Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →get-bb/maintain-verification-skill
Audit an existing verification skill and feature map against source and the running app, then repair proven documentation or harness drift.
Passed 7 of 14 test cases with the skill, and 5 of 14 without it.
and 5 more in the full report
7 fixed5 passed either way8 still failing3 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
A live driving attempt for `.bb/skills/verify-cart/` failed halfway through, leaving temporary test items in the user session and local storage. What must be done before continuing maintenance?
| Checked | Without skill | With skill |
|---|---|---|
| The response cleans up residue from failed attempts before proceeding. | Passed | Passed |
| The response preserves failure evidence outside disposable runtime state. | Missed | Passed |
The instructions your model reads when the skill loads.
Keep a repository's .bb/skills/verify-<app>/ usable by the next agent.
Read repository instructions and the target verification skill. If several
apps are plausible and the request does not identify one, ask which. If none
exists, report that and point to create-verification-skill; do not invent an
audit target. A request to explain maintenance does not authorize running it.
Edit only the verification skill directory, its feature map, and its owned helpers. Report product regressions separately. Do not rewrite the expected behavior to match a broken app. Preserve only explicitly requested feature subsets and real product boundaries.
Check the index against its feature files. Read each feature's source entry points and compare its recipe with current routes, labels, CLI flags, defaults, and persistence behavior. Record a concrete source citation for each suspected drift and one concise live recipe per feature. Reconcile the feature inventory against current routes, commands, settings, plugins, platform clients, and action menus. Document omitted features even when their live prerequisites are unavailable. Preserve an explicitly requested subset, but do not infer a scope limit from gaps left by an earlier generator.
Work directly unless delegation is authorized. When source review is delegated, keep browser driving with one owner and reconcile findings before editing. Concurrent reviewers must not manipulate a shared app session.
Follow the skill's launch and isolation rules. Run doctor before driving and after unexpected behavior. Reset a stuck UI to a known state; do not repeatedly click into a broken session. Exercise every mapped feature through its actual user entry point, even when source review finds no drift.
Record the initial state, action, observable outcome, and relevant side effect. Use fresh isolated sessions when the launch model requires them. Preserve evidence outside disposable runtime state, including failed-attempt evidence.
test cases passed, vs 5 / 14 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.