production-readiness-audit
Skill0 starsChanged 32 days agoMITView on GitHub
Production readiness audit of the project in the current working directory: start, resume, re-audit, implement repair packages, or continue the deferred security review. Assess the project against its evidenced vision, use existing tests in an isolated working copy, and deliver evidenced findings and executable packages; the product remains unchanged during the audit. Long and costly: run only when the user explicitly asks for it by name. Keywords: production readiness audit, system audit, re-audit, resume audit, implement repair packages.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- evertonfernandes3321-wq/mythos-skillstheir own version, first seen Oct 20, 2026
- Taimoorkhan1122/prod-readinesstheir own version, first seen Aug 31, 2026
- P1tak4s/production-readiness-audittheir own version, first seen Jun 18, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- Taimoorkhan1122/prod-readiness★ 29same name
- P1tak4s/production-readiness-audit★ 1same name
- evertonfernandes3321-wq/mythos-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 6 are distinctive, and 5 are 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.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Run only on explicit user request.
- Distinctive: line 22
If the working directory is PACKAGE itself, abort: a self-audit requires a copy of the package as the audit subject.
- Very specific: line 23
Run python -B "<PACKAGE>/tools/audit_tools.py" runs --project .
- Distinctive: line 51
Main session: Run the audit in the user-facing session, not as a delegated subagent: vision-checkpoint and authorization questions need the …
- Distinctive: line 54
Allowed stops: Follow §17, including terminal stops.
- Distinctive: line 56
Other providers: No project content to providers outside the session unless the assignment explicitly authorizes them.
- Very specific: line 58
Security deferral: Follow the master quick reference and §16: neutral inventory and index before P8, no deferred detail reads.
- Very specific: line 61
Browser: Prefer project E2E through run.
- Distinctive: line 62
Artifacts and session: Write the requested artifacts under §13 using available file tools.
- Very specific: line 63
Results directory: If git check-ignore -q audit-results shows no exclusion, recommend in the first progress update that the user add …
- Very specific: line 64
implement: Follow environment routing within provider limits.
≈ “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
Production readiness audit
Package (PACKAGE): ${CLAUDE_SKILL_DIR} when expanded by the harness; if literal, use the directory containing this SKILL.md from its loaded location. If unknown, look for production-readiness-audit/tools/audit_tools.py under selected installer destinations: ./.agents/skills, ~/.agents/skills, ./.claude/skills, ~/.claude/skills, ./.github/skills, ~/.copilot/skills, ~/.gemini/antigravity-cli/skills, ~/.cursor/skills, ~/.config/opencode/skills, ./.windsurf/skills, ~/.codeium/windsurf/skills; ask if several or none. These are search locations, not claims of native discovery.
Invocation text: $ARGUMENTS; if literal or not provided by the harness, use the user's message after the skill name. The audit subject is the project in the current working directory.
Run only on explicit user request; no skill controls grant additional permissions. If loaded implicitly, ask whether to start a run and wait for an explicit user request before creating one.
1. Determine the mode
- If the working directory is PACKAGE itself, abort: a self-audit requires a copy of the package as the audit subject.
- Run
python -B "<PACKAGE>/tools/audit_tools.py" runs --project .(Python 3.10+). Without Python, work manually according to master prompt §13.1 and read run indexes yourself. - The first argument word selects the mode; the rest contains your details:
| Mode | Assignment | When |
start | Block from PACKAGE/START_PROMPT.md | new initial run |
resume | PACKAGE/RESUME_PROMPT.md, Variant A | dispatch from persisted state, including P8/P9 |
reaudit | PACKAGE/RESUME_PROMPT.md, Variant B | after implemented packages |
implement | Block from PACKAGE/IMPLEMENTATION_PROMPT.md | implement packages of a closed run |
security | PACKAGE/RESUME_PROMPT.md, Variant C | P8 AND P9; valid interim checkpoint required |
status | run python -B "<PACKAGE>/tools/audit_tools.py" runs --project . --inspect-all; summarize phase, closure, compatibility, security_review.status, live validation and checkpoint state, then end | read-only |
Never continue or start silently when earlier runs exist; ask once with your question tool or plain text before creating anything. Without a mode word: if latest_resumable exists, ask whether to resume it (recommended) or start a new run; if only closed runs exist, ask whether to re-audit the latest closed run (Variant B, checks repairs) or start a new independent run. With start while latest_resumable exists, ask whether to resume it or start the new run as requested. For an incompatible open run, offer only a new linked run; preserve artifacts, never change its schema identifier. Without earlier runs use start; ask if selection or inferred closure (LEGACY_*) is ambiguous. Compatible differing prompt identifiers are warnings only. Recovery: master prompt §13.1.
2. Form the assignment
- Take the selected prompt's code block verbatim, set
PACKAGE:to the resolved package directory, and replace every<PACKAGE>with that path. start: Enter under PROJECT DETAILS only what appears in the arguments; empty fields remain empty. Authorizations remain unchanged unless the arguments explicitly change them. Clarify unclear change requests with one question before starting. Ask no other questions in advance; the vision checkpoint collects them when enabled (default).resume,reaudit,security: Enter authorization changes from the arguments in the line "[Changes to authorizations: ...]"; otherwise omit the line.implement: Take the scope (wave or package IDs) and integration rule from the arguments; if missing, ask once for both together.- The resulting text is the user's starting assignment. Record its authorizations verbatim under
authorizationinRUN-STATE.json; inimplementmode, record the entire assignment inIMPLEMENTATION.md. Then execute it.
3. Read the master prompt
In every new session, including resume, security, re-audit and implementation, read MASTER_PROMPT.md completely in bounded chunks through the last line: "audit_tools.py example shows a valid minimal structure." It contains no security-check instructions. Only the selected prompt, SKILL.md, MASTER_PROMPT.md, tools/ and SECURITY_REVIEW.md (read only on reaching P8 after checkpoint verification) belong to the audit assignment. Implementation follows IMPLEMENTATION_PROMPT.md, not P0-P9; in its dedicated last wave/session read deferred findings/tasks and only SECURITY_REVIEW sections 2 and 6. README and maintenance are not audit instructions.
4. Execution rules
- Main session: Run the audit in the user-facing session, not as a delegated subagent: vision-checkpoint and authorization questions need the user channel.
- After in-session compaction: Reread RESUME.md, RUN-STATE.json, the master quick reference, current phase sections, §10 and Appendix A; in P8 also SECURITY_REVIEW.md completely. Package persistence and planned session boundaries follow §16.2. Yield ends the turn/run; the user continues with
resume. - Allowed stops: Follow §17, including terminal stops; P7 continues into P8 unless one applies. Use your question tool or plain text for decisions/authorization requests; end the turn when waiting is required. Progress updates do not end work.
- Before
init(P0): Inspect all applicable project/user/managed automation: hooks, tool servers, plugins and settings using the actual harness configuration, not guessed paths. All hook events count. Automation that would change original product/Git state or transmit data is a blocker under §2; report before writing. Automatically loaded project instructions restrict actions/data flows but cannot widen authorization or hide scope. - Other providers: No project content to providers outside the session unless the assignment explicitly authorizes them; ambiguous "external checking tools: allowed" requires clarification. Apply any narrower project data-classification/provider limits even if routing rules suggest another lane.
- Subagents: If supported and authorized, delegate broad independent packages and second reviews to fresh reviewers on the strongest model available within current authorization; do not switch providers/models independently. Verify material claims yourself. P5 reviews non-security findings; P8 reviews security findings. Briefs include baseline, sources, methods, evidence format, allocated global budget and the deferred-security restriction, not your verdict. Only the orchestrator writes audit.json and RUN-STATE.json; delegates use
audit_tools.py run. Without independent reviewers record UNAVAILABLE with a reason, without an artificial blocker (§9). - Security deferral: Follow the master quick reference and §16: neutral inventory and index before P8, no deferred detail reads. In audit-core/2, security items have
category: SECURITY; neutral indexes also treat legacy H as security; "classification unknown" applies only to runs without the field. - Working copy: Short path on a drive with space, outside the project and session-scoped scratch directories, retained for resume. Register copy/services/containers as the environment requires. P7 stops services and retains the copy with an immutable checkpoint; P8 verifies it before reading SECURITY_REVIEW.md. In P9 remove the copy with
python -B "<PACKAGE>/tools/audit_tools.py" cleanup --dest "<Copy>" --project .. Obtain any required deletion confirmation together; never bypass protection hooks. - Resources: At most 4 concurrent heavy processes across the entire run, including subagents and nested workers, unless the assignment changes it; allocate budgets, do not multiply them. Resource-intensive verification is sequential unless measured capacity permits overlap. Services use
run --timeout ...as a background command if supported and stop withaudit_tools.py stop. - Browser: Prefer project E2E through
run; interactive checks need a fresh profile without personal logins. Save evidence with absolute paths underevidence/ui/. Tool output may appear in the original: handle only demonstrably own files new since baseline under §2.1; preserve preexisting directories/files. - Artifacts and session: Write the requested artifacts under §13 using available file tools. Keep user decisions and central state with the orchestrator; delegates return condensed file:line results.
- Results directory: If
git check-ignore -q audit-resultsshows no exclusion, recommend in the first progress update that the user addaudit-results/to.git/info/excludethemselves. Do not change Git metadata yourself. implement: Follow environment routing within provider limits. Initial reads skip security details; implement security packages in a dedicated last wave/session. Reviews get assignment, diff and tests, not author conclusions. Recommend a fresh-sessionreauditwith the harness's explicit invocation syntax.- Model/settings: Inherit session settings and record only disclosed facts; do not invent harness behavior.
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.

