derivation-audit
Skill2 starsChanged 8 days agoMITView on GitHub
Audit formal claims, registered checks, and symbolic derivation scope. Use when asked to review a computer-algebra derivation, a paper proposition, or a solver-backed proof. Do not use for ordinary prose claims, a whole-tree engineering audit, or a contest paper workflow.
In this repository21 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agent-artifactSkill
- agent-discoverySkill
- agent-run-supervisionSkill
- api-contract-migrationSkill
- changelog-updateSkill
- change-reviewSkill
- ci-reviewSkill
- claims-auditSkill
- dependency-auditSkill
- durable-cacheSkill
- engineering-auditSkill
- infer-conventionsSkill
- operational-signal-intakeSkill
- planned-releaseSkill
- preferred-stackSkill
- pressure-benchmarkingSkill
- public-surface-reconSkill
- rfc-processSkill
- security-auditSkill
- session-reviewSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
Keep the statement, the domain, and the quantifiers.
- Very specific: line 50
Write durable findings under docs/issues per docs-conventions.
- Distinctive: line 58
When a named skill is unavailable, preserve the finding and name the deeper review.
≈ “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.
Files
Derivation audit
Review a formal claim and the checks that claim to support it. This skill does not replace claims-audit. It owns identity, implication, registered rows, and solver scope when a consumer asks for that job by name.
The consumer owns the executable and the style exemplar. Stay on concepts. Do not name a computer algebra system.
Quick reference
keep statement domain quantifiers -> classify the claim -> inspect registered checks -> treat solver output -> record unresolvedClaim classes
Keep the statement, the domain, and the quantifiers. Distinguish:
- identity;
- equivalent solution sets;
- one-way implication.
A numeric example does not replace a quantified claim. Solver or test order is not uniqueness. When a condition is rewritten, show both sides and the domain of the rewrite.
Registered checks
A Boolean check table certifies only its registered rows. A nonempty table is not a proof of the surrounding proposition. Completeness, consistency, and quality are different questions.
A green gate certifies the assertions it ran, not the surrounding product claim. Missing runtime is not a successful static pass.
Solver output
Timeout, unevaluated output, and a failed solver check stay unresolved until a counterexample or a closed proof exists. Do not treat an unresolved result as a refutation.
A cited standard or theorem used as support needs a recorded statement with its hypotheses, not a name-drop.
Freeze
Quantities that come from the derivation stay unverifiable until they are frozen from the producing artifact. Do not hand-edit the freeze. A later file does not prove an earlier gate.
Record
Write durable findings under docs/issues per docs-conventions. Outcomes follow claims-audit: supported, partially supported, unsupported, outdated, contested, unverifiable. Add unresolved when the solver did not close the claim.
Routing
claims-auditfor documentary facts, quotations, and freeze of prose quantities.engineering-auditwhen the surrounding product claim is a green test gate.rfc-processwhen an RFC states a theorem; reuse these statuses only when that RFC asks.
When a named skill is unavailable, preserve the finding and name the deeper review.
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
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.