verification-before-completion · Verification Before Completion
A pre-delivery verification discipline with one iron law: evidence before assertions — before claiming tests pass, a bug is fixed, or a build succeeds, run the verification command and read the full output first.
It ships a claim-to-evidence mapping (what each common claim requires: tests, lint, build, fix, regression, requirements), plus a list of red-flag phrasings ("should", "probably", "looks right") and rationalization traps. Use it before committing, opening a PR, or declaring a task done.
Example invocation: "I've made the changes — verify everything before we call it done."
Full brief
Positioning
verification-before-completion is a gate that enforces verification before any claim of completion, fix, or passing: skipping any step of the gate counts as not verifying.
Core capabilities
- The gate: identify the command that proves the claim → run it fully → read the full output, check the exit code, count failures → claim only if the output supports it.
- Claim-to-evidence mapping: tests pass needs a test run with 0 failures; linter clean needs linter output; build success needs exit code 0; bug fixed needs the original symptom re-tested; regression tests need a verified red-green cycle; requirements met needs a line-by-line checklist.
- Red-flag detection: bans "should", "probably", "seems right"; verification comes before expressing satisfaction, committing, PRs, and handoffs; never trusts an agent's success report — re-verify independently; partial checks prove nothing.
- Rationalization prevention: an excuse-vs-reality table — "I'm confident" ≠ evidence, "just this once" = no exceptions, "linter passed" ≠ compiled, "the agent said success" = verify independently.
Workflow
- Identify: what command proves this claim
- Run: execute the verification command fully (fresh, complete)
- Read: full output, exit code, failure count
- Verify: does the output support the claim — if not, state the actual status
- Claim: only with evidence attached
Inputs & outputs
| Input | Notes |
|---|---|
| The claim to be made | Required — e.g. "tests pass", "bug fixed" |
| Verification command | Optional — the skill can derive it |
Output: an evidence-backed conclusion, or an honest status report.
Boundaries with adjacent skills
- systematic-debugging: the investigation process when something is broken (root cause before fixes).
- verification-before-completion (this): the gate before claiming completion (evidence before assertions). systematic-debugging's fix phase calls on this skill to confirm the fix works.
Fit
- Before committing code or opening a PR
- Before claiming a bug is fixed or requirements are met
- Verifying success reports from delegated subagents
- Before task handoffs or moving to the next phase
Before you start
- No special setup. The whole point is discipline: run the command before any success statement — rephrasing doesn't change the rule.
verification-before-completion is part of the Aiglade Skill library. Invoke it from the Aiglade chat box in plain language.