agentleFS
simiancraftsimiancraft/simiancraft-skills/skills/prove-work-on-github/SKILL.md

prove-work-on-github

Skill8 starsChanged 4 months agoMITView on GitHub

Prove that work claimed on a pull request or issue actually landed and is sound, with verifiable evidence a reader can independently re-check. A claim is narrative; a receipt is proof. Use when deciding what evidence a change owes, capturing it (screenshots, GIFs, logs, database shape, over-the-wire results, test and coverage output), storing it in a durable evidence branch, rendering it inline in a PR or issue comment, judging whether the evidence is adequate to merge, or checking whether existing proof has gone stale. Triggers: "prove this works", "what evidence does this PR need", "attach proof to the PR", "is this enough to merge", "evidence locker", "is this proof still valid". Skip for trivial changes whose proof is the diff itself, and for work that never touches GitHub.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 5 are distinctive, and 0 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.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Make every receipt re-checkable by someone else: If a second reader cannot click it, re-run it, or verify its hash, it is not proof.

    line 73
  2. Distinctive:

    When the signal is genuinely new, extend the rubric (in your copy, or propose it upstream).

    line 109
  3. Distinctive:

    Every load-bearing claim about the change has a receipt.

    line 114
  4. Distinctive:

    No receipt leaks a secret, and no capture mutated the system under proof.

    line 118
  5. Distinctive:

    If any item fails, you have a story, not proof.

    line 120

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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 Work on GitHub

A claim is narrative; a receipt is proof. The remote is the witness, not your intent to push.

A statement about a change ("the fix works", "CI is green", "the screen looks right", "it is ready to merge") is narrative until it is paired with a receipt: verifiable evidence that you observed and that a reader can independently re-check. "I ran the command" is not "the remote advanced." A green checkmark is not "the feature is correct." This skill is the discipline that turns claims into receipts, sizes how much proof a change owes, stores the receipts durably on the pull request, and judges whether they are enough to merge.

Proof is asymptotic

Proof is never perfect; it only approaches the asymptote of perfection. The categories below are open sets, not closed enums: when you find a new signal that makes a change more provable, add it (extend your copy, or propose it upstream). The rubric ships incomplete by design and improves by contribution. A skill that claims a finished, total account of proof is lying.

When proof is owed

Proof is owed when a claim about a change would change what a reviewer or a merge gate does, and the claim is not self-evident from the diff. Match the rigor to the change (see references/physical.md); a typo and a schema migration do not earn the same receipt. Skip when the diff is its own proof, or when the work never reaches GitHub.

The model: three aspects and a judgement

Three measured aspects of a change, and a judgement that interprets them. Judgement is last.

AspectQuestion
Physicalwhat is the change, and how much does it therefore owe (surface area, complexity, and reversibility)?
Correctnessis the change sound (alignment, verifiability, durability, and security)?
Evidencewhat receipts does that change require, by type, and at what fidelity?
Judgementis the evidence adequate, and how confident are you to merge?

Physical sizes the bar; Correctness names what must be proven; Evidence supplies the receipts; Judgement reads the three and returns an action: merge, gather more, or block.

The lifecycle (what you do)

#Step
1Size the change (Physical) and name what it must prove (Correctness)
2Acquire the evidence the change owes, by type
3Store it in the evidence branch and render it inline
4Judge adequacy and confidence, and act
5Re-judge when the repo moves and the proof goes stale

Cross-cutting non-negotiables

  • Pin to an immutable referent. Cite a full commit SHA, a content-addressed artifact, or a permalink with the commit hash; never a moving target. The embedded SHA is what fixes the bytes under the reference; append-only is a convention kept on top of that, not something the branch enforces.
  • Never let a receipt leak. A pasted log can dump a token; a screenshot can catch a secret. Redact before you store or present.
  • Keep capture read-only. Acquiring proof of a change must not alter the system under proof.
  • Make every receipt re-checkable by someone else. If a second reader cannot click it, re-run it, or verify its hash, it is not proof.

Reference index

NeedRead
Size the change: surface area, complexity, reversibility, and the metrics behind themreferences/physical.md
Is the change sound: alignment, verifiability, durability, and securityreferences/correctness.md
Judge adequacy versus confidence and map the score to an actionreferences/judgement.md
Capture the evidence a change type owesreferences/acquire.md
The evidence branch: storage, protection, naming, and fidelityreferences/evidence-locker.md
Prepare an artifact for GitHub: format, compression, and sizingreferences/optimize-assets.md
Compose and render proof inline in the commentreferences/render.md
The per-artifact metadata that ties it togetherreferences/artifact-manifest.md
When proof has gone stale and what to reacquirereferences/freshness-and-reproof.md
Worked examples of strong and fake proofreferences/catalog.md

Extending this skill

A living document; the categories grow as you find new ways to prove. Every new fact has one home:

A new fact is...Home
a sizing signal for the changereferences/physical.md
a soundness signal (alignment, verifiability, durability, or security)references/correctness.md
a judgement element or a scoring rulereferences/judgement.md
a capture method for a change typereferences/acquire.md
a storage, protection, or naming rule for the evidence branchreferences/evidence-locker.md
an artifact format or optimizationreferences/optimize-assets.md
a rule for rendering proof inlinereferences/render.md
an attribute of an artifactreferences/artifact-manifest.md
a staleness or reproof rulereferences/freshness-and-reproof.md
a cross-cutting non-negotiablethis SKILL.md
a worked example of strong or fake proofreferences/catalog.md

When the signal is genuinely new, extend the rubric (in your copy, or propose it upstream); see "Proof is asymptotic."

Acceptance: a claim is proven when

  • Every load-bearing claim about the change has a receipt; nothing rests on narrative alone.
  • Each receipt cites an immutable referent and is re-checkable by a second reader.
  • The evidence covers what the change owes (its Correctness claims) at the depth its Physical size demands.
  • Judgement returns merge: adequacy clears the bar, and confidence is not undercut by an uncoverable blind spot on a load-bearing claim.
  • No receipt leaks a secret, and no capture mutated the system under proof.

If any item fails, you have a story, not proof.

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.