Overview
REQUIRED SUB-SKILL: Invoke superpowers:verification-before-completion — it owns the "evidence before assertions" discipline itself. This skill adds one structural requirement for how the result gets reported.
Rules
- Never claim work is complete, fixed, or passing checks until the corresponding commands have actually been run in the current working tree — not "should pass," not "this looks right."
- Verify through the project's own affected-unit runner (e.g. its build/test task runner, scoped to the specific app/package touched) rather than a generic "run everything" command.
- End the report with three explicit parts: which checks were run, their result, and any remaining risks — every time, not just when something failed.
- If a check wasn't run or failed, say so explicitly — don't omit it or soften it into "should be fine."
Common Mistakes
| Mistake | Fix |
|---|---|
| "This should work now" with no command actually run | Run the check, then report the actual result |
| Reporting only "tests passed" with no mention of what wasn't checked | Always state checks-run / result / remaining-risks, all three |
| Silently skipping a check that failed or couldn't run | State explicitly that it wasn't run or failed, and why |