| name | verification-before-completion |
| description | Use an explicit claim-to-evidence gate for High-risk work, disputed or unreliable claims, or multiple independent material claims; routine command sets alone do not trigger it. |
Verification Gate
Router-, user-, or repository-required gates cannot be skipped. Otherwise use ordinary scoped checks when a separate gate adds no discipline. Multiple routine commands alone do not trigger this skill.
Reuse fresh evidence from active workflows instead of rerunning commands; map it to the claims it proves.
For destructive operations, run a go/no-go check before execution and verify outcomes afterward.
For production deployment, verify the target and revision, rollout health, rollback readiness, and runtime evidence; a Git push is not deployment evidence. For production-data changes, verify target identity, dry-run scope, recovery readiness, integrity, and post-operation counts without exposing sensitive records.
- List each material claim and the command or inspection that proves it.
- After the final relevant change, run the narrowest sufficient checks, batching compatible checks.
- Inspect exit status, failures, and meaningful warnings; do not infer success from partial output.
- Compare the actual diff and final state with acceptance.
- Report only what the evidence proves, including unavailable or failing checks.
Any later relevant mutation invalidates affected evidence only; rerun the corresponding checks.
Broaden checks only when repository policy or the named risk requires it. Preserve user work and distinguish verified outcomes from assumptions or pre-existing failures.