agentleFS
metratonmetraton/gaia/skills/execution/SKILL.md

execution

Skill4 starsChanged 4 months agoMITView on GitHub

Use when the specialist that requested a T3 operation or COMMAND_SET is resumed after the user granted it

10 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use review.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • SentrySkill

    review

    Fixture review skill.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read the approval with gaia approvals show <approval_id>: its state, its commands in order, their directory, and how much of the grant's …

    line 20
  2. Very specific:

    Run exactly one command per tool call using command-execution, byte for byte as requested, in its sealed directory.

    line 23
  3. Distinctive:

    After every result, checkpoint the exact command, index and exit status.

    line 33
  4. Distinctive:

    After successful mutations, verify desired state with separate read-only checks.

    line 53

“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Approved Execution

The approved work runs in the specialist that requested it, resumed by the orchestrator; the signature belongs to that agent and session. Approval does not turn the orchestrator into an executor and does not broaden scope.

Source boundary

For Gaia components, edit only the canonical gaia/ source tree. Never write, copy, generate, or stage anything under .claude/; installation propagates source changes. The same prohibition applies to fixtures and bulk operations.

Ordered execution

  1. Read the approval with gaia approvals show <approval_id>: its state, its commands in order, their directory, and how much of the grant's window is left.
  2. Run exactly one command per tool call using command-execution, byte for byte as requested, in its sealed directory. When that directory is not the one your Bash runs in, run exactly cd <sealed directory> && <sealed command>, the only compound accepted. Never join commands otherwise, skip an index, substitute an equivalent spelling, or add an unapproved command. A still-pending approval is not a reason to stop: attempt the command. If Gaia refuses it because the user has not answered yet, end the turn with APPROVAL_REQUEST and the same approval id; the orchestrator asks the user and resumes you after the answer. This is the same in Claude Code and OpenCode, and nothing is asked in your session.
  3. After every result, checkpoint the exact command, index and exit status.
  4. A command that exits with a code its request declared with --expect-exit has succeeded for the set: continue with the next index. Only a declaration made when the set was requested counts; you cannot declare one now. Any other failure stops the set; apply command-execution's COMMAND_SET fail-fast rule, then use agent-protocol to reconcile the result for its consumer. An unsigned step of the set (gaia approvals show marks it "signed": false) runs in its position like any other command of yours: it holds no index, so it neither spends nor advances the signature.
  5. A call the host refused before running it (Claude Code's Permission to use Bash ... has been denied) spends nothing: the item keeps its index and the single-command signature stays usable. Retry the same sealed bytes once, as a new call; if the host refuses again, stop and report the refusal with the approval id instead of requesting anew. A call that never reports back is recorded as no result, not as a failure, and does not advance the set. Read its state before deciding anything, and request anything still needed as a new request. A single-command signature whose command exited non-zero is given back once, while its window lasts: retry the same sealed bytes from the same directory, and after a second failure investigate and request anew.
  6. After successful mutations, verify desired state with separate read-only checks. Success exit codes alone are insufficient.
  7. Checkpoint verification and emit NEEDS_VERIFICATION for a plan-task-bound producer; only an eligible unbound turn/verifier may reach COMPLETE.

This skill sequences the approved work; it does not duplicate the failure rule or the consumer reconciliation owned by those two skills.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.