AmmarBibi
UserSelf-verification hooks + a reasoning skill for Claude Code - keep the agent honest across your prompts, its claims, your tests, parked work, and memory. Fail-silent, stdlib-only, no network.
Categories
Indexed Skills (3)
consistency-audit
Catch consistency drift in a deliverable (report/doc/slide deck) - numbers, figures, or claims in the prose that no longer match the source-of-truth data (CSVs, computed results, sweep outputs). Run before submitting or re-exporting a report, after regenerating results, at each revision, and whenever the user asks "do the numbers in the report still match the data / are all figures referenced / is anything stale or fabricated / did a value drift". The mechanical half (extract every cited value + figure ref, cross-check against the source data with tolerance) is a bundled script; the reasoning half (is a claim actually SUPPORTED, is the interpretation consistent across sections) is Claude's - which is why this is a skill, not just a hook.
meta-review
A repeatable meta-hardening pass for any project. Claude (the reasoner) audits for parked-but-unscheduled work, instance-only fixes lacking a durable mechanism, code/structure/perf optimization gaps, things missing or wrong, and improvements that would lead to a better outcome - then schedules or fixes each. Use at milestones, before a release/cutover, or whenever the user asks "am I missing anything / is this optimized / did we only patch the instance". This is the REASONING half that hooks cannot do; hooks only surface state.
completeness-audit
Guarantee a plan/roadmap stays a TRUE completeness ledger for its authoritative sources - catch both parked/deferred-and-forgotten items AND source content silently missing from the plan. Run when a plan is created or rebuilt, at milestones / after each cluster ships, before declaring any source or the project "done", and whenever the user asks "did we forget anything / is this complete / did you defer-and-forget". Prevents the failure where a plan asserts completeness ("each gap has a home", "essentially all built") without it being verified against the sources.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.