The child's PlacementStamp when recorded.
placement.line	JSON	The rendered comparison placement.
as_of	JSON	The selected commit's turn and sha, or null for the current snapshot.
as_of.turn	JSON	The selected commit's turn number.
as_of.sha	JSON	The selected commit identifier.

Sources: session_state_as_dict, session_state_as_dict, branch_facts, manifest_header, session_snapshot.

attach --json approval and commit fields

Support: public. Return the wire form of a SessionState, what attach --json and the web serialize.

Source: viewmodel.session.view.session_state_as_dict.

Field	Type	Meaning
head	JSON	The question part added to each pending approval.
payload	JSON	The command part added to each pending approval.
label	JSON	The picker label added to each commit.

budget rendered fields

Support: public. Return the wire form of a BudgetView, the one attach --json and sessions show share.

Source: viewmodel.session.view.budget_dict.

Field	Type	Meaning
usd_text	JSON	The rendered session spend, then this execution's against its cap.
plan_text	JSON	The account's reading: the binding window's percent, its reset and when it was read; "" without a reading.

Machine snapshots

Support: public.

A machine's attach --json and /api/machine/<name> read MachineState with the computed fields below.

Field	Type	Meaning
machine	str	The spec's declared name.
version	int	The spec's version.
initial	str	The initial state's name.
current	str	Where the machine is, or is about to run.
states	tuple[MachineStateView, ...]	Every state in spec order, position-flagged.
transitions	tuple[TransitionView, ...]	The path taken, in order.
ended	MachineResult \| None	The end record, None while the machine has not ended.
notifications	tuple[NotificationView, ...]	The recent machine.notify events, oldest first.
status	JSON	The directory-aware machine status when a directory is supplied.
status_level	JSON	The status presentation level.
refusals	JSON	Reasons keyed by machine verb, with an empty reason for an allowed verb.
worker_lost	JSON	The missing-worker diagnostic for a stopped machine.
worker_lost.reason	JSON	The reason the machine reads stopped.
worker_lost.state	JSON	The state where the worker was lost.
spend	JSON	The machine's cumulative usage.
spend.cost_usd	JSON	Cumulative spend in USD.
spend.usd_partial	JSON	Whether unpriced usage makes spend a lower bound.
spend.text	JSON	The rendered spend.
spend.input_tokens	JSON	Recorded input tokens.
spend.output_tokens	JSON	Recorded output tokens.
spend.in_flight_state	JSON	The agent-state execution still in flight, or null.

Sources: machine_state_as_dict, machine_snapshot.

Web API

Web UI describes authentication, routes, limits and status codes. The response builders below name keys read directly from their dictionary literals and assignments; conditional keys appear only on the paths that write them. Request bodies use the typed Body records below. Git blob responses carry bytes rather than JSON.

/api/session/<id>/events frame fields

Support: public. Send a session snapshot when its state changes.

Source: ui.web.sse.stream_session.send_frame.

Field	Type	Meaning
stream_dead	JSON	Whether the worker died without a journaled end.
last_event_age_s	JSON	Seconds since the last event at frame emission.

/api/machine/<name>/events snapshot frame

Support: public. Stream one machine until it ends or the client leaves.

Source: ui.web.sse.stream_machine.

Field	Type	Meaning
machine	JSON	The machine snapshot with its cumulative spend.
reasoning	JSON	The newest agent-state session snapshot, empty before an execution.

/api/machine/<name>/events frames

Support: public. Stream one machine until it ends or the client leaves.

Source: ui.web.sse.stream_machine.

Field	Type	Meaning
machine	JSON	The machine snapshot with its cumulative spend.
reasoning	JSON	The newest agent-state session snapshot, empty before an execution.
type	JSON	The error frame kind when the machine cannot be read.
error	JSON	The machine-read failure diagnostic.

/api/machine/<name>/events reasoning age

Support: public. Fill the reasoning fold's idle age in from its epoch.

Source: ui.web.sse._with_idle_age.

Field	Type	Meaning
last_event_age_s	JSON	Seconds since the reasoning snapshot's last event at frame emission.

/api/meta payload

Support: public. Dispatch a GET by path.

Source: ui.web.server._Handler._route.

Field	Type	Meaning
version	JSON	The running agent6 version.
target	JSON	The target selected when the server started.
target_kind	JSON	Whether the target is a session, machine or draft.

/api/config/entry_tables payload

Support: public. Dispatch a GET by path.

Source: ui.web.server._Handler._route.

Field	Type	Meaning
tables	JSON	The supported named config tables.
tables[].path	JSON	The table's dotted config path.
tables[].label	JSON	The table's display name.
tables[].leaf	JSON	Whether each entry is a scalar instead of a subtable.
tables[].fields	JSON	The editable fields of an entry.
tables[].fields[].name	JSON	The field name accepted by the config writer.
tables[].fields[].choices	JSON	Suggested values for the field.
tables[].fields[].description	JSON	The config model's field meaning.
tables[].defaults	JSON	Known provider defaults, only on the providers table.

/api/config/suggest/<key> payload

Support: public. Dispatch a GET by path.

Source: ui.web.server._Handler._route.

Field	Type	Meaning
values	JSON	Suggestions for the requested config key.

/api/login result

Support: public. Answer a sign-in: the token for the server's secret, 403 for any other.

Source: ui.web.server._Handler._login.

Field	Type	Meaning
ok	JSON	Whether the secret was accepted.
token	JSON	The bearer token on a successful login.
error	JSON	The refusal reason on a failed login.

Web API sign-in refusal

Support: public. Send 401; login tells the page to show its sign-in form.

Source: ui.web.server._Handler._send_sign_in_required.

Field	Type	Meaning
error	JSON	The authentication refusal reason.
login	JSON	Whether the client must sign in.

Web POST result envelope

Support: public. Send the payload under ok, or the error as 422.

Source: ui.web.server._Handler._ok_or_err.

Field	Type	Meaning
ok	JSON	Whether the requested action succeeded.
error	JSON	The failure reason when the action failed.

Web POST result fields

Support: public. Dispatch a POST by path.

Source: ui.web.server._Handler._route_post.

Field	Type	Meaning
session_id	JSON	The session created by new work.
draft	JSON	The machine-authoring draft identifier.
message	JSON	The action's result text.

/api/session/<id>/fork result

Support: public. Fork a run at its latest checkpoint into a new, unstarted run.

Source: ui.web.actions.fork_run.

Field	Type	Meaning
new_session_id	JSON	The forked session identifier.

/api/session/<id>/undo result

Support: public. Fork a finished run at the state before its last operator message, unstarted.

Source: ui.web.actions.undo_session.

Field	Type	Meaning
new_session_id	JSON	The child session created by undo.
undone_text	JSON	The instruction removed by undo.
notice	JSON	The undo result message.

/api/session/<id>/review result

Support: public. Review a finished run's record through agent6 sessions review, a call of minutes.

Source: ui.web.actions.review_run.

Field	Type	Meaning
review	JSON	The completed run review as Markdown.

/api/hub payload

Support: public. Return the hub: sessions, machines, drafts, machine files, presets.

Source: ui.web.model.hub_payload.

Field	Type	Meaning
sessions	JSON	Session listing rows, newest activity first.
machines	JSON	Machine-instance listing rows.
machine_files	JSON	Authored machine files available to run.
drafts	JSON	Machine-authoring session listing rows.
presets	JSON	Available preset names.
preset_default_label	JSON	The composer's no-flag preset label.

/api/hub machine row

Support: public. Return one machine-instance row for the hub.

Source: ui.web.model._machine_row.

Field	Type	Meaning
name	JSON	The machine instance name.
mtime	JSON	The latest activity time in epoch seconds.
when	JSON	The rendered latest activity time.
status	JSON	The machine's status word.
status_label	JSON	The rendered status, with its reason when it has one.
status_level	JSON	The status presentation level.
machine	JSON	The declared machine name, absent for unreadable instances.
current	JSON	The current state, absent for unreadable instances.

/api/hub machine-file rows

Support: public. Return the hub's machine-file rows.

Source: ui.web.model.list_machine_files.

Field	Type	Meaning
[].path	JSON	The path accepted by machine run.
[].name	JSON	The machine file's basename.

/api/routes payload

Support: public. Return the new-work composer's model picker.

Source: ui.web.model.routes_payload.

Field	Type	Meaning
routes	JSON	The provider/model routes the config can run.
default_label	JSON	The composer's no-flag model label.

GET /api/session/<id>/merge payload

Support: public. Return what a merge confirm states about a run before the click.

Source: ui.web.model.merge_preview_payload.

Field	Type	Meaning
same_paths	JSON	The line naming the unmerged runs that changed paths this run changed, or "" when there is none.

/api/session/<id>/resume_defaults payload

Support: public. Return the resume row's no-flag preset and model labels under a picked preset.

Source: ui.web.model.resume_defaults_payload.

Field	Type	Meaning
preset_label	JSON	The resume's no-flag preset label.
model_label	JSON	The resume's no-flag model label under the selected preset.

/api/session/<id>/conversation payload

Support: public. Return a run's conversation and the operator's past inputs, from one read of the log.

Source: ui.web.model.conversation_payload.

Field	Type	Meaning
items	JSON	The folded conversation items.
operator_inputs	JSON	The operator's past inputs for composer history.

/api/session/<id>/conversation item

Support: public. Fold events into rendered conversation items.

Source: ui.web.model.conversation_items.

Field	Type	Meaning
kind	JSON	The transcript item kind.
lines	JSON	Collapsed lines of text/style span pairs.
full	JSON	Expanded lines of text/style span pairs, only when different from lines.

/api/session/<id>/restate payload

Support: public. Return /restate for the web composer, over the session's whole journal.

Source: ui.web.model.restate_payload.

Field	Type	Meaning
text	JSON	The rendered restatement of the session.

/api/machine/<name>/conversation payload

Support: public. Return the conversation of the machine's newest agent-state execution, or no items.

Source: ui.web.model.machine_conversation_payload.

Field	Type	Meaning
items	JSON	The newest agent-state conversation, empty before any execution.

/api/machine/<name>/reasoning fields

Support: public. Return the session state of the machine's newest agent-state execution.

Source: ui.web.model.machine_reasoning_snapshot.

Field	Type	Meaning
state_dir	JSON	The agent-state execution directory to echo when answering a prompt.

/api/session/<id>/diff payload

Support: public. Return one commit's patch, or the chain base..sha when cumulative and a base is known.

Source: ui.web.model._diff_payload.

Field	Type	Meaning
cumulative	JSON	Whether the patch covers the chain rather than one commit.
patch	JSON	The unified patch for the selected commit or chain.
old_spans	JSON	Base paths to per-line [start, end, class] spans in UTF-16 units. Line 1 is index 0; undisplayed lines and unhighlighted sides are empty.
new_spans	JSON	Head paths to per-line [start, end, class] spans in UTF-16 units. Line 1 is index 0; undisplayed lines and unhighlighted sides are empty.

/api/config setting

Support: public. Return one leaf's JSON view, shared by the full dump and the single-key path.

Source: viewmodel.config_view._leaf_json.

Field	Type	Meaning
value	JSON	The effective raw config value, masked when secret.
effective	JSON	The resolved config value, masked when secret.
default	JSON	The built-in config default.
source	JSON	The layer that supplied the value.
modified	JSON	Whether a config layer set the value.
adaptive	JSON	Whether the effective value was resolved from an automatic setting.
type	JSON	The config editor's type name.
choices	JSON	Allowed enum values, or null for open input.
description	JSON	The config model's field meaning.
choice_meanings	JSON	Short meanings keyed by fixed choice value.
display	JSON	The rendered effective value.
default_display	JSON	The rendered default value.
model_role	JSON	The role whose model picker the leaf opens; "" for every other leaf.
description_points	JSON	The description's points, one per line in an editor.

/api/config editing fields

Support: public. Return the effective config per leaf, keyed by dotted key.

Source: ui.web.model.config_payload.

Field	Type	Meaning
input	JSON	The value formatted for round-trip editing.
entry	JSON	The named config entry, only for entry-table settings.
entry.table	JSON	The entry table's dotted path.
entry.name	JSON	The entry's name.
entry.source	JSON	The highest config layer holding the entry.

/api/git/refs ref row

Support: public. Return one Git ref for the picker.

Source: ui.web.git_model._ref_row.

Field	Type	Meaning
name	JSON	The ref's display name.
full	JSON	The fully qualified ref name.
kind	JSON	The ref group used by the picker.
sha	JSON	The ref's commit identifier.
head	JSON	Whether this ref is the current branch.

Git changed-file row

Support: public. Return one changed file.

Source: ui.web.git_model._file_row.

Field	Type	Meaning
status	JSON	Git's change-status letter.
path	JSON	The file's path on the head side, or the base path for a deletion.
old_path	JSON	The base path of a renamed file, or an empty string.
additions	JSON	The number of added lines.
deletions	JSON	The number of deleted lines.
binary	JSON	Whether Git reports a binary change.
staged	JSON	Whether the index differs from HEAD for the file (working tree only).
unstaged	JSON	Whether the working tree differs from the index (working tree only).
untracked	JSON	Whether git does not track the file (working tree only).

/api/git/graph commit row

Support: public. Return one commit graph row.

Source: ui.web.git_model._log_row.

Field	Type	Meaning
sha	JSON	The full commit identifier.
short	JSON	The seven-character commit identifier.
parents	JSON	The parent commit identifiers in order.
author	JSON	The commit author's name.
date	JSON	The author date from Git.
when	JSON	The rendered commit time.
subject	JSON	The commit's first message line.
is_merge	JSON	Whether the commit has more than one parent.

/api/git/graph ref decoration

Support: public. Return the refs at each sha, in listing order.

Source: ui.web.git_model._decorations.

Field	Type	Meaning
name	JSON	The ref's display name.
kind	JSON	The ref's group.
head	JSON	Whether this ref is the current branch.

/api/git/compare endpoint

Support: public. Return one side of a compare: the rev as sent, the sha it names, and its subject.

Source: ui.web.git_model._endpoint.

Field	Type	Meaning
rev	JSON	The revision as requested.
sha	JSON	The resolved commit, empty tree or WORKTREE identifier.
short	JSON	The abbreviated identifier, or WORKTREE.
subject	JSON	The commit subject or working-tree label.

/api/git/refs payload

Support: public. Return every ref with its kind and tip, and HEAD's sha and branch.

Source: ui.web.git_model.refs_payload.

Field	Type	Meaning
head	JSON	The checkout's HEAD identity.
head.sha	JSON	The HEAD commit identifier.
head.branch	JSON	The current branch name.
refs	JSON	The selectable ref rows.
remotes	JSON	The remotes a fetch can name.
remotes[].name	JSON	The remote's name.
remotes[].url	JSON	Its URL, an http(s) URL's user and password left out.

/api/git/graph payload

Support: public. Return one page of the commit graph.

Source: ui.web.git_model.graph_payload.

Field	Type	Meaning
commits	JSON	Commit rows with their ref decorations.
commits[].refs	JSON	The ref decorations attached to this commit.
tips	JSON	The resolved commit tips to reuse for the next page.
has_more	JSON	Whether another commit page exists.
skip	JSON	The number of commits skipped.
limit	JSON	The effective page size.

/api/git/commit/<rev> payload

Support: public. Return a commit's header, message, refs and the files it changed from its first parent.

Source: ui.web.git_model.commit_payload.

Field	Type	Meaning
sha	JSON	The full commit identifier.
short	JSON	The seven-character commit identifier.
parents	JSON	The parent commit identifiers in order.
author	JSON	The commit author's identity and date.
author.name	JSON	The author's name.
author.email	JSON	The author's email address.
author.date	JSON	The author date from Git.
committer	JSON	The committer's identity and date.
committer.name	JSON	The committer's name.
committer.email	JSON	The committer's email address.
committer.date	JSON	The committer date from Git.
subject	JSON	The commit's first message line.
body	JSON	The commit's remaining message.
refs	JSON	The ref decorations at this commit.
is_merge	JSON	Whether the commit has more than one parent.
base	JSON	The first parent, or the empty tree for a root commit.
files	JSON	Changed-file rows against the base.
truncated	JSON	Whether the file list reached its cap.

/api/git/compare payload

Support: public. Return the files that differ between two revisions, as git diff base head lists them.

Source: ui.web.git_model.compare_payload.

Field	Type	Meaning
base	JSON	The base revision as an endpoint object.
head	JSON	The head revision as an endpoint object.
files	JSON	Changed-file rows between the endpoints.
truncated	JSON	Whether the file list reached its cap.

/api/git/worktree payload

Support: public. Return the working tree's uncommitted files, tracked changes first.

Source: ui.web.git_model.worktree_payload.

Field	Type	Meaning
base	JSON	HEAD's commit, or the empty tree on an unborn branch.
files	JSON	Changed-file rows, with tracked changes before untracked files.
truncated	JSON	Whether the file list reached its cap.

/api/git/diff payload

Support: public. Return one file's unified diff between two revisions.

Source: ui.web.git_model.diff_payload.

Field	Type	Meaning
base	JSON	The resolved base revision.
head	JSON	The resolved head revision, or WORKTREE.
base_digest	JSON	The base input's SHA-256 digest, or null when unavailable.
head_digest	JSON	The head input's SHA-256 digest, or null when unavailable.
full_context	JSON	Whether all unchanged lines were requested.
context	JSON	File text for a full-context change without text hunks, or null.
path	JSON	The file's head path, or its base path for a deletion.
old_path	JSON	The base path for a rename, or an empty string.
patch	JSON	The unified patch, subject to the byte cap.
old_spans	JSON	Base paths to per-line [start, end, class] spans in UTF-16 units. Line 1 is index 0; undisplayed lines and unhighlighted sides are empty.
new_spans	JSON	Head paths to per-line [start, end, class] spans in UTF-16 units. Line 1 is index 0; undisplayed lines and unhighlighted sides are empty.
binary	JSON	Whether Git reports a binary patch.
truncated	JSON	Whether the displayed content reached its cap.
max_bytes	JSON	The byte cap, or null when whole content was requested.

/api/git/search payload

Support: public. Return the refs whose name contains the text and the commits whose message or sha does.

Source: ui.web.git_model.search_payload.

Field	Type	Meaning
refs	JSON	Matching ref rows.
commits	JSON	Matching commit rows.
truncated	JSON	Whether either result list reached its cap.

/api/git/tree payload

Support: public. Return the files at a revision or in the working tree.

Source: ui.web.git_model.tree_payload.

Field	Type	Meaning
rev	JSON	The resolved revision, or WORKTREE.
files	JSON	TreeFile records in the selected tree.
ignored	JSON	Whether the working-tree listing includes Git's ignored files.
truncated	JSON	Whether the file list reached its cap.

/api/git/text payload

Support: public. Return a file's text, highlight spans and rendered markdown at a revision.

Source: ui.web.git_model.text_payload.

Field	Type	Meaning
rev	JSON	The resolved revision, or WORKTREE.
path	JSON	The repository-relative file path.
digest	JSON	The SHA-256 digest of the complete checked file bytes.
text	JSON	The file text, or an empty string for binary content.
rendered_html	JSON	Rendered markdown with raw HTML escaped; null for other or incomplete text.
lines	JSON	The number of displayed text lines.
binary	JSON	Whether the file is binary.
truncated	JSON	Whether the displayed text reached its cap.
max_bytes	JSON	The byte cap, or null when whole content was requested.
language	JSON	The tree-sitter language name, or an empty string when unknown.
spans	JSON	Per-line highlight spans as start, end and CSS class triples.

/api/git/blame payload

Support: public. Return per-line attribution with the text route's caps, refusals and digest.

Source: ui.web.git_model.blame_payload.

Field	Type	Meaning
rev	JSON	The resolved revision, or WORKTREE.
path	JSON	The repository-relative file path.
digest	JSON	The SHA-256 digest of the complete checked file bytes.
lines	JSON	BlameLine records for the displayed text lines.
binary	JSON	Whether the file is binary.
truncated	JSON	Whether the displayed text reached its cap.
max_bytes	JSON	The byte cap, or null when whole content was requested.

/api/git/findings payload

Support: public. Return findings with exact comparison matches and links to their recorded targets.

Source: ui.web.review_routes.comparison.

Field	Type	Meaning
findings	JSON	One comparison entry per recorded finding.
read_failures	JSON	Diagnostics for unreadable review evidence.

/api/git/findings finding

Support: public. Return findings with exact comparison matches and links to their recorded targets.

Source: ui.web.review_routes.comparison.

Field	Type	Meaning
session_id	JSON	The finding's session, empty for a saved standalone review.
version	JSON	The review version used for action concurrency checks.
panel	JSON	The panel containing the selected observation.
finding	JSON	The finding's shared history and status.
observation	JSON	The selected observation of the finding.
matches	JSON	Whether both displayed trees match the observation's target.

/api/reviews/<panel_id> payload

Support: public. Read a recorded comparison or file without consulting a mutable checkout.

Source: ui.web.review_routes.panel.

Field	Type	Meaning
error	JSON	The refusal diagnostic on an unsuccessful read.
session_id	JSON	The session owning the panel, or an empty string for a saved review.
review	JSON	The shared review evidence.
panel	JSON	The selected PanelView.
files	JSON	Changed-file records between the reviewed trees.
truncated	JSON	Whether the changed-file list reached its cap.

/api/reviews/<panel_id>/diff payload

Support: public. Return a file diff from the recorded review trees.

Source: ui.web.review_routes._diff.

Field	Type	Meaning
base	JSON	The recorded base tree.
head	JSON	The recorded head tree.
path	JSON	The head-side path, or the base path for a deletion.
old_path	JSON	The base path of a renamed file, or an empty string.
base_digest	JSON	The base input's SHA-256 digest, or null when unavailable.
head_digest	JSON	The head input's SHA-256 digest, or null when unavailable.
patch	JSON	The unified patch between the recorded trees.
old_spans	JSON	Base paths to per-line [start, end, class] spans in UTF-16 units. Line 1 is index 0; undisplayed lines and unhighlighted sides are empty.
new_spans	JSON	Head paths to per-line [start, end, class] spans in UTF-16 units. Line 1 is index 0; undisplayed lines and unhighlighted sides are empty.
binary	JSON	Whether Git reports a binary patch.
truncated	JSON	Whether the patch reached its byte cap.
max_bytes	JSON	The byte cap, or null when whole content was requested.
error	JSON	The refusal diagnostic on an unsuccessful read.

/api/reviews/<panel_id>/file payload

Support: public. Return a file from one recorded review tree.

Source: ui.web.review_routes._file.

Field	Type	Meaning
rev	JSON	The recorded tree on the requested side.
path	JSON	The repository-relative file path.
text	JSON	The displayed file text, empty for binary content.
binary	JSON	Whether the file is binary.
truncated	JSON	Whether the text reached its byte cap.
max_bytes	JSON	The byte cap, or null when whole content was requested.
digest	JSON	The SHA-256 digest of the complete file bytes.
lines	JSON	The number of displayed lines.
language	JSON	The tree-sitter language name, or an empty string when unknown.
spans	JSON	Per-line highlight spans as start, end and CSS class triples.
error	JSON	The refusal diagnostic on an unsuccessful read.

/api/session/<id>/findings review fields

Support: public. Return JSON-ready evidence, including nested boundary models and merge facts.

Source: viewmodel.review.as_dict.

Field	Type	Meaning
summary	JSON	The shared review label, distinguishing absent panels from unreadable evidence.
merge_guard	JSON	The open-block and unreadable-evidence merge facts.
rule_sources	JSON	The latest panel's recorded rule sources.
rules_complete	JSON	Whether all rule sources were supplied, or null when unrecorded.

/api/session/<id>/findings finding fields

Support: public. Return JSON-ready evidence, including nested boundary models and merge facts.

Source: viewmodel.review.as_dict.

Field	Type	Meaning
line_origin	JSON	The latest observation's attribution evidence, added to each finding.

/api/session/<id>/receipt fields

Support: public. Return the JSON-safe receipt fields.

Source: viewmodel.receipt.as_dict.

Field	Type	Meaning
revert_request	JSON	A proposed task for preparing a reversal, not a command to execute.
text	JSON	The shared receipt rendering without retained output bodies.

Reviewer choices

Support: public. /api/config/reviewer_choices returns the enum choices from config.write.reviewer_choices, keyed by reviewer field.

Field	Type	Meaning
effort	list[str]	reasoning effort; route's value when unset; effort = "off" an explicit override

Typed records

SessionManifest is the private, durable manifest.json record. HarnessStamp.gate separates the pinned command and source from execution availability. ComparisonStamp holds group facts on the parent. PlacementStamp links each child to it. Other records are nested objects used by the CLI and web serializers above. Properties are absent unless a serializer adds them explicitly.

ChildLaunch

Support: public. What a child needs to start, and what it knows about itself that no config layer holds.

Source: agent6.app.child_launch.ChildLaunch.

Field	Type	Meaning
worktree	pathlib.Path	The child's clone, the run's cwd.
record_state	pathlib.Path	The state dir of the repository that dispatched the child, where the child's record lives from its first event.
session_id	str	The child's session id.
task	str	The task text, exactly as the child's manifest records it.
pins	tuple[str, ...]	The parent's pins, out of band of the task.
max_usd	float \| None	The parent's forwarded --max-usd; None keeps the config's.
allowance	BudgetAllowance \| None	What the parent had left when the child started, which sets every cap; None for the config's caps.
auto_approve	bool	Whether the parent ran with --auto-approve.
away	Literal['wait', 'deny', 'approve']	The child's away mode: wait parks a question for agent6 attach, deny answers it empty.
config	pathlib.Path	The child's config snapshot, beside this file; the child's sole --config.
notes	tuple[str, ...]	The snapshot's serialization loss notes, recorded in the child's manifest.
parent	str	The session every listing nests the child under.
group	str	The group the child belongs to.
number	int	The child's number, from 1.
gate_source	Literal['', 'configured', 'inferred', 'adopted']	Where the verify gate in the snapshot came from; "" for none.
operator_task	str \| None	The task's words before composition; None when they are the task.
limits	ChildLimits \| None	The turn cap and deadline the child runs under; None for the config's own.
contract	ConversationContract \| None	The bounded conversation a reviewer child drives; None for a run child.

CompletionRecord

Support: public. What a bounded conversation delivered, written once at its end.

Source: agent6.sessions.completion.CompletionRecord.

Field	Type	Meaning
text	str	The last response's text, or the finish summary when the model finished.
stop_reason	str	The session's end reason, as the journal's session.stop carries it.
output_tokens	int	The output tokens of every attempt.
calls	int	The provider attempts made, retries included.
diagnostics	tuple[str, ...]	What the ceiling or the contract cut, one line each; empty when nothing was.
session_id	str	The producing read-only session, empty in legacy records.
review_binding	ReviewBinding \| None	The parent-assigned panel, reviewer and target, or null outside a panel.
response_stop_reason	str	The last provider response's stop reason.
response_output_tokens	int	The last response's output tokens.
reasoning_chars	int	The last response's reasoning character count.
asked_for_tools	bool	Whether the last response requested tools.

DispatchRecord

Support: public. The record of one dispatch.

Source: agent6.sessions.child_dispatch.DispatchRecord.

Field	Type	Meaning
version	int	The record's shape; a newer one refuses the read.
identity	str	The dispatch's id, the group id <parent>-p<n>.
source	Literal['operator', 'model', 'panel']	Who asked.
join_policy	Literal['merge', 'record_only', 'rank']	Ordered work merge, harness-owned record-only import, or the best-ranked child alone.
ranking	RankRecord \| None	A rank join's order; None before it ranked and under the other policies.
owner	str	The execution that made it.
turn	int	The turn it was dispatched after.
base	str	The chain tip the children were cut from.
segments	tuple[SegmentRecord, ...]	The requested tasks.
children	tuple[ChildRecord, ...]	The children, in child order.
state	Literal['prepared', 'running', 'settling', 'joining', 'complete']	How far the dispatch got.
message	str	The result delivered for it, "" until complete.
delivered	bool	Whether the result reached the conversation.
written_at	float	When this version of the record was written, as epoch seconds.

SessionManifest

Support: public. The typed manifest.json a session starts with and later stamps.

Source: agent6.sessions.manifest.SessionManifest.

Field	Type	Meaning
version	int	The manifest shape, MANIFEST_VERSION.
agent6_version	str	The agent6 that wrote it.
session_id	str	The session id.
mode	str	The session mode. No default: it is the privilege gate's only input, and a manifest that lost the key must not read as the more-privileged run.
start_ts	str	When the session started.
user_task	str	A 4000-character fallback for listings whose journal is missing.
continued_task	str	The latest resume steer, clipped the same way; the fallback for the continuation line.
base_sha	str	The commit the run's un-landed work starts from: where it started, or the commit a resume continued from after its work landed.
base_branch	str	The branch that commit was on.
review_base_tree	str	The base tree of a /review session, whose AGENTS.md binds it.
run_branch	str \| None	The run's own branch, or None.
git_control	str	[git].control at start; a model run has no per-turn chain or branch. The git surfaces refuse through model_git_refusal.
models	ModelsBrief	The models the run resolved.
harness	HarnessStamp	The strategy the run started with.
policy	PolicyStamp	The policy the run launched under.
execution_targets	dict[str, ExecutionTargetStamp]	Recorded target facts by reference, empty in historical manifests.
budget	BudgetStamp	The caps the run's or a resume's budget flags set (--max-usd and its siblings).
parked_task	str	The verbatim task of a run submitted but never started; agent6 resume starts it fresh and its manifest rewrite clears this.
parked_operator_task	str \| None	The parked task before composition, when recorded.
parked_reason	str	Why it was parked: the checkout was busy, or the tree was dirty and the operator chose to wait.
source_session_id	str \| None	The run --from source, which contributes context but not this session's mode, checkout or history.
parent_session_id	str \| None	The fork parent, or None.
forked_from_turn	int \| None	The turn the fork rolled back to, or None.
forked_from_sha	str \| None	The commit the fork started on, or None.
worktree	pathlib.Path \| None	The absolute path of the linked worktree agent6 fork or run --worktree added; None for a session in the operator's checkout. An /undo fork names its source's.
worktree_git_dir	pathlib.Path \| None	The repository git dir granted to a linked checkout's jail, recorded before any model command, including for an operator-owned checkout. Resume and same-checkout forks inherit it; a writable .git pointer cannot replace it.
worktree_set_up	bool	Whether the setup command ran in the session's own worktree or clone; a resume runs it only while this is False. A fork's new worktree starts False, and an /undo fork carries its source's.
merged	MergeStamp \| None	The merge record, None until the run branch is merged.
includes	IncludedRun \| None	The unmerged run this run took in under dirty_tree = "include", stamped merged when this run lands; None when it took none, or more than that run's tip.
parallel	ParallelLineage \| None	A child's lineage, or None.
placement	PlacementStamp \| None	A child's rank and group reference, or None.
comparison	ComparisonStamp \| None	The group's ranking and judge spend on its parent, or None.

agent6 ps JSON row

Support: public. One live session or machine instance.

Source: agent6.ui.cli.ps_cmd._Row.

Field	Type	Meaning
directory	str \| None	The absolute checkout path, or None when the id does not lead back to one.
repo_id	str	The state dir's name for the repo.
id	str	The session id or machine name.
mode	str	The session mode, or "machine".
status	str	The raw status word, without its detail.
status_detail	str	The detail explaining the status, or "" when there is none.
pid	int \| None	The worker pid, when readable.
attached	bool	A front end is live on the session.
parent	str	For a child, the live parent row it nests under.
children	tuple[agent6.ui.cli.ps_cmd._Row, ...]	The group's live children, nested.
worktree	str \| None	The linked worktree the run works in, None for a run in the checkout.

MergeReviewGuard

Support: public. Report open grounded blocks and unreadable evidence without enforcing a merge.

Source: agent6.viewmodel.review.MergeReviewGuard.

Field	Type	Meaning
open_block_ids	tuple[str, ...]	The open grounded findings that block a merge.
read_failure	str \| None	The evidence-read diagnostic that blocks a merge, or null.

ReceiptView

Support: public. Keep a session's evidence derived from its journal and manifest.

Source: agent6.viewmodel.receipt.ReceiptView.

Field	Type	Meaning
session_id	str	The session identifier.
execution_id	str \| None	The last execution identifier, or null when unrecorded.
source_tip	str \| None	The source checkpoint commit, or null when unrecorded.
end_tree	str \| None	The final observed tree, or null when unrecorded.
verifications	tuple[VerificationEvidence, ...]	The recorded VerificationEvidence attempts.
final_verify_id	str \| None	The attempt covering the final tree, or null when absent.
review	ReviewView	The review evidence compared against the final tree.
revert	RevertPlan	The reversal plan derived from recorded merge facts.
evidence_notes	tuple[str, ...]	Diagnostics for absent or unreadable evidence.
merged	MergeStamp \| None	The MergeStamp, or null without a recorded landing.
targets	tuple[ExecutionTargetEvidence, ...]	The recorded execution targets, without runtime queries.
setups	tuple[SetupEvidence, ...]	Every recorded setup result.
jobs	tuple[JobsObservation, ...]	Job-settlement observations.
environments	tuple[EnvironmentObservation, ...]	Environment lifecycle observations.
acquisitions	tuple[AcquisitionObservation, ...]	The container leases the run's executions acquired.
children	tuple[ChildView, ...]	The session's children, from its dispatch records or its children's lineage.

ChildView

Support: public. One child of a session, as the surfaces show it.

Source: agent6.viewmodel.children.ChildView.

Field	Type	Meaning
session_id	str	The child's session id.
number	int	Its number in its group.
group	str	The group it was dispatched in.
source	str	Who dispatched it: operator, model or panel; "" when no record says.
task	str	Its task text; "" when unknown.
model	str \| None	Its selector, None for the worker's own settings.
route	str	The provider/model its manifest records; "" when it never started.
status	str	Its own status word; "" when it never started.
outcome	str	How it joined: joined, no_changes, conflict or failed; "" when no join is recorded, as for a group's children.
join_sha	str	The merge commit, "" when none.
imported_ref	str	The branch the join read, "" when none.
error	str	The join's detail, "" when none.
spend_usd	float	What its parent settled for it; the child's own cost when no record says.
usd_partial	bool	Whether unpriced calls make the spend a lower bound.
settled	bool	Whether the parent charged its usage; False while it is unknown.
launched	bool	Whether a session dir of its own exists.

DispatchView

Support: public. One dispatch record, as the surfaces show it.

Source: agent6.viewmodel.children.DispatchView.

Field	Type	Meaning
identity	str	The group id.
source	str	Who dispatched it.
state	str	The record's state, or unreadable.
turn	int	The turn it was dispatched after.
base	str	The chain tip the children were cut from.
message	str	The delivered result, or why the record could not be read.
children	tuple[ChildView, ...]	Its children, in child order.

FindingAction

Support: public. Validate an operator request against a known review-state version.

Source: agent6.review_records.FindingAction.

Field	Type	Meaning
request_id	str	The caller's identifier for idempotent action delivery.
expected_version	int	The review version the caller read.
action	Literal['fix', 'resolve', 'reopen']	The action kind supplied by the route rather than the HTTP body.
resolution	FindingResolution \| None	The conclusion for a resolve action, or null.
reason	str \| None	The explanation for a reopen action, or null.
resume	bool	Whether the operator approved resuming a stopped session to fix the finding.

ActionResult

Support: public. Return the recorded outcome, or an explicit pending delivery.

Source: agent6.app.review.actions.ActionResult.

Field	Type	Meaning
request_id	str	The caller's action identifier.
finding_id	str	The finding the action targets.
version	int	The review version after the action.
status	Literal['resolved', 'reopened', 'submitted', 'failed', 'delivery_unknown', 'queued']	The action's delivery or completion status.
message	str	The result or next-action explanation.

BlameLine

Support: public. One line's last commit; an empty sha marks uncommitted content.

Source: agent6.vcs.git.files.BlameLine.

Field	Type	Meaning
line	int	The one-based displayed line number.
sha	str	The attribution commit, empty for uncommitted lines.
author	str	The attribution author, or not committed for uncommitted lines.
date	str	The attribution date in UTC ISO 8601, empty for uncommitted lines.
filename	str	The file's historical path in the attribution commit.
original_line	int	The one-based historical line number.

TreeFile

Support: public. One entry of a tree listing.

Source: agent6.vcs.git.files.TreeFile.

Field	Type	Meaning
path	str	The repository-relative path.
size	int	The size in bytes; a symlink's is its target's length, a submodule's 0.
kind	str	"file", "symlink" or "submodule".
state	str	"tracked", "untracked", "modified" or "ignored" in the working tree, else "".
target	str	A symlink's target as git or the filesystem stores it, else "".

NewWorkBody

Support: public. The /api/new body: a session of mode on task, under a preset and a model.

Source: agent6.ui.web.server.NewWorkBody.

Field	Type	Meaning
mode	str	The session mode to start.
task	str	The operator's task text; empty only with seed_from.
preset	str	The preset override, empty for configured selection.
model	str	The provider/model override, empty for configured selection.
worktree	bool	Work in a new git worktree (run --worktree); run mode only.
seed_from	str	The session whose context seeds the run (run --from); a plan becomes the task when task is empty.
max_usd	float \| None	The metered spend ceiling (--max-usd), or null for the config's.

SteerBody

Support: public. A steer body.

Source: agent6.ui.web.server.SteerBody.

Field	Type	Meaning
text	str	The instruction.
state	str	For a machine, the state dir the prompt was rendered from; "" is the newest.

ApproveBody

Support: public. An approval answer.

Source: agent6.ui.web.server.ApproveBody.

Field	Type	Meaning
id	str	The prompt's id.
answer	Literal['yes', 'no', 'session', 'session-deny']	The operator's literal choice. The asking side decides what a session answer grants.
state	str	For a machine, the state dir the prompt was rendered from; "" is the newest.

AnswerBody

Support: public. An ask_user answer: one entry per question, by index.

Source: agent6.ui.web.server.AnswerBody.

Field	Type	Meaning
id	str	The question prompt identifier.
answers	list[str]	One answer per question in prompt order.
state	str	The machine execution directory, empty to select the newest.

MergeBody

Support: public. A merge body; an empty method takes the config's.

Source: agent6.ui.web.server.MergeBody.

Field	Type	Meaning
method	str	The merge method override, empty to use config.

PruneBody

Support: public. A prune body; delete_squashed is the CLI's own opt-in flag.

Source: agent6.ui.web.server.PruneBody.

Field	Type	Meaning
delete_squashed	bool	Whether to delete branches confirmed squash-merged by their manifests.

StopBody

Support: public. A stop body; after_turn lets the current turn's results and auto-commit land.

Source: agent6.ui.web.server.StopBody.

Field	Type	Meaning
after_turn	bool	Whether to let the current turn finish before stopping.

ResumeBody

Support: public. A resume body; each empty field means as the run recorded.

Source: agent6.ui.web.server.ResumeBody.

Field	Type	Meaning
text	str	The continuation instruction, empty to continue without new work.
preset	str	The preset override, empty to use the recorded selection.
model	str	The provider/model override, empty to use the recorded selection.

MachineCreateBody

Support: public. A machine create body.

Source: agent6.ui.web.server.MachineCreateBody.

Field	Type	Meaning
task	str	The machine-authoring task.

MachineRunBody

Support: public. A machine run body naming a listed machine file.

Source: agent6.ui.web.server.MachineRunBody.

Field	Type	Meaning
file	str	The path of a listed machine definition.

MachinePokeBody

Support: public. A poke body; a JSON data payload wins over a message string.

Source: agent6.ui.web.server.MachinePokeBody.

Field	Type	Meaning
message	str	The text message, used when data is null.
data	Any	The JSON message payload, or null to use message.

EntryBody

Support: public. The add-entry form's table, name and string fields.

Source: agent6.ui.web.server.EntryBody.

Field	Type	Meaning
table	str	The named config table's dotted path.
name	str	The entry name.
fields	dict[str, str]	Entry field values as strings.
repo	bool	Whether to write repo config rather than global config.

EntryRemoveBody

Support: public. The entry and config layer to remove it from.

Source: agent6.ui.web.server.EntryRemoveBody.

Field	Type	Meaning
table	str	The named config table's dotted path.
name	str	The entry to remove.
repo	bool	Whether to remove from repo config rather than global config.

ReviewerBody

Support: public. The add-reviewer form's four fields and save target.

Source: agent6.ui.web.server.ReviewerBody.

Field	Type	Meaning
focus	str	The reviewer's required focus text.
model	str	A provider/model route, empty to inherit the reviewer role.
effort	str	The reasoning effort override, empty for the default.
temperature	str	The sampling temperature as text, empty for the default.
repo	bool	Whether to write repo config rather than global config.

RemoveReviewerBody

Support: public. The zero-based reviewer position and save target.

Source: agent6.ui.web.server.RemoveReviewerBody.

Field	Type	Meaning
index	int	The zero-based reviewer position to remove.
repo	bool	Whether to write repo config rather than global config.

LoginBody

Support: public. A sign-in body: the secret as typed, normalized before it is compared.

Source: agent6.ui.web.server.LoginBody.

Field	Type	Meaning
secret	str	The server secret as typed, normalized before comparison.

PresetBody

Support: public. A preset table to create or remove.

Source: agent6.ui.web.server.PresetBody.

Field	Type	Meaning
name	str	The preset name.
repo	bool	Whether to create it in the repo config rather than the global one.

FixManyBody

Support: public. A Fix of several findings, or of every open one without a queued Fix.

Source: agent6.ui.web.server.FixManyBody.

Field	Type	Meaning
finding_ids	tuple[str, ...]	The findings to fix, in order; empty with all_open.
all_open	bool	Fix every open finding without a queued Fix.
request_id	str	The caller's retry identity.
resume	bool	Whether the operator confirmed resuming an idle session to fix.

GitPathsBody

Support: public. Files for a Git page stage or unstage.

Source: agent6.ui.web.server.GitPathsBody.

Field	Type	Meaning
paths	tuple[str, ...]	The repo-relative paths, at least one.

GitPathBody

Support: public. One tracked file for a Git page discard.

Source: agent6.ui.web.server.GitPathBody.

Field	Type	Meaning
path	str	The repo-relative path.

GitCommitBody

Support: public. A Git page commit of the index.

Source: agent6.ui.web.server.GitCommitBody.

Field	Type	Meaning
message	str	The commit message.

GitBranchBody

Support: public. A Git page branch to create or switch to.

Source: agent6.ui.web.server.GitBranchBody.

Field	Type	Meaning
name	str	The branch name.
start_point	str	For a new branch, the revision to cut from; "" is HEAD.

GitFetchBody

Support: public. A Git page fetch from one remote.

Source: agent6.ui.web.server.GitFetchBody.

Field	Type	Meaning
remote	str	The remote's name.
pull	int \| None	A pull request's number: its refs/pull/<n>/head lands at <remote>/pr/<n>. None fetches the remote's branches.

ConfigSetBody

Support: public. A config write; unset removes the key from the target layer instead.

Source: agent6.ui.web.server.ConfigSetBody.

Field	Type	Meaning
key	str	The dotted config key.
value	str	The string value parsed by the config writer.
repo	bool	Whether to write repo config rather than global config.
unset	bool	Whether to remove the setting instead of assigning value.

BudgetAllowance

Support: public. One allowance in the three currencies, each with the cap's meaning.

Source: agent6.budget.BudgetAllowance.

Field	Type	Meaning
usd	float	The USD cap: -1 unlimited, 0 refuse, > 0 the amount.
tokens	int	The fallback-token cap, the same way.
plan_points	float	The plan percentage-point cap, the same way.

ChildLimits

Support: public. The ceiling a child runs under.

Source: agent6.child_records.ChildLimits.

Field	Type	Meaning
max_turns	int \| None	The turn cap the child's harness.max_turns is clamped to; None keeps the config's value.
deadline_at	float \| None	The wall-clock second (epoch) the child must have ended by; None for none. Wall-clock, since it crosses the process boundary into the child.

ConversationContract

Support: public. Bound one conversation the Harness drives in place of a worker's.

Source: agent6.kinds.ConversationContract.

Field	Type	Meaning
system	str	The system prompt, in place of the worker's assembled one.
opening	str	The first user message, in place of the task block.
max_turns	int	The turns allowed; a last turn that still asks for tools ends the conversation max_turns with its text kept and those tools not run.
temperature	float \| None	The sampling temperature; None keeps the worker's.
review_binding	ReviewBinding \| None	Parent-owned review provenance, or None for other conversations.
allowed_tools	tuple[str, ...]	The tools the parent hands down: the ones its own model is offered now, within the reviewer mode's surface.

ReviewBinding

Support: public. Bind a completion to the parent-assigned panel, reviewer and immutable target.

Source: agent6.review_records.ReviewBinding.

Field	Type	Meaning
panel_id	str	The parent-assigned panel identity.
reviewer_index	int	The zero-based roster position.
target	ReviewTarget	The exact retained comparison.

RankRecord

Support: public. The order a rank join landed by, written before the first join so no resume ranks twice.

Source: agent6.sessions.child_dispatch.RankRecord.

Field	Type	Meaning
order	tuple[str, ...]	The ranked children's session ids, best first; empty when no child had a result.
ranked_by	Literal['judge', 'mechanical']	judge or mechanical.
rationale	str	The judges' words; "" for the mechanical order.
winner	str	The one child the join lands; "" when the order crowns none.

SegmentRecord

Support: public. One requested task and its child spec.

Source: agent6.sessions.child_dispatch.SegmentRecord.

Field	Type	Meaning
task	str	The task text.
spec	str	The child spec as given.
children	int	How many children the segment expanded to.
node_id	str \| None	The DAG node the segment got, None without a graph.

ChildRecord

Support: public. One child of the dispatch.

Source: agent6.sessions.child_dispatch.ChildRecord.

Field	Type	Meaning
session_id	str	The child's session id, fixed before launch.
number	int	Its number in the group.
task	str	Its task text.
model	str \| None	Its selector, None for the worker's own settings.
limits	ChildLimits \| None	Its turn cap and deadline, None for the config's own.
join	JoinRecord \| None	How it joined, None until it did.
report	str	What it said at its end, its finish summary; "" until it joined or when it gave none.
usage	UsageRecord \| None	What it used, None until settled with known usage.
settled	bool	Whether its usage was charged; False while it is unknown.

ModelsBrief

Support: public. The models the run resolved.

Source: agent6.sessions.manifest.ModelsBrief.

Field	Type	Meaning
driver	ModelBrief \| None	The model that drove the run, the worker or a plan's planner; None unset.
reviewer	ModelBrief \| None	The reviewer; None unset.
helper	ModelBrief \| None	The helper; None unset.
reviewers	dict[str, ModelBrief]	Each reviewer's facts, in roster order, labelled by index and focus.
driver_from_flag	bool	The driver came from a --model, the run's or a resume's.

HarnessStamp

Support: public. The in-loop strategy the run started with, so resume re-applies it.

Source: agent6.sessions.manifest.HarnessStamp.

Field	Type	Meaning
review_when	str	[review].when at launch.
preset	str	The preset in force at launch.
gate	GateStamp	The pinned gate, its source and this execution's availability.
preset_from_flag	bool	The preset came from --preset rather than a config file.

PolicyStamp

Support: public. The policy the run launched under, so every surface and agent6 exec read one record.

Source: agent6.sessions.manifest.PolicyStamp.

Field	Type	Meaning
run_commands	str	[sandbox].run_commands at launch.
isolation	str	[sandbox].isolation at launch.
network	str	The network the run's commands join, session or host; "" when the stamp names no level. A manifest an older agent6 wrote holds the configured word.
commit_per_turn	bool	[git].commit_per_turn for the live execution; False means a dirty tree at the end is the deliverable, never a stranded commit.

ExecutionTargetStamp

Support: public. The persistence-safe projection of an ExecutionTarget, with unknown facts preserved.

Source: agent6.kinds.ExecutionTargetStamp.

Field	Type	Meaning
transport	str	The recorded command transport, distinct from local confinement.
host_root	pathlib.Path	The host workspace used by file tools.
command_root	pathlib.Path	The workspace used by commands.
user	str \| None	The selected command user, or None when unobserved.
home	str \| None	The command HOME, or None when unobserved.
environment	ExecutionEnvironmentPolicy	Inheritance policy and variable names, without values.
controls	tuple[EffectiveControl, ...]	Configured and effective controls with their enforcement owners.
runtime_id	str \| None	Observed runtime identity, or None when unknown.
container_id	str \| None	Observed container identity, or None when unknown.
launcher_instance	str \| None	Supervisor-reported instance identity, or None when unknown.
protocol_version	int \| None	Observed supervisor protocol version, or None without a handshake.
launcher_digest	str \| None	The launcher binary's sha256:<hex> where it was transferred, or None when unobserved.
lifecycle_ref	str \| None	Recorded lifecycle ownership reference, or None when unknown.

BudgetStamp

Support: public. The caps the budget flags set; a resume without a flag runs under the recorded caps.

Source: agent6.sessions.manifest.BudgetStamp.

Field	Type	Meaning
max_usd	float \| None	--max-usd, None when no flag set it.
max_tokens_fallback	int \| None	--max-tokens-fallback, None when no flag set it.
max_plan_points	float \| None	--max-plan-points, None when no flag set it.

MergeStamp

Support: public. The record of the run branch's merge.

Source: agent6.sessions.manifest.MergeStamp.

Field	Type	Meaning
into	str	The base branch.
sha	str	The merge commit in the base, or NO_MERGE_COMMIT.
ts	str	When it was merged.
tip	str	The run branch tip that was merged. sessions prune --delete-squashed deletes only while the branch still points here: a resumed run's later commits exist in no other ref.
into_tip	str	For a merge that added nothing, the base's own tip at the time, the commit that already held the run's content; the delete checks it as it checks sha.
via	str	The run whose merge landed this one, when a later run took its work in under dirty_tree = "include"; "" for the run's own merge.
method	Literal['ff_only', 'merge', 'squash', 'tasks'] \| None	The landing method, or null when unrecorded.
target_before	str \| None	The target commit before landing, or null when unrecorded.

IncludedRun

Support: public. An unmerged run whose whole work this run's start tree held and its commits carry.

Source: agent6.sessions.manifest.IncludedRun.

Field	Type	Meaning
session_id	str	That run.
tip	str	Its chain tip at this run's start, the tree this run started on.

ParallelLineage

Support: public. A child's place.

Source: agent6.sessions.manifest.ParallelLineage.

Field	Type	Meaning
group	str	The group it was dispatched in: the parent's id for run --parallel, <parent>-p<n> for a /parallel group.
number	int	Its number in the group.
parent	str	The session that dispatched it, which every listing nests it under.
config_notes	tuple[str, ...]	Resolved sampling values the child's TOML snapshot could not preserve.

PlacementStamp

Support: public. A child's auto-compare placement.

Source: agent6.sessions.manifest.PlacementStamp.

Field	Type	Meaning
rank	int	The child's place, 1 first.
of	int	How many children were ranked.
winner	bool	The child ranked first.
group	str	The comparison group recorded on its parent.

ComparisonStamp

Support: public. A group's ranking and judge spend, owned by its parent.

Source: agent6.sessions.manifest.ComparisonStamp.

Field	Type	Meaning
group	str	The comparison group id.
ranking	tuple[str, ...]	The ranked child ids, best first.
ranked_by	Literal['judge', 'mechanical']	The ranking method, judge or mechanical.
rationale	str	The judge's reason, at most 2000 characters; empty for mechanical ranking.
judge_cost_usd	float	The group's judge spend in USD, zero when no judge call was made.
judge_cost_partial	bool	The cost is a lower bound because some judge usage was unpriced.

SessionCore

Support: public. The facts both folds share: the lifecycle and the usage.

Source: agent6.viewmodel.session.records.SessionCore.

Field	Type	Meaning
lifecycle	Lifecycle	Where the session stands.
budget	BudgetView	Its usage, each amount with one lifetime.

TaskNodeView

Support: public. One node of the live task tree, flattened in DFS pre-order with its depth.

Source: agent6.viewmodel.session.records.TaskNodeView.

Field	Type	Meaning
id	str	The graph's id for the task.
title	str	The task's title.
status	Literal['pending', 'in_progress', 'passed', 'failed', 'skipped', 'obsolete']	The task's status word.
depth	int	The nesting depth, for tree rendering.
is_cursor	bool	The worker is on this task.
created_by	str	Who added the task; "" for a session dir written before the field existed.
standing	bool	The task is the standing goal.
note	str	The operator's mark beside a task they own ("queued by you", "standing goal"), or "".
short_id	str	The id as the operator reads and types it.
glyph	str	The status as every surface draws it; the cursor's task draws as in progress.

ModelCallView

Support: public. The last model call and what it streamed.

Source: agent6.viewmodel.session.records.ModelCallView.

Field	Type	Meaning
role	str	The role that called.
model	str	The model called.
in_flight	bool	The call has not returned.
provider	str	The provider that dialled the model; pairs with model for the registry's context-window lookup.
context_tokens	int	The full prompt in tokens at the last completed call (fresh input plus cache reads and writes); 0 until a result lands.
streamed_text	str	The live text, reset on every call, appended per delta, kept to the last state._STREAM_TAIL characters.
streamed_thinking	str	The live reasoning, with the same lifecycle.

ToolCallView

Support: public. One tool call in the bounded history.

Source: agent6.viewmodel.session.records.ToolCallView.

Field	Type	Meaning
name	str	The tool's name.
args_preview	str	The args rendered with each value truncated, for the inline table.
args_full	str	The args rendered with a generous per-value cap, for the detail view.
result_summary	str	The result's summary once it landed.
ok	bool \| None	The tool's verdict; None while in flight.
task_id	str \| None	The task in focus when the call ran, for filtering.
call_id	int \| None	The per-dispatch correlation id; None on a log without ids.

VerifyView

Support: public. The last verify gate run.

Source: agent6.viewmodel.session.records.VerifyView.

Field	Type	Meaning
command	tuple[str, ...]	The gate's argv.
exit_code	int \| None	The exit code, None when no exit was observed.
duration_s	float	How long it ran.
stdout_tail	str	The capped stdout tail.
stderr_tail	str	The capped stderr tail.
timeout_s	float	The gate's wall-clock limit, 0.0 when the event carried none.
outcome	str \| None	The finished attempt's outcome; None while in flight.
reason	str	Why the command did not run, separate from its output.

ReviewView

Support: public. Fold durable review evidence in journal append order.

Source: agent6.viewmodel.review.ReviewView.

Field	Type	Meaning
panels	tuple[PanelView, ...]	Recorded review panels in append order.
findings	tuple[FindingView, ...]	Findings with their observations and action history.
version	int	The concurrency version for finding actions.
open_counts	SeverityCounts	Counts of open grounded findings by severity.
completion	Literal['unknown', 'running', 'complete', 'inconclusive', 'skipped', 'reused']	The latest panel's completion state.
freshness	Literal['unknown', 'current', 'stale']	Whether the evidence matches the compared trees and paths.
legacy_notice	str	The notice for review data without structured evidence.
read_failure	str \| None	The evidence-read diagnostic, or null.
comparison	ReviewTarget \| None	The target used to determine freshness, or null.
event_ids	tuple[str, ...]	The durable event identifiers already folded.

EnvironmentView

Support: public. The run's container as the journal tells it: a start in progress, then the lease held.

Source: agent6.viewmodel.session.records.EnvironmentView.

Field	Type	Meaning
execution_id	str \| None	The execution that started or acquired it; the lease rows precede session.start, so the start keeps a view bound to its own execution.
starting	tuple[str, ...]	The start command running, () when none runs.
role	str	owner or borrower; "" before the run holds the container.
reason	str	Why the run holds that role.
started	bool	This run's start command made the container appear.
start_s	float \| None	How long the start took; None when none ran.
outcome	str	The lifecycle's end (left_running, stopped, unknown); "" while the run holds it.

ExecutionTargets

Support: public. Fold immutable target history and its current execution binding.

Source: agent6.evidence.ExecutionTargets.

Field	Type	Meaning
recorded	tuple[ExecutionTargetEvidence, ...]	Valid target stamps in observation order.
current	ExecutionTargetEvidence \| None	The prepared or started execution's target, None when unproved.
error	str	The latest target record's validation error, empty for a valid record.

BudgetView

Support: public. The run's usage as every surface shows it, each amount with one lifetime.

Source: agent6.viewmodel.session.records.BudgetView.

Field	Type	Meaning
session	SessionUsage	The usage across every execution.
execution	ExecutionUsage	The current accounting lifetime's usage beside its caps.
account	AccountReading \| None	The latest subscription account reading; None before a plan call runs.

ApprovalPrompt

Support: public. One approval the run asked for.

Source: agent6.viewmodel.session.records.ApprovalPrompt.

Field	Type	Meaning
id	str	The prompt's id.
prompt	str	The prompt's words.
standing	bool	An "allow all" is on offer for this prompt.
scope	str \| None	What a standing answer covers; None when none is on offer or the journal did not record it.
answered	bool	An answer was folded.
approved	bool \| None	The answer; None until answered.
asked_at	float \| None	When it was asked, for the waiting status's age.

QuestionPrompt

Support: public. An ask_user prompt: related questions the operator answers together.

Source: agent6.viewmodel.session.records.QuestionPrompt.

Field	Type	Meaning
id	str	The prompt's id.
questions	tuple[Question, ...]	The questions asked.
answered	bool	An answer was folded.
answers	tuple[str, ...]	One answer per question, aligned to questions.
from_harness	bool	agent6 itself asked (a start gate such as the dirty-tree question): it was asked before the session started or after it finished, when no model runs.
asked_at	float \| None	When it was asked, for the waiting status's age.
wait	bool	Whether the call waited for the answers, as asked.
blocking	bool	Whether the run is paused on it now: a waiting question in the execution that asked it; a queued question, or one an earlier execution asked, blocks nothing.
source	str	Who answered; late for a queued question's answers taken at a boundary.

DiffView

Support: public. One auto-commit diff plus the task in focus when it landed.

Source: agent6.viewmodel.session.records.DiffView.

Field	Type	Meaning
patch	str	The auto-commit's unified patch.
task_id	str \| None	The task in focus when the commit landed, or null.
sha	str	The auto-commit identifier.

RunCommit

Support: public. One per-turn commit of the run, which the dashboards select among.

Source: agent6.viewmodel.session.records.RunCommit.

Field	Type	Meaning
turn	int	The turn that produced the commit.
sha	str	The commit identifier.
subject	str	The commit's subject.

MachineStateView

Support: public. One state in the overview: its name, kind and position.

Source: agent6.viewmodel.machine.state.MachineStateView.

Field	Type	Meaning
name	str	The state's name.
kind	str	The state's kind from the spec.
is_current	bool	The machine is in this state, or is about to run it.
is_visited	bool	The machine entered or left this state.
mark	str	The mark before the name, as every surface draws it.

TransitionView

Support: public. One journaled transition: state --label--> goto.

Source: agent6.viewmodel.machine.state.TransitionView.

Field	Type	Meaning
seq	int	The transition's sequence number.
state	str	The state left.
label	str	The transition's label.
goto	str	The state entered.
detail	str	Bounded failure evidence (a failed tool's exit code and last output line, a failed agent state's stop reason); "" on success.
line	str	format_transition of the fields, as every surface prints it.

MachineResult

Support: public. How a drive ended.

Source: agent6.machine.contracts.MachineResult.

Field	Type	Meaning
status	Literal['ok', 'failed', 'incomplete', 'waiting', 'stopped']	ok or failed from a journaled end; incomplete for a replay that ends before a terminal; waiting for an --exit-on-wait park; stopped for a stop.
reason	str	The terminal's reason, the failure, or where the machine parked.
state	str	The state the machine is in.
transitions	int	The transitions taken.

NotificationView

Support: public. One journaled machine.notify (a state's notify message), in order.

Source: agent6.viewmodel.machine.state.NotificationView.

Field	Type	Meaning
ts	str	The journaled notification time.
state	str	The state that emitted the notification.
message	str	The notification text.
level	str	The notification's presentation level.

VerificationEvidence

Support: public. Record one verification attempt and the content it observed.

Source: agent6.evidence.VerificationEvidence.

Field	Type	Meaning
verify_id	str \| None	The verification attempt identifier, or null when unrecorded.
execution_id	str \| None	The execution that ran the attempt, or null when unrecorded.
command	tuple[str, ...]	The verification command's argv.
tree_before	str \| None	The observed tree before the attempt, or null when unknown.
tree_after	str \| None	The observed tree after the attempt, or null when unknown.
scoped	bool	Whether verification ran over selected paths instead of the full gate.
selected_paths	tuple[str, ...]	The paths selected for scoped verification.
outcome	Literal['passed', 'failed', 'timed_out', 'interrupted', 'denied', 'unavailable']	The attempt's observed outcome.
exit_code	int \| None	The process exit code, or null without an observed exit.
duration_s	float	The elapsed attempt time in seconds.
timeout_s	float	The configured timeout in seconds.
stdout_tail	str	The retained standard-output tail.
stderr_tail	str	The retained standard-error tail.
stdout_truncated	bool \| None	Whether standard output was truncated, or null when unrecorded.
stderr_truncated	bool \| None	Whether standard error was truncated, or null when unrecorded.
baseline	bool	Whether this was the pre-run baseline attempt.
reason	str	Why the command did not run, separate from its output.
target_id	str \| None	The execution-target stamp, or null when unrecorded.

RevertPlan

Support: public. Explain reversal from recorded git facts without executing it.

Source: agent6.viewmodel.receipt.RevertPlan.

Field	Type	Meaning
kind	Literal['not_recorded', 'squash', 'tasks', 'merge', 'ff_only', 'manual']	The reversal method supported by the recorded evidence.
commits	tuple[str, ...]	The commits to reverse in order.
mainline	int \| None	The merge parent to keep, or null when not needed.
explanation	str	The evidence supporting the plan or requiring manual inspection.

ExecutionTargetEvidence

Support: public. Bind observed target facts to an execution, without inventing launcher identity.

Source: agent6.evidence.ExecutionTargetEvidence.

Field	Type	Meaning
target_id	str	A host-issued evidence reference, not a launcher instance identifier.
execution_id	str \| None	The execution that selected the target, or None when unrecorded.
target	ExecutionTargetStamp	Persistence-safe target facts; unknown observations remain None.

SetupEvidence

Support: public. Retain setup identity and output independently of later baseline attempts.

Source: agent6.evidence.SetupEvidence.

Field	Type	Meaning
execution_id	str \| None	The execution that ran setup, or None when unrecorded.
target_id	str \| None	The setup target reference, or None when unrecorded.
command	tuple[str, ...]	The setup command's argv.
exit_code	int \| None	The observed exit code, or None when unknown.
duration_s	float	Elapsed setup time in seconds.
exec_failed	bool	Whether the command could not be executed.
stdout_tail	str	Retained standard output.
stderr_tail	str	Retained standard error.

JobsObservation

Support: public. Record whether the command owner confirmed job settlement, not environment shutdown.

Source: agent6.evidence.JobsObservation.

Field	Type	Meaning
execution_id	str \| None	The execution whose jobs were observed, or None when unrecorded.
target_id	str \| None	The observed target reference, or None when unrecorded.
outcome	Literal['settled', 'unknown']	Whether the command owner confirmed settlement.
reason	str	Why settlement is unknown; empty when confirmed.

EnvironmentObservation

Support: public. Record an environment lifecycle observation without granting stop authority.

Source: agent6.evidence.EnvironmentObservation.

Field	Type	Meaning
execution_id	str \| None	The execution observing the lifecycle, or None when unrecorded.
target_id	str \| None	The observed target reference, or None when unrecorded.
outcome	Literal['left_running', 'stopped', 'unknown', 'not_applicable']	The observed lifecycle state; not_applicable means the host is not managed.
reason	str	The observation's explanation, not an authorization to stop anything.

AcquisitionObservation

Support: public. Record the container lease an execution acquired, as the lifecycle said it.

Source: agent6.evidence.AcquisitionObservation.

Field	Type	Meaning
execution_id	str \| None	The acquiring execution, or None when unrecorded.
lifecycle_ref	str	The lease key, the target's lifecycle_ref.
role	str	owner when this run's start made the container appear, else borrower.
reason	str	Why the run holds that role.
started	bool	This run's start command made the container appear.
start_s	float \| None	How long the start took in seconds; None when none ran.

FindingResolution

Support: public. Record the operator's conclusion without inferring verification success.

Source: agent6.review_records.FindingResolution.

Field	Type	Meaning
disposition	Literal['fixed', 'false_positive', 'intentional', 'wont_fix', 'duplicate', 'pre_existing']	The operator's conclusion about the finding.
reason	str	The operator's explanation.
duplicate_of	str \| None	The other finding for a duplicate resolution, or null.
commit	str \| None	The associated correction commit, or null.
verify_id	str \| None	The associated verification attempt, or null.

ReviewTarget

Support: public. Identify the immutable trees and path coverage of a review.

Source: agent6.review_records.ReviewTarget.

Field	Type	Meaning
base_commit	str \| None	The base commit, or null for a tree-only comparison.
base_tree	str	The immutable base tree identifier.
head_commit	str \| None	The head commit, or null for a working-tree snapshot.
head_tree	str	The immutable head tree identifier.
snapshot_commit	str	The commit retaining the reviewed tree.
paths	tuple[str, ...]	The reviewed path scope, empty for all paths.
source	Literal['revisions', 'worktree']	Whether the comparison came from revisions or the working tree.
hidden_paths	tuple[str, ...]	Paths withheld from the review.
snapshot_ref	str	The session-owned hidden ref retaining the comparison; empty on old records.

JoinRecord

Support: public. How one child's work joined, with the chain tip on either side of the merge.

Source: agent6.sessions.child_dispatch.JoinRecord.

Field	Type	Meaning
branch	str	The child's branch.
status	Literal['joined', 'no_changes', 'conflict', 'failed', 'ranked']	The join's outcome.
sha	str	The merge commit, "" when none.
detail	str	The failure's reason, or what stopped a child whose work still joined; "" when neither.
target_before	str	The chain tip before the join.
target_after	str	The chain tip after it; equal to target_before when nothing merged.

UsageRecord

Support: public. What a child used, as settled against the parent.

Source: agent6.sessions.child_dispatch.UsageRecord.

Field	Type	Meaning
usd	float	The child's metered spend.
usd_partial	bool	Whether unpriced calls make it a lower bound.
tokens	int	Its fallback-ledger tokens.

ModelBrief

Support: public. The route and effective effort of a resolved role.

Source: agent6.sessions.manifest.ModelBrief.

Field	Type	Meaning
provider	str	The configured provider name.
model	str	The provider's model identifier.
effort	str \| None	The effective effort; None sends no value.
effort_source	str	The origin, including inheritance; empty when not recorded.

GateStamp

Support: public. The pinned gate and its availability for this execution.

Source: agent6.sessions.manifest.GateStamp.

Field	Type	Meaning
command	tuple[str, ...]	The pinned argv; only operator config can replace it on resume.
source	Literal['', 'configured', 'inferred', 'adopted']	configured from operator config, inferred from repo signals, adopted mid-run, or "" for no gate.
availability	Literal['ready', 'pending', 'withheld']	ready after resolution, pending before resolution, or withheld when this execution cannot run the command.

ExecutionEnvironmentPolicy

Support: public. Persist environment policy and variable names, never their potentially secret values.

Source: agent6.kinds.ExecutionEnvironmentPolicy.

Field	Type	Meaning
inherit	bool	Whether commands inherit the target environment before applying variables.
variable_names	tuple[str, ...]	Names explicitly supplied to commands, sorted; values are omitted.

EffectiveControl

Support: public. One configured sandbox leaf and the enforcement actually selected for it.

Source: agent6.kinds.EffectiveControl.

Field	Type	Meaning
leaf	str	The sandbox configuration field.
configured	str \| int \| tuple[str, ...]	Its configured value before target resolution.
effective	str \| int \| tuple[str, ...]	Its resolved value for this target.
behaviour	str	The control's effective behaviour.
owner	Literal['agent6', 'container', 'host file tools', 'unsupported']	What enforces the control, or unsupported when it is not enforced.
probe	Literal['not probed', 'resolved', 'prepared']	How far target preparation confirmed the control.
reason	str	The refusal or warning reason, if any.

Lifecycle

Support: public. Where a session stands, as every fold reads it.

Source: agent6.viewmodel.session.records.Lifecycle.

Field	Type	Meaning
started	bool	An execution has begun; a parked or created run has none.
finished	bool	A session.stop was folded and no later execution began.
mode	str	The session's mode, "" until a start names it.
execution_id	str	The current execution's id, "" until its start is read.
executions	int	One plus the resumed executions.
task_label	TaskLabel	The first task and the latest resume steer.
stop_reason	str	The session.stop reason word, "" while an execution runs.
verification	Literal['passed', 'failed', 'unverified', 'not_applicable'] \| None	The final verification word, or None when unknown.
verify_scoped	bool	The judging gate ran scoped.
stranded	bool	This execution finished with its edits uncommitted and no commit landed.
end_summary	str	The end sentence the journal carries, "" when it carries none.
finish_summary	str	The finish tool's summary, the agent's closing statement.
undone_to	str	The session /undo forked, "" when it was never undone.
provider_error	str	The last failed driving call's error, "" when none failed.
last_verify_rc	int \| None	This execution's last verify.end exit code; None before one ends.
verify_timeout_s	float	The limit that verify ran out, 0.0 when it ended on its own.
unattended_questions	int	This execution's questions answered empty because nobody was attached.
pins	tuple[str, ...]	The operator's pinned instructions in force, most recent last.
environment	EnvironmentView \| None	The container this execution holds; None outside one.

PanelView

Support: public. Keep a panel's roster, grounded verdicts and completion evidence.

Source: agent6.viewmodel.review.PanelView.

Field	Type	Meaning
panel_id	str	The panel identifier.
execution_id	str	The execution that requested the panel.
target	ReviewTarget \| None	The recorded comparison target, or null.
roster	tuple[ReviewerAssignment, ...]	The resolved reviewer assignments.
reviewers	tuple[ReviewVerdict, ...]	The recorded verdicts from individual reviewers.
result	ReviewPanel \| None	The completed panel result, or null while unavailable.
completion	Literal['unknown', 'running', 'complete', 'inconclusive', 'skipped', 'reused']	The panel's completion state.
reason	str \| None	The explanation for a skipped or incomplete panel, or null.
rule_sources	tuple[RuleSource, ...]	The base-tree rule sources supplied to the panel.
rules_complete	bool \| None	Whether all rule sources were supplied, or null when unrecorded.
rules_reason	str \| None	The explanation for missing rules, or null.

FindingView

Support: public. Keep a finding's observations and operator history independently of panel verdicts.

Source: agent6.viewmodel.review.FindingView.

Field	Type	Meaning
finding_id	str	The durable finding identifier.
observations	tuple[FindingObservation, ...]	The finding's observations in journal order.
status	Literal['open', 'resolved']	Whether the finding is open or resolved.
resolution	FindingResolution \| None	The operator's conclusion, or null while open.
actions	tuple[ActionRecord, ...]	The finding's recorded operator actions.
anchor_state	Literal['unknown', 'current', 'outdated']	Whether the location matches the displayed comparison.
fix_state	Literal['none', 'delivery_unknown', 'submitted', 'failed']	The latest fix delivery state.
opened_at	float \| None	The first observation time in epoch seconds, or null when unrecorded.

SeverityCounts

Support: public. Count open findings by their grounded severity.

Source: agent6.viewmodel.review.SeverityCounts.

Field	Type	Meaning
block	int	The number of open grounded block findings.
warn	int	The number of open grounded warning findings.
nit	int	The number of open grounded nit findings.

SessionUsage

Support: public. The session's usage across every execution, what the listings' usage cell shows.

Source: agent6.viewmodel.session.records.SessionUsage.

Field	Type	Meaning
usd	float	The spend across every execution.
usd_partial	bool	Some execution's model had no price, so usd is a lower bound.
plan_observed_points	float	The plan movement observed across every execution, in points.
unmetered_tokens	int	Fallback tokens across every execution, including settled descendants.
settled_children	tuple[str, ...]	Child sessions whose spend is durably included in the journal.

ExecutionUsage

Support: public. The current accounting lifetime's usage beside the caps that bound it.

Source: agent6.viewmodel.session.records.ExecutionUsage.

Field	Type	Meaning
input_tokens	int \| None	Input tokens; None until a budget row carries the counters.
output_tokens	int \| None	Output tokens; None until a budget row carries the counters.
cache_read_tokens	int \| None	The cached side of the input; None when the row lacked it.
cache_write_tokens	int \| None	Cache-write tokens; None when the row lacked it.
usd	float	The metered spend.
usd_partial	bool	Some model had no price, so usd is a lower bound.
unmetered_tokens	int	Input and output tokens of calls the meter could not price.
usd_running_children	float	What the children still running have spent so far: inside the cap, outside usd until each settles.
tokens_running_children	int	Their unpriced tokens so far, outside unmetered_tokens.
plan_observed_points	float	The plan movement observed, in points.
usd_cap	float \| None	[budget].max_usd.
tokens_fallback_cap	int \| None	[budget].max_tokens_fallback.
plan_cap	float \| None	[budget].max_plan_points.

AccountReading

Support: public. The latest reading of a subscription account: an observation of the account, not the run.

Source: agent6.viewmodel.session.records.AccountReading.

Field	Type	Meaning
provider	str	The provider entry whose account was read; "" in a journal written before the reading named it.
used_percent	float	The binding window's used percent.
resets_at	float	When the binding window resets, as epoch seconds; 0 when unknown.
windows	tuple[PlanWindow, ...]	Every window the backend reported, the primary first.
credits_has	bool	Whether the account holds purchased credits.
credits_unlimited	bool	Whether the credits are unlimited.
credits_balance	str	The credit balance as the backend sent it; "" when none.
observed_at	float	When the reading was journaled, as epoch seconds.

Question

Support: public. One question within an ask_user prompt.

Source: agent6.viewmodel.session.records.Question.

Field	Type	Meaning
question	str	The question's text.
options	tuple[str, ...]	Selectable presets; the operator may also type a free-text answer.

ReviewerAssignment

Support: public. Identify the resolved route and settings of one reviewer.

Source: agent6.review_records.ReviewerAssignment.

Field	Type	Meaning
reviewer_index	int	The reviewer's zero-based position in the roster.
model_route	str	The resolved provider/model route.
focus	str	The reviewer's assigned review focus.
temperature	float \| None	The configured sampling temperature, or null for the provider default.

ReviewVerdict

Support: public. Hold one indexed reviewer's verdict, including an explicit abstention on failure.

Source: agent6.review_records.ReviewVerdict.

Field	Type	Meaning
reviewer_index	int	The reviewer's zero-based position in the roster.
focus	str	The reviewer's assigned review focus.
model_route	str	The provider/model route that reviewed.
verdict	Literal['pass', 'block', 'abstain']	The reviewer's pass, block or abstention decision.
findings	tuple[Finding, ...]	The grounded findings from this reviewer.
summary	str	The reviewer's closing summary.
error	str \| None	The failure diagnostic for an abstention, or null.
session_id	str	The reviewer's session, empty in legacy verdicts.

ReviewPanel

Support: public. Validate a completed panel's durable evidence, including its worker feedback.

Source: agent6.review_records.ReviewPanel.

Field	Type	Meaning
panel_id	str	The durable panel identifier.
execution_id	str	The execution that requested the panel.
turn	int	The turn that triggered the panel.
trigger	str	The review trigger.
target	ReviewTarget	The immutable reviewed comparison.
roster	tuple[ReviewerAssignment, ...]	The reviewers that sat; empty on a record written before rosters were kept.
findings	tuple[FindingObservation, ...]	The panel's grounded finding observations.
injected_finding_ids	tuple[str, ...]	Findings supplied to the worker as feedback.
block_on	Literal['none', 'any', 'all', 'model_quorum']	The configured worker-finish blocking rule.
raw_blocked	bool	Whether the panel met the blocking rule before disarming.
blocked	bool	Whether the panel blocks worker completion.
disarmed	bool	Whether the harness disarmed the review block.
n_block	int	The number of distinct blocking provider/model routes.
n_abstain	int	The number of abstaining reviewers.
completion	Literal['complete', 'inconclusive']	Whether the panel completed conclusively.

RuleSource

Support: public. Identify one base-tree rule blob supplied whole to the reviewer.

Source: agent6.review_records.RuleSource.

Field	Type	Meaning
path	str	The rule file's path in the base tree.
blob_id	str	The immutable rule blob identifier.
supplied_bytes	int	The number of rule bytes supplied to reviewers.

FindingObservation

Support: public. Associate a finding with its panel and the reviewers that reported it.

Source: agent6.review_records.FindingObservation.

Field	Type	Meaning
finding_id	str	The finding identifier shared across observations.
panel_id	str	The panel that recorded this observation.
finding	Finding	The grounded finding text and location.
reviewer_indexes	tuple[int, ...]	The zero-based reviewers that reported the finding.
line_origin	LineOrigin	The line-attribution evidence for this observation.

ActionRecord

Support: public. Validate the journal fields retained in a finding's action history.

Source: agent6.viewmodel.review.ActionRecord.

Field	Type	Meaning
type	str	The journaled action event name.
finding_id	str	The target finding identifier.
request_id	str	The caller's idempotency identifier.
source	Literal['cli', 'tui', 'web', 'acp'] \| None	The front-end that requested the action, or null when unrecorded.
version	int \| None	The resulting review version, or null when unrecorded.
resolution	FindingResolution \| None	The recorded conclusion, or null.
reason	str \| None	The operator's explanation, or null.
target	ReviewTarget \| None	The reviewed target associated with the action, or null.
action	str \| None	The requested action kind, or null when unrecorded.
execution_id	str \| None	The execution that handled the action, or null.
error	str \| None	The action failure diagnostic, or null.

PlanWindow

Support: public. One rate-limit window of a subscription plan.

Source: agent6.budget.PlanWindow.

Field	Type	Meaning
name	str	The backend's label: primary, secondary, a per-model family.
used_percent	float	Percent of the included allowance used.
window_minutes	int	The window's length; 0 when the backend gave none.
resets_at	float	When the window resets, as a Unix time.

Finding

Support: public. Hold one review finding at a structured location.

Source: agent6.review_records.Finding.

Field	Type	Meaning
category	str	The finding's issue category.
severity	Literal['block', 'warn', 'nit']	The grounded block, warn or nit severity.
title	str	The short finding title.
detail	str	The supporting explanation and evidence.
location	ReviewLocation	The path, side and optional line range.

LineOrigin

Support: public. Record byte-level evidence without dating a defect or changing its severity.

Source: agent6.review_records.LineOrigin.

Field	Type	Meaning
kind	Literal['pre_existing', 'changed', 'unknown']	Whether the lines are pre-existing, changed or of unknown origin.
reason	str	The explanation for the attribution result.
base_location	ReviewLocation \| None	The mapped base range for pre-existing lines, or null.
commits	tuple[str, ...]	Attribution commits for pre-existing lines.

ReviewLocation

Support: public. Name a repository-relative path on one side, with an optional inclusive line range.

Source: agent6.review_records.ReviewLocation.

Field	Type	Meaning
path	str	The repository-relative path.
side	Literal['base', 'head']	The base or head side of the comparison.
start_line	int \| None	The inclusive one-based start line, or null for a file-level finding.
end_line	int \| None	The inclusive one-based end line, or null for a file-level finding.