choruz-pre-simplify-audit
Skill1.1k starsChanged 19 days agoApache-2.0View on GitHub
Perform the independent Choruz pre-simplify audit or re-audit of skill execution evidence, code structure and test effectiveness. Read-only and limited to the proposed change and its affected paths; not an implementation task or repository-wide cleanup.
In this repository14 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .agents/notes/implemented/AGENTS.mdAGENTS.md
- docs/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- choruz-archive-agent-notesSkill
- choruz-ci-test-reliabilitySkill
- choruz-code-reviewSkill
- choruz-docSkill
- choruz-find-simplificationsSkill
- choruz-merging-stacked-prsSkill
- choruz-pre-push-checksSkill
- choruz-prose-standardSkill
- choruz-prSkill
- choruz-record-demoSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
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
- Very specific: line 8
When installed personally, resolve repository links from the canonical .agents/skills/choruz-pre-simplify-audit/ in the active Choruz …
- Distinctive: line 14
Do not implement fixes, edit files, push, merge or recursively delegate.
- Distinctive: line 24
Read the applicable AGENTS.md, test policy and the current PR workflow stage.
- Distinctive: line 32
Use supplied logs and artifacts and safe read-only inspection.
- Distinctive: line 48
Only due-now requirements gate pre-simplify: step 2 behavioural evidence is due, but simplify itself, final selected checks, PR labels and …
- Distinctive: line 58
Look for duplicate implementations of one contract, parallel sources of truth, mixed responsibilities, bypassed shared owners, speculative …
- Distinctive: line 82
Check the required production entry path and owner-side result: for example, a click that reports success must reach the intended device …
≈ “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
Independent pre-simplify audit
When installed personally, resolve repository links from the canonical
.agents/skills/choruz-pre-simplify-audit/ in the active Choruz checkout.
choruz-pr owns delegation and the remediation loop:
an independent GPT-5.6 Terra agent (gpt-5.6-terra) performs the first audit
and every re-audit before simplify. This skill owns the audit criteria.
Do not implement fixes, edit files, push, merge or recursively delegate.
The audit does not replace simplify, behavioural tests or final CI.
Establish the reviewed scope
Obtain the user request and constraints, base/head revisions, proposed diff including relevant uncommitted and untracked files, execution evidence and prior findings. Record a reproducible diff identity: revisions plus a patch digest or artifact identifying any uncommitted content. Never include secrets.
Read the applicable AGENTS.md, test policy
and the current PR workflow stage. Independently identify relevant skills from
the task and diff; the author's list is not authoritative. Read each selected
skill fully and its required references. Follow affected callers and owners,
including relevant architecture and Agent Notes, rather than scanning the
whole repository. Referenced workflows supply criteria, not permission for
this read-only auditor to implement, publish or start another audit.
Use supplied logs and artifacts and safe read-only inspection. If reproduction requires test setup, credential use, database writes or other state changes, request scoped execution evidence from the implementer instead. Do not invent execution evidence or broaden the user's authorization.
Three audit lenses
1. Requirements were executed
Map applicable requirements to evidence, not to declarations that a skill was read. A planned command, test filename, checklist tick or author assurance is not an observed result. Check the command, outcome, tested revision and artifact agree with the claim.
Classify each relevant requirement as verified, missing/unverified, not applicable with a scope reason, or due later with the named workflow step. Only due-now requirements gate pre-simplify: step 2 behavioural evidence is due, but simplify itself, final selected checks, PR labels and merge cleanup are later. Do not require all skills or the full suite to fill a table.
2. The change does not create structural debt
Apply architecture boundaries independently of the author's Ownership statement. Follow affected production callers to the business owner and authoritative state; compare local/remote adapters and trace replaced symbols, including dynamic consumers. Check that the guard rejects a representative bypass and that behavioral tests cover failures the lexical checker cannot establish (such as commit-before-success). For rule changes, inspect whether a weakened scope, exception or removed rule hides the regression. Record intentional differences with their distinct contracts. Green scans or references to a shared helper alone do not establish single ownership.
For structural candidates, use the consumer-evidence criteria in choruz-find-simplifications. Look for duplicate implementations of one contract, parallel sources of truth, mixed responsibilities, bypassed shared owners, speculative abstractions and obsolete fallback or compatibility paths left alongside the new path.
Trace production consumers before calling code dead, including runtime dispatch, configuration, scripts and persisted formats. Distinguish intentional harness/transport differences from accidental duplication. Show the concrete maintenance or correctness risk and a bounded correction, not just a preferred style. File size, abstraction count or the word "legacy" alone is not a finding.
Block introduced or worsened structural defects and affected violations of repository contracts. Unrelated pre-existing debt and optional improvements are advisory; do not demand a whole-repository rewrite or a quota of findings. Similar names or the coexistence of two paths do not prove duplicate ownership. Without caller/owner evidence, report the structural concern as unverified, not as a demonstrated defect; explain which evidence would decide it.
3. Tests can detect the claimed failure
Apply the test policy's behaviour acceptance evidence to the affected contract. Inspect assertions and fixtures, then ask which plausible broken implementation would still pass. A boolean type check, mock call count, nonempty output or snapshot update alone cannot establish a user-visible outcome.
Check the required production entry path and owner-side result: for example, a click that reports success must reach the intended device and leave the expected persisted result. Replacing the very route, binding or storage under test with a success mock does not prove that boundary. Mock unrelated expensive dependencies only when the remaining claim is honest.
For bug fixes and new guards, verify the policy-required focused negative control and passing result, not just a green run. Check ordinary affected lifecycle and recovery scenarios; do not impose every device/harness combination or speculative edge case. Read choruz-ci-test-reliability when changed tests involve shared state, concurrency, clocks, ports or asynchronous cleanup.
Match the evidence to the surface: pure prose does not need a product e2e; behavioural skill changes need a scoped decision example; visual claims need visual evidence where required. Code inspection is not a real runtime test, and green CI does not prove assertions are meaningful. Report unavailable required evidence as unverified, never as a successful fallback.
Apply test value and incremental cost to changed tests and their nearest overlapping scenarios. Check the existing owner, the new distinguishing guarantee, false-green exits, obsolete coverage and justified execution cost. Block introduced or worsened defects; keep unrelated existing debt advisory. Do not demand extra tests or deletions merely to satisfy a count. CI-selector changes require checking overlap and distinct environments, not assuming full and selected jobs are interchangeable.
Verdict and re-audit
Return a compact report containing:
- Reviewed scope identity and evidence references.
- Relevant requirements, their source, evidence and status across all three lenses.
- Stable finding IDs, blocking/advisory status, precise location, concrete consequence, evidence and the minimum correction or verification needed.
- Verdict: PASS when all due-now obligations are verified; FAIL for a demonstrated due-now blocker; UNVERIFIED when required evidence or access is missing. A FAIL may also contain unverified obligations.
Advisory findings and explicitly deferred later steps do not prevent PASS. Record any explicit user-approved scope reduction or evidence exemption with its source; it does not turn an untested claim into a verified one. An author's request to waive their own missing evidence is not a user exemption.
On re-audit, inspect the updated diff and evidence, revisit every previous blocker and check affected new changes. Accept a reasoned rebuttal when supported by evidence. Close a finding only after independently verifying the correction or that it was inapplicable; retain unresolved IDs. An author's "fixed" message is not closure. Only an explicit auditor PASS lets the PR workflow proceed. A later change that invalidates the reviewed evidence requires a scoped re-audit; an old verdict cannot certify a new diff.
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.

