agentleFS
netanelavrnetanelavr/dotcursor/skills/kit-prove/SKILL.md

kit-prove

Skill6 starsChanged 27 days agoMITView on GitHub

Require live proof for any “works / tested / validated” claim — CLI, raw JSON/logs, timestamps, or a real runtime path. Use when verifying a fix, smoke, or “validate it works”. Not for stripping a finished diff or design debate.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 5 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    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. Very specific:

    Not stripping extras from a finished diff (kit-strip-diff)

    line 13
  2. Very specific:

    Not a substitute for kit-review judgment

    line 15
  3. Very specific:

    Not production incident work (kit-debug-prod)

    line 16
  4. Distinctive:

    Not leaving probes / TEMP bypasses in git

    line 17
  5. Distinctive:

    When the claim is a bug fix, a diff or a new passing test alone is not enough.

    line 32
  6. Distinctive:

    If the tree is already fixed, temporarily restore the broken path long enough to recapture before, then restore and recapture after — or …

    line 37
  7. Distinctive:

    If only a derived visual exists: label reconstruction and still attach the raw source.

    line 54
  8. Very specific:

    Never commit validation junk under docs/ (or similar).

    line 56
  9. Distinctive:

    Prefer short reproducible commands over long narrative.

    line 80

≈ “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

Prove

Inverted burden of proof: a claim is guilty until there is a live artifact.

What this is NOT

  • Not stripping extras from a finished diff (kit-strip-diff)
  • Not “I think this approach is right” without a run — that’s opinion, not proof
  • Not a substitute for kit-review judgment
  • Not production incident work (kit-debug-prod)
  • Not leaving probes / TEMP bypasses in git

Claim → proof

ClaimAcceptable proof
Command / test passedFull command + exit + relevant stdout/stderr
API behavesRaw request/response JSON (redact secrets)
UI fixedReal UI screenshot of the actual run, or recorded steps + DOM/network evidence
Reproduced bugSteps + failing artifact before fix
“Deployed / published”Registry/CI URL or CLI output with version + timestamp
Feature works at runtimeExercised real path (adapter below) + artifact — tests alone are not enough when the change is risky

Bugfix before / after

When the claim is a bug fix, a diff or a new passing test alone is not enough.

  1. Before — reproduce the failure on the real path; save the artifact (HTTP/JSON, logs, UI, CLI).
  2. After — same steps, same environment, against the fix; save the matching artifact.
  3. Show both side by side in the summary so the delta is obvious.
  4. If the tree is already fixed, temporarily restore the broken path long enough to recapture before, then restore and recapture after — or say clearly what you could not reproduce.

Same endpoint / query / UI flow / log line for both captures.

Rejected as proof

  • Mockups, HTML→PNG “screenshots”, reconstructed diagrams sold as authentic
  • “I ran it” with no artifact
  • Summaries without the underlying command + output

Flow

  1. List claims that need proof.
  2. Isolated checker (prefer for multi-step or risky changes): spawn a new subagent. Give it only claims, done criteria, how to run proof, and paths in scope. Do not pass the build chat, your rationale, or “it should pass.” That checker runs this skill and reports artifacts or unproven. Tiny obvious fixes may stay inline; say so.
  3. Depth: Tiny change + strong tests + obvious path → you may skip deep probes (extra adapters/instrumentation). Say why. This is not permission to claim “it works” without an artifact.
  4. For each claim you still assert: run or request the real path. Pick an adapter when the user asked to validate/verify a feature (not only a one-line smoke).
  5. Paste/attach the raw artifact (keep timestamps).
  6. If only a derived visual exists: label reconstruction and still attach the raw source.
  7. Cleanup — remove probes; git diff + search for TEMP/debug markers must be clean before push.
  8. Never commit validation junk under docs/ (or similar); keep proofs in chat/PR comments.

Adapters (pick one when you need a runtime path)

Discover ports from compose/env/package.json / listening processes — never assume defaults.

SituationApproach
Local HTTP APIHit real endpoints; assert status/body
Hard-to-reach backend pathTemporary checkpoints → local collector / logs (remove before commit)
UI / browserDrive the real UI or attach to an existing session
User must clickPrep harness first, then instruct — don’t ask them to set up

Output

  • Claims vs artifacts
  • What was exercised (if a runtime path)
  • Gaps / unproven
  • Confirm probes removed

Rules

  • Missing artifact → say unproven. Do not soften to “likely works”.
  • Early stop only skips deep probes — any “it works” / “validated” claim still needs an artifact, or mark unproven.
  • Prefer short reproducible commands over long narrative.

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.