verify-before-done — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited verify-before-done (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
A clean compile is not verification. Types passing is not verification. "I updated the handler" is not verification.
You can state facts about what you ran. You cannot state facts about what should happen.
No "done" without evidence. Run a command that exercises the change, read the output, then claim completion.
| You changed… | Minimum proof |
|---|---|
| App logic / API | Targeted test or curl/request that hits the path |
| UI behaviour | Render + interaction check (pair with visual-quality for pixels) |
| Build / types | build, tsc --noEmit, cargo check, etc. |
| Refactor only | Tests that cover touched modules |
| Config / env | Command that loads config or a smoke script |
| SQL / migration | Dry-run, EXPLAIN, or migration against a scratch DB |
If nothing can be run, say "unverified — manual step required" and list exactly what the human must run. Do not imply it works.
Use Bash (or the project's test tool). Prefer the narrowest command that still falsifies the bug.
Examples (adapt to the repo):
npm test -- --runInBand path/to.test.ts
pnpm vitest run src/foo.test.ts
pytest tests/test_checkout.py -q
cargo test -p mycrate checkout::
go test ./internal/checkout/...
xcodebuild test -scheme App -destination 'platform=iOS Simulator,name=iPhone 16'
npm run build
pnpm lint && pnpm typecheckRead failures literally. Fix, re-run, until green or you document a known flake with evidence.
Include:
When verification is genuinely complete, end your assistant message with:
VERIFY-BEFORE-DONE: PASS — <command> → <outcome>Example:
VERIFY-BEFORE-DONE: PASS — pytest tests/test_invoice.py -q → 4 passedThe Stop hook looks for this line after code edits. Without it, the turn may be blocked.
False positive (docs-only change, user said skip tests): touch .verify-before-done-off in the project root, or export VERIFY_BEFORE_DONE_GATE=off.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.