agentleFS
grovegrove/promise-to-proof/skills/productivity/repair-gaps/SKILL.md

repair-gaps

Skill4 starsChanged 36 days agoApache-2.0View on GitHub

Repair named implementation or evidence gaps from a NOT PROVEN proof, report the change honestly, and require fresh proof without declaring acceptance.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 7 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When the controller supplies a stall perspective reset, reopen the original promise and accepted requirements rather than anchoring on the …

    line 17
  2. Distinctive:

    Use only the bounded recent strategies first, reading existing deeper receipts when the difference matters.

    line 19
  3. Distinctive:

    Use its supplied exact input identity.

    line 30
  4. Distinctive:

    Before repair, read the acceptance contract protocol.

    line 43
  5. Distinctive:

    Never weaken the contract to fit a repair.

    line 56
  6. Very specific:

    Do not publish, commit, push, or declare PROVEN.

    line 59
  7. Very specific:

    If inputs are stale, ambiguous, or unavailable, report BLOCKED without editing.

    line 60
  8. Distinctive:

    Run the focused check and inspect the resulting diff.

    line 73
  9. Distinctive:

    If this run reveals durable, non-obvious knowledge that is likely to help future delivery work, include it as a learning candidate.

    line 102
  10. Very specific:

    When the enclosing controller requests structured JSON, put these in its learning_candidates array.

    line 107
  11. Very specific:

    After the handoff, end with Next steps: and a numbered list (1., 2., ...) of applicable actions in order, so each can be referenced by …

    line 135

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

Delegated continuation

In an enclosing autonomous delivery, apply a covering in-scope recovery recommendation and continue to both independent verifiers without asking again. For a read-only diagnosis dispatch, inspect mechanisms and return the controller's requested diagnosis schema; make no edits and do not claim REPAIRED. Choose a concrete different method after repeated gaps, including evidence or local prerequisite work. Name an exact unavailable input only when recovery cannot act. When the controller supplies a stall perspective reset, reopen the original promise and accepted requirements rather than anchoring on the previous fix. Use only the bounded recent strategies first, reading existing deeper receipts when the difference matters. Give the executable method a stable strategy_key, explain difference_from_prior, and name in expected_result the concrete falsifiable check and observed behavior that will change. Do not rename the old repair and claim novelty, or treat a single transient error as a structural sizing issue. A real unavailable prerequisite or authority stays a blocker. Evidence-only recovery must leave product bytes unchanged. Planning decisions return to the delegated planner/auditor; repair cannot weaken the agreement.

A controller's read-only diagnosis can follow implementation, review or proof; it does not require a NOT PROVEN proof for an implementation or agreement gap. Use its supplied exact input identity. The controller owns authority validation, candidate capture, Git generation mapping and durable records; do not reconstruct those checks or audit its instruction archive. Read every applicable protocol section using its headings, skipping unrelated workflow sections. The standalone repair workflow below requires the matching proof and storage steps.

/repair-gaps is the repair step after /prove. It consumes the source contract, the proof result, the matching candidate, and specific unresolved requirement IDs. It repairs only those gaps and never replaces fresh proof. Review-only findings belong to an authorized /implement-contract invocation; they do not replace the matching NOT PROVEN proof this skill requires.

Before repair, read the acceptance contract protocol. Make the smallest complete repair for the named requirements: narrow in scope, complete in depth.

Guardrails

  • Confirm that the proof's exact contract revision and captured text match the current contract, and that the exact candidate identity still matches.
  • Edit only the implementation or evidence needed for the named requirements.
  • Preserve the source requirements and existing valid checks.
  • Implement all state, invariant, persistence, and failure behavior necessary for the named outcomes. Those IDs do not authorize unrelated improvements or speculative machinery. Reuse the agreed seams and independent oracles.
  • Never weaken the contract to fit a repair. If a changed promise or consequential new seam is needed, report the decision as blocked and hand it back for explicit resolution. An authorized material change requires a new revision and proof.
  • Do not publish, commit, push, or declare PROVEN.
  • If inputs are stale, ambiguous, or unavailable, report BLOCKED without editing.

Workflow

  1. Accept .p2p/work/<slug>/contract.md and discover its linked inputs, candidate.json, and proof.md under .p2p/work/<slug>/; also accept an explicit proof reference that resolves to the same work item. Read the exact local contract revision and matching NOT PROVEN proof result. Recheck binding parent/spec hashes and comparison base; local .p2p/ record updates preserve candidate identity and must stay out of Git history.
  2. Confirm the candidate identity and the unresolved requirement IDs.
  3. Apply the smallest complete implementation or evidence repair for the named requirements, when editing is authorized. Preserve authorization already given.
  4. Run the focused check and inspect the resulting diff.
  5. Recheck the contract and report the candidate before and after, changed files, addressed requirements, changed evidence, and remaining gaps. When gaps remain, recommend one smallest concrete next action. Name the affected IDs and the exact command, missing input, or human decision needed; None is valid only when no gaps remain. The candidate may change only in the scoped repair; the acceptance requirements stay intact.
  6. Save and reread .p2p/work/<slug>/repair.md, retained evidence, and the updated compact candidate.json identity. Keep payload required to resume an active run in ignored local storage, not in the durable candidate record. Preserve previous reports and candidate records under the protocol's history rule before replacement. Exclude .p2p/ from the candidate. Changed product or binding inputs make old review/proof historical; neither accepts the repaired candidate. Hand off the changed candidate for separate /review-implementation and fresh proof. End with: Fresh /prove required before acceptance.

Outcomes

  • REPAIRED: the scoped change was made and focused checks passed.
  • NO CHANGE: no safe or necessary change was made; explain why.
  • BLOCKED: the inputs or required capability are unavailable or stale.

None of these outcomes accepts the ticket. A new /prove run must evaluate the changed candidate against every requirement.

Preserve reusable delivery learning

If this run reveals durable, non-obvious knowledge that is likely to help future delivery work, include it as a learning candidate. Empty output is valid and preferred over inventing a lesson. A candidate must state its scope, the reusable lesson, the concrete evidence that suggested it, and important uncertainty.

When the enclosing controller requests structured JSON, put these in its learning_candidates array. Otherwise add a short ## Learning candidates section to the report.

Learning candidates are leads for /retrospect, not accepted advice. They must not change the contract, requirement IDs, stage verdict, scope, repair allowance, or authority. Do not promote them into project rules or the advisory learning register from this skill.

Handoff format

# <REPAIRED | NO CHANGE | BLOCKED>: <source>

Contract: <source and exact revision>
Contract snapshot: <same immutable reference or captured text and digest as proof>
Candidate before: <identity>
Candidate after: <identity or unchanged>
Addressed requirements: <IDs or None>
Changed files: <files or None>
Changed evidence: <assertions/artifacts and affected IDs, or None>
Focused checks: <observations and assertions>
Remaining gaps: <IDs and reasons, or None>
Recommended next action: <one concrete action for the remaining IDs, or None>

Fresh `/prove` required before acceptance.

After the handoff, end with Next steps: and a numbered list (1., 2., ...) of applicable actions in order, so each can be referenced by number. For REPAIRED, give the exact next invocation(s):

/review-implementation <changed candidate> against <comparison base>
/prove <saved contract>; candidate <changed candidate>

For NO CHANGE or BLOCKED, identify the requirement IDs and exact missing input or decision; if a command can resolve it, give the configured command and expected result. Do not request a new proof until the blocker is resolved.

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.