/maintain-verification-skill

name: maintain-verification-skill
description: "Re-sync the project's verification skill when its commands drift — re-run each recorded command, update expectations and the feature map to what is true now, record the verdict actually earned."
aliases: []
metadata:
  credits:
    verification-skill-maintenance: pstack
    author: Lauren Tan
    url: "https://github.com/cursor/plugins/tree/main/pstack"
    port: pstack-claude
    port-author: Michael Denyer
    port-url: "https://github.com/michael-denyer/pstack-claude"

Maintain Verification Skill

A verification skill that names dead commands is worse than none: it fails loudly on trivia and passes quietly on what matters. This skill re-syncs the project's `verify` skill to the repo's current reality.

When to use

• verification errors on a command that no longer exists or was renamed
• a new surface (package, app, e2e suite) has no check recorded
• after a toolchain or CI migration

Workflow

1. **Audit the recorded skill**: list every command and manual check it names; for each, does the command still exist and still mean the same thing (flags, scripts, config)? 2. **Re-run, don't re-read**: execute each surviving command and compare the observed outcome to the recorded expectation. A command whose passing shape changed gets its expectation rewritten, with the new evidence. 3. **Find uncovered surface**: what does the repo build/check now that the feature map does not mention? Add the check by running it, or record the gap as uncovered — the map stays honest. 4. **Update the skill and feature map in place**: same file, corrected commands, corrected coverage table, dated revision line at the top. 5. **Record** the resync through the ledger's locked writer — `appendLedgerEntry` and `defaultLedgerPath` from the installed package's `dist/features/verification-ledger/index.js` (`babysit`'s ledger step demonstrates the exact invocation); never append raw rows to `.omc/state/verification-ledger.tsv` — the writer owns header creation, locking, and cell sanitization. The verdict is the one the rerun actually earned — `unit-test-verified` when the test suite ran, `live-ui-verified` for a live-UI flow, `type-check-only` when only build/lint/typecheck commands were rerun — and those commands as evidence. Never claim a test ran that did not.

Rules

• The rerun is the resync; nothing is marked current without execution.
• Coverage that shrank is reported out loud, not silently dropped from the

map.

• If the project's verification has outgrown the single skill shape (multi-

package, per-app checks), say so and split per surface rather than fattening one file.