Verification Before Completion Skill

Verification Before Completion is a skill from Superpowers (skill source). It activates whenever an agent is about to claim that work is complete, fixed or passing, before committing or opening a pull request. Its rule is short: evidence before claims, always.

The iron law: "No completion claims without fresh verification evidence." If the agent has not run the command in this message, it cannot say it passes.

Key Features

• The gate function: identify the command that proves the claim, run it fresh and in full, read the output and exit code, and only then state the result with the evidence.
• Claim table: "tests pass" needs a test run with 0 failures; "build succeeds" needs the build to exit 0, and a passing linter is not enough; "bug fixed" needs the original symptom re-tested.
• Regression proof: write the test, run it, revert the fix and see it fail, restore and see it pass.
• Delegation rule: when a subagent reports success, check the VCS diff yourself before reporting.
• Red-flag words: "should", "probably", "seems to", and premature "Great!" or "Done!" all trigger a stop.

Use Cases

• The last step before any commit, pull request or "task complete" message.
• Reviewing work returned by another agent.
• Long autonomous sessions where optimistic summaries creep in.

Pricing

Free and open source under the MIT license. Prime Radiant, the company behind Superpowers, sells commercial support to enterprises. The practical cost is tokens: process skills add questions, reviews and subagent runs.

Getting Started

Superpowers installs as one plugin, so this skill arrives with the rest of the library (15 skills in v6.4.1). In Claude Code run /plugin install superpowers@claude-plugins-official. Cursor uses /add-plugin superpowers, Gemini CLI uses gemini extensions install https://github.com/obra/superpowers, and the README lists commands for Codex, GitHub Copilot CLI, OpenCode, Pi and other harnesses. Skills trigger on their own; you can also ask for one by name. You will notice it in the agent's final messages: results arrive with the command and the counts that prove them.

Limitation: running full test suites and builds before every claim takes time on large projects, and the skill cannot verify things with no runnable check, such as visual design.

FAQ

Is this different from TDD?

Yes. TDD governs how code is written; this governs what the agent may claim afterward.

Does it apply to partial progress reports?

Yes. Any wording that implies success counts.

Alternatives

• Unlazy: acceptance gates as runnable checks.
• Claude Hooks: enforce checks deterministically with shell hooks instead of instructions.
• Requesting Code Review: a second pair of eyes after verification.

Conclusion

A small skill that fixes the most common agent failure: saying "done" when it is not. More in the skills hub.