Use when preparing to claim success, hand work off, commit, merge, open a PR, publish, deploy, or close a task, including in headless and SDK-driven sessions.
Scanned 8/31/2026
Install to Claude Code
npx -y skills add lawzava/megapowers --skill verify-and-finish --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Verify And Finish?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/lawzava-verify-and-finish)More formats (shields.io, HTML) on the badges page.
---
name: verify-and-finish
description: Use when preparing to claim success, hand work off, commit, merge, open a PR, publish, deploy, or close a task, including in headless and SDK-driven sessions.
---
# Verify and Finish
Evidence precedes every load-bearing claim. For each acceptance criterion,
identify the oracle, run it fresh against the current artifact, read the whole
result, and state only what it proves. A focused test does not prove the full
suite; artifact inspection does not prove runtime behavior; confidence proves
nothing.
Name the load-bearing safety fact behind an impact claim and the proof level it
needs. Separate declared configuration from effective runtime behavior. When a
result can be stale, bind it to the exact artifact identity and commit SHA.
Keep these evidence classes explicit:
- Executed check: behavior observed now in the named environment.
- Artifact inspection: what a file or configuration declares.
- Inference: a hypothesis that still needs a check.
Separate local verification from external proof. A local build cannot prove a
deployment, published package, API effect, or user-visible result. When the
criterion requires an external oracle, exercise the real boundary and record
the environment and correlation identity. If a required tool or environment is
unavailable, report the criterion as unverified rather than substituting a
nearby check.
For a user-facing product, run the real user journey. Use an agreed substitute
oracle only when the real journey is unavailable, and name what it cannot prove.
Before completion, reconcile each affected repository-owned behavior
specification with verified behavior. Do not create a durable specification
without repository convention or user approval.
Before a handoff, commit, PR, merge, release, or cleanup, run the repository's
canonical checks and inspect the current diff and workspace state. Sweep the
diff for generated excess: provenance comments, redundant code, churn. When the
task names a target branch, confirm the checked-out branch matches that named
target before the commit; on a mismatch, stop and resolve it first. Do not
commit, publish, merge, delete a branch, or remove a worktree unless the user
authorized that destination. Destructive cleanup requires explicit target
confirmation and ownership evidence.
Open the report with `VERIFIED: <claim>` only when every required criterion has
fresh oracle evidence. Otherwise use `NOT VERIFIED. Remaining: <gap>`.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!