Proof Reporttrust artifact
evidence-first autonomy

Evidence before autonomy scales.

A Proof Report turns autonomous execution into reviewable operational evidence before PR, merge, deployment, escalation, or rollback.

Proof Report/v1/proof-packets/sample
Trace

WorkGraph event cursor, runtime receipts, tool calls, actor identities, and capability refs.

6/6 required lanes present
Checks

typecheck, lint, build, diff hygiene, and capability-contract assertions.

validation gate required
Artifacts

patches, logs, screenshots, generated files, capability readbacks, and proof report refs.

artifact rail visible
Risk

branch race, path ownership, external side effects, and deployment blast radius.

human review required
Rollback

baseline ref, revert command, supersession marker, replay cursor, recovery owner.

rollback path pinned
Approval Queue/v1/governance/decisions/sample
sensitive action

Proof Report must be accepted before external or sensitive action

human_reviewerapproval_required
External side effect

No external send, production mutation, or customer-visible change from the console

owner onlyblocked
Runtime escalation

Specialized execution lanes require explicit policy and spend scope

governance_corereview
Deployment

Release handoff needs health checks and rollback review

release_handoffhold
Destructive shell

Denied until explicit command, target, and rollback are present

toolmeshblocked
Rollback Inspector/v1/events/feed
origin/mainbaseline

last fetched remote before autonomous work

anchor
workgraph://verified-prstate checkpoint

objective, leases, runtime choices, policy decisions

ready
review://report/sampleevidence bundle

checks, logs, trace, artifacts, reviewer checklist

compiling
artifact://patchreversible diff

supersede or revert before sensitive action

review
event://replayreplay cursor

resume from last verified event if runtime fails

waiting
Tool Permissions View/v1/tool-registry/capabilities
GitHubchecks/read by default; PR/write gated

WorkGraph lease + Proof Report

approval_required
Shellsandboxed commands only; destructive denied

Execution receipt + side-effect summary

leased
Browserobserved navigation, screenshots, console logs

Execution Chamber artifact

leased
Protected resourcesSensitive systems exposed by contract

ToolMesh capability + denial receipt

review
SkillsExecution Host/OpenAI/agent skills wrapped as governed actions

policy scope + review ref

ready
EvaluationsQuality reviews feed the release decision

quality gate + risk note

approval_required
review gate

Objective

Summary, repo, branch, actor, risk class, and acceptance criteria are frozen before execution.

review gate

Receipts

Every runtime/tool action records input, output, side effects, artifacts, and policy decision.

review gate

Verification

Tests, evals, screenshots, build logs, review notes, and capability readbacks are linked.

review gate

Approval

Human can approve, reject, narrow, retry, stop, or rollback with the review in view.

review gate

Lineage

Commit, artifact, runtime version, source repo, and rollback target remain traceable.

field

objective

Required for governed handoff and auditability.

field

plan

Required for governed handoff and auditability.

field

execution trace

Required for governed handoff and auditability.

field

artifacts

Required for governed handoff and auditability.

field

tests/evals

Required for governed handoff and auditability.

field

review notes

Required for governed handoff and auditability.

field

risks

Required for governed handoff and auditability.

field

approvals

Required for governed handoff and auditability.

field

rollback

Required for governed handoff and auditability.

field

lineage

Required for governed handoff and auditability.

field

handoff

Required for governed handoff and auditability.