BB impostor-hunt
Hunt the impostor in a "finished" deliverable. An impostor is a result-correct artifact whose causal chain is NOT the one the user's purpose required — the test set leaked, only the happy path is wired, correlation got relabeled as cause, the config file exists but no code path reads it, the wrapper greps stdout for "OK" without checking the exit code, the agent "can refactor" backed by one cherry-picked example. The output looks right. The logic that produced "right" is the wrong logic. AUTO-INVOKE when a turn declares completion AND the user's original purpose is recoverable from context. Completion signals include: "done", "implemented", "finished", "works now", "ready to merge", "all tests pass", "task complete", "feature shipped", an agent posting a summary with checkmarks, a PR description, a CHANGELOG entry, or a closing message that asserts the work is over. Original purpose is recoverable from: the user's first message in the thread, the issue/ticket text, the PR title and description, the commit message that opened the branch, the README's stated goal, or an explicit prompt the user pasted at the start. If both signals are present, invoke without waiting for an explicit request. Also invoke when the user explicitly asks to audit completion truth, detect fake completion, check for hidden goal misalignment, look for mock-driven success, test-set leakage, happy-path-only delivery, correlation-as-causation, configured-but-unread settings, logs-say-success-but-return-code-unchecked, or similar patterns. Triggers in any language; the report mirrors the user's language. DO NOT invoke for ordinary bug hunting (route to code-review), style or refactor cleanup (route to simplify), or behavior-by-execution verification (route to verify / run). DO NOT invoke when the user is mid-implementation and has not yet declared the work done — interrupting an in-flight task with a completion audit is noise. DO NOT invoke when the original purpose is not recoverable from cont
Hunt the impostor in a "finished" deliverable.
As a process B 75/100 · Nearly there — weak spots: inputs and preconditions, running it twice, progress reporting
How to improve
- Shorten the description to 1024 characters.
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 0
✓ No critical or high findings
Files scanned: 8. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- error
description-longdescription is 2043 chars, limit 1024 - note
description-budgetdescription takes 2043 of the ~15000-char shared budget for all skills
Process rating: all ten parameters 75/100
- 0Inputs and preconditions. Does not say what the process needs to start
- 0Progress reporting. Says nothing while it works
- 30Running it twice. 6 mutating operations with no state check
- 60Result and completion. Output format stated, no completion criterion
- 70When it triggers. States when to use, but not when not to
- 100Tools and files. No external tools needed
- 100Steps. 38 steps
- 100Failures and branches. 3 branches, has a failure section
- 100Consistency. Name and required fields are in place
- 100Execution cost. Instruction body is 2561 tokens
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +4Description does not say when NOT to use the skill (false activations)
- +3Description length 2042: 120–800 characters recommended
- +1No license
- +2Single-language instructions
- +5Description quotes 9 example trigger phrases
- +4Structure: 22 headings
- +3Step-by-step instructions: 38 items
- +3Output format is stated explicitly
- +4Has examples (1 code blocks)
- +4Reference files are cited in the instructions (6 of 6)
Quality base 70; lint remarks subtract, signals add up to 100. Result: 65.