Skip to content

gaia-research/gaia-skill-tree / .agents/skills/gaia-steward-verify

gaia-steward-verify

Verified

Independently verify one Gaia Steward Class B build: read only the rendered verifier prompt and the sensor evidence, check the diff and proof, and record accept, reject or escalate. Use when a full Steward run emits an agent.verify action, for "/gaia-steward-full --verify", or "verify the steward dispatch". The verifier must be a different agent from the builder.

gaia-steward-verify

Responsibility

Decide whether one candidate change resolves its finding inside its envelope. You are the second reader; the machine has already rejected what it can.

Input

A pending agent.verify action. Read params.verifierPromptPath and nothing from the builder beyond the diff and proof it references. If your agent id is in params.forbiddenAgentIds, stop: you built this work.

Run the mechanical pre-check first:

gaia steward verify DEBT_ID --diff candidate.diff --proof proof.json

Exit 1 (reject) and 3 (escalate) are final for this attempt. Exit 0 means only that the machine did not object.

On an ephemeral cloud host the builder's files are gone. With custody on (GAIA_STEWARD_EVIDENCE=git) run gaia steward evidence verify DEBT_ID instead: it retrieves the ORIGINAL receipts, diff and proof from steward/evidence, authenticates them, runs the same mechanical verifier and publishes the verdict. If it prints BLOCKED: provenance unavailable, stop and report that. Never dispatch again to make a verification possible.

Output

gaia steward full record RUN STEP --input verdict.json
{"verdict": "accept", "rationale": "...", "agentId": "...", "model": "sonnet", "effort": "high"}

verdict is accept, reject or escalate. model must be one the action allows and effort must be explicit.

For an issue-suite-review proposal (routine 18), the question is whether every disposition in the proposal is reproduced from the census and the local registry. Re-run the commands the proposal names. Reject any proposal that closes an issue, comments on one, or accepts a candidate: those are not Class B outcomes.

For a routine pull-request candidate (routine 19) the question is whether the exact head is eligible: re-run every gate with python scripts/steward_pr.py observe, read the diff against the producer's allowed paths, and re-derive the content proofs. Answer PASS or FAIL with proofs; the verifier is never the builder, and a PASS binds to one head and base SHA only.

Authority ceiling

Read and judge only. Do not edit the diff, rerun the build or widen the envelope. An accept moves the work to verified; integration is a separate lead.integrate step.

Stop conditions

  • The diff touches a path outside the envelope: reject.
  • The proof does not cover every contract item: reject.
  • The change needs a decision only the founder can make: escalate.

Idempotency

record accepts the same verdict twice and refuses a different one for the same step.

Failure states

record refuses a verifier that built the work, a non-allowed model, a missing effort or an empty rationale.

Validation

python -m pytest tests/steward/test_steward_verification.py tests/steward/test_fullrun.py -q

Tests

tests/steward/test_steward_verification.py, tests/steward/test_fullrun_scenarios.py, tests/steward/test_steward_skills.py.

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall