← ClaudeAtlas

implementauditlisted

Execute audit-governed work to closure or handoff. Activate for /implementaudit and audit closure. For scheduling/dispatch/resource ceilings, read references/child-agents.md.
theislampill/IMPLEMENTAUDIT.md · ★ 16 · AI & Automation · score 66
Install: claude install-skill theislampill/IMPLEMENTAUDIT.md
# /implementaudit Concise runtime bootloader for audit-governed implementation. Details live in the packaged references, templates, and helper scripts named below. Read this file once for the spine, then use progressive disclosure: inspect only the owner/source sections needed for the current gate. Source checkout layout is conventional and name-matched: `skills/implementaudit/SKILL.md` with sibling `references/`, `scripts/`, and `templates/`. Release archives flatten that directory only as a build artifact: installed runtime paths are `SKILL.md`, `references/`, `scripts/`, and `templates/` under the active skill directory. Every finding closes. No orphan items. No unsafe actions. No proof claim without evidence. Continue until every item is terminally `done`, `changed`, `blocked`, `deferred`, or `unverified`. First executable dogfood rule: Do not read this entire skill or installed payload. Baseline the target repo first, then use progressive disclosure for only the owner/source sections required by the current gate. --- ## Dogfood Bootstrap / Read Map Do not read this entire installed `SKILL.md` before acting, and do not chunk through a truncated payload. Baseline first, discover actual host syntax, then use headings and targeted `rg`/grep for the live owner/source files. Package proof uses deterministic checks, not model-visible full-file readback: manifests, archive listing, checksums, `build-release-asset.sh --check`, installed file existence, and `verify-package.