Verification before completion
Skill IcodeNet/agent-skills/skills/verification-before-completion
Gate every completion claim behind executed evidence. Use before saying "done", "fixed", "works", "passing", "delivered", or "merge-ready" — for any code change, bug fix, refactor, or configuration change. Requires running the relevant command and observing the output in this session; code inspection alone never counts as verification.From its SKILL.md
npx -y skills add IcodeNet/agent-skills --skill verification-before-completionAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
2.7 KB, 542 tokens by cl100k_base, as published. Nobody here has run it
Verification Before Completion
"The code looks right" is not a result. A completion claim is a claim about behavior, and behavior is only known by executing something and reading the output.
The rule
Before any completion claim:
- Name the claim precisely. "Fixed the timeout bug" / "feature X delivered" / "refactor is behavior-preserving" — a vague claim can't be verified.
- Run the verifying command in this session. The relevant test suite, the failing repro, the build, the endpoint call, the UI flow. Prefer the project's own package scripts. Stale results from earlier in the session don't count if the code changed since.
- Read the output, don't pattern-match it. Exit code 0 with skipped tests is not a pass. A green suite that doesn't exercise the changed behavior is not evidence for the claim.
- Match evidence to claim. Bug fix → the original repro no longer fails and a regression test passes. Feature → the user-visible behavior was exercised, not just unit internals. Refactor → the pre-existing suite passes unchanged.
- Report command + outcome verbatim. State exactly what ran and what it showed. If something was skipped, say so.
When verification cannot run
Environment broken, credentials missing, suite requires services that are down:
- Say the claim is unverified and why.
- State the exact command someone must run to complete verification.
- Do not soften to "should work" — the status is unverified, not probably fine.
Common self-deceptions (reject all of these)
| Deception | Reality |
|---|---|
| "The change is simple, no need to run it" | Simple changes break builds daily |
| "Tests passed before my last small edit" | The last edit is the one that breaks |
| "The type checker is happy" | Types don't verify behavior |
| "I ran the unit tests" (for a user-visible bug) | The user's flow was never exercised |
| "Lint + build succeeded" | Neither executes the changed code path |
| "The logs look like it worked" | Find the assertion, not the vibe |
Verification
- Claim stated precisely
- Verifying command executed after the final code change
- Output actually read; skips/warnings accounted for
- Evidence type matches claim type (repro for bugs, behavior for features, suite for refactors)
- Report includes command + observed result, or an explicit UNVERIFIED status
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most quality gates skills give in 542 tokens
Counted across 1,524 of the 2,830 authors here whose files we hold, read 2026-09-06
- Read full output and check exit codein 45 of 1524, across 40 files
- Verify output confirms the claimin 44 of 1524, across 39 files
- Identify the command that proves the claimin 43 of 1524, across 39 files
- Execute the full verification commandin 36 of 1524, across 30 files
- Produce a verification reportin 34 of 1524, across 18 files
- Review git diff changesin 30 of 1524, across 16 files
- Fix build failures immediatelyin 29 of 1524, across 9 files
- Group findings by severityin 28 of 1524
- State claim only with evidencein 27 of 1524, across 22 files
- Verify regression tests with red-green cyclein 26 of 1524, across 22 files
- Run the full test suitein 26 of 1524, across 25 files
- Run test suite with coveragein 25 of 1524, across 10 files
Said here and by no other author read
- Name the completion claim precisely
- Report command and outcome verbatim
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.