IMPLEMENTAUDIT
On this page
  1. Splash
  2. Overview pages
Claude Code + Codex skill - v0.3.3.3 correction in progress

IMPLEMENTAUDITImplementation runs that close like audits.

Reusable meta-engineering for repo changes that need evidence: sustained planning, design, implementation, improvement, recovery, integration, and closure, not only a one-off code audit.
Give it one request. It then:
  • creates or binds a work record, with a run folder for phased work
  • writes state and phase checklists
  • runs before/after checks
  • records failed-evidence paths
  • reaches AUDIT_COMPLETE or leaves an honest handoff.
A skill or agent-tooling repo can itself be the audit object. This repo proves self-application here, not effectiveness across every agent, host, repository, or external ecosystem.
v0.3.3.3 remains the public identity for runtime 0.3.3. Its premature first publication is retained as superseded evidence; the final same-identity candidate remains pending publication and fresh public download/install readback.
Releasev0.3.3.3 final correction / runtime 0.3.3
Example runguided
One pasted request binds the work, writes state, runs checks, stops on failed evidence, and closes only after final audit.
direct requestexisting repomarkdown evidence
example phased run folder (run root): .IMPLEMENTAUDIT/runs/demo
plan the work contains run-folder tracking, which contains phase checks; failure can interrupt any layer, and final audit closes only after evidence. Formal loop names live in the runtime model.
/implementaudit add tests for login timeout

RUN create or bind work record - run folder when phased
STATE write STATE.md - IN_PHASE - phase 3/5
PLAN phase checklist - where the change belongs (owner/source)
CHECK before-checks - baseline (Smoke A)
STOP ANDON_PROBE - stop, classify, inspect cause
FIX countermeasure - bounded fix; escalate only with new evidence
RERUN evidence - passing vs baseline (Smoke B)
AUDIT_VERIFY - every in-scope item checked
AUDIT_COMPLETE - final checkable evidence recorded
IMPLEMENTAUDIT_RUN_COMPLETE
Overview

Start with the splash, then use the pages.

Start hereUse Quick Start when you want the first pasted command.
Use the sidebarPortal pages live in the sidebar, with stable next and previous links.
Runtime modelThe full loop model is a page, not a collapsed overview drawer.
EvidenceThe proof page shows the files, markers, and final audit surfaces.