Skip to content

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

gaia-steward-full

Verified

Run Gaia Steward end to end: observe every routine's detectors, triage debt by authority, land Class A repairs on the integration branch, drive a bounded batch of Class B dispatches through independent verifiers, re-observe, and leave a quiet founder desk or a precise list of decisions. Use for "/gaia-steward-full", "run the steward", "full maintenance pass", "what does the repo need", "resume the steward run", or a scheduled maintenance wake. Thin entrypoint: each phase is an atomic gaia-steward-* skill.

FileKindSize
SKILL.mdskill md9.9 KB
cloud-custody.mdmarkdown3.6 KB

cloud-custody.md

View on GitHub
---
name: cloud-custody
description: Shared contract for gaia-steward-full, -lane and -verify on ephemeral cloud agents (issue #2448).
---

# Ephemeral cloud-subagent evidence (daily runs)

The daily cloud job is an external scheduled task whose agents are destroyed when
they finish; it is not the GitHub Actions Class A pulse (`.github/workflows/steward.yml`).
**`steward/ledger` carries full-run state only. Checkout-local `.gaia/steward/`
receipts, diffs and proofs die with the container**, so cloud Class B work runs with
evidence custody on: `--evidence git` (or `GAIA_STEWARD_EVIDENCE=git`) on `gaia steward
full run|record`, and on standalone `dispatch`, `lane next` and `verify`. Custody
requires the shared git ledger: `--ledger git`; `auto` never falls back and `local` is
refused (BLOCKED, exit 4, even on a quiet day). Custody
stores the original scan and dispatch receipts byte-for-byte (and the exact builder
prompt), the candidate diff, the validated `steward-proof-transcript-v1` and the
verification receipt on the append-only `steward/evidence` ref (fast-forward CAS, no
force, read back before success). Contract:
`founder/handovers/2026-11-13-issue-2448-cloud-evidence-custody.md`.

| Boundary | Command (Lead unless noted) | If it fails |
|---|---|---|
| Dispatch issued | automatic with `--evidence git`; or `gaia steward evidence publish-dispatch DEBT` | no prompt, no builder, run `blocked` |
| Builder accepts | builder runs `gaia steward evidence accept DEBT` first | builder stops |
| Patch + proof | `gaia steward evidence publish-build DEBT --diff D --proof P --summary S --model M --effort E` (automatic in `full record`) | outcome refused; step stays pending, record again after repair |
| Lost prompt / adoption | automatic: the ORIGINAL prompt is restored from custody | pending work shows `BLOCKED`; resume after repair; nothing is re-rendered |
| Verifier retrieves | different checkout: `gaia steward evidence verify DEBT` (automatic in `full run --verify`) | work deferred, never re-dispatched |
| Outcome recorded | automatic: the verdict receipt is published, bound to that exact build | deferred |
| Before READY / PR | `gaia steward evidence ready DEBT --pin-bundle ID --pin-commit SHA`; paste its ids and hashes in the PR | not READY |

A verdict is bound to the build it judged: its debt, dispatch, diff hash and proof
hash must match that build's bytes, at publication, at every retrieval and in `ready`.
`--pin-bundle` / `--pin-commit` check the chain against identifiers recorded somewhere
the ref's writers do not control (the ledger work item, or the Lead's PR handoff);
`--pin-commit` also detects rewritten history. Writer, model, effort and summary are
part of each bundle's identity. `ready` never claims the ref is protected: that is
repository configuration, verified only by `scripts/check_steward_evidence_protection.py`.

Non-negotiable: never re-dispatch, re-render or write a receipt to recreate lost
authority; a missing or altered original means `BLOCKED: provenance unavailable`
(exit 4), reported verbatim, with no silent local fallback and no founder item
disguised as success. A mechanical verdict stays `pending`: the independent
verifier still decides, and a recorded founder disposition of one candidate (e.g.
#2445) is not a pass of `gaia steward verify` and authorizes no future bypass.
Without `--evidence git`, `full run --verify` refuses to re-dispatch when the
original receipt is missing.

Recover by repairing the remote or push permission, then re-running the same
command: publishing is idempotent and resumed runs never duplicate a dispatch.
Stop by abandoning the run (`gaia steward full abandon RUN --reason WHY`); nothing is deleted.
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: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated 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: 323kUpdated Claude CodeCodexVerifiedInstall