Skip to content
Testing

Verify Before Completion

Connect completion claims to fresh checks that actually establish the claimed result.

Install with Skills CLI
npx skills add obra/superpowers --skill verification-before-completion

What it does

An agent can report success after editing code without demonstrating that the intended behavior works. This skill makes the relationship between a claim, a verification command, and the observed result explicit.

What changes after install

  • Identifies the check needed to substantiate each completion claim.
  • Requires inspecting current output and exit status rather than relying on earlier runs.
  • Separates build, test, requirement, and runtime evidence.

Install

Run this command with Node.js 22.20 or later and select your agent in the Skills CLI.

Install with Skills CLI
npx skills add obra/superpowers --skill verification-before-completion

Before using Verify Before Completion

  • Identify a command that verifies the specific completion claim and ensure the project’s environment can run it. Capture fresh output and its exit status before reporting success.

Read the reviewed setup instructions for the selected workflow.

When to use it

  • Checking a coding task before handoff or commit.
  • Reviewing an agent's claim that a fix, build, or test suite is complete.

When not to use it

  • Treating a single green check as proof of unrelated requirements.

Match the check to the claim

If an agent says “the CSV import is fixed,” ask what established that result. A type check may prove that the code compiles, but it does not show that a quoted delimiter is parsed correctly. A test of one parser case may prove that case, but not that the upload screen handles the resulting error. This skill encourages distinguishing those claims instead of treating any green command as universal evidence.

The reviewed instructions require identifying a suitable check, running it, reading the complete outcome, and stating only what the result supports. They also distinguish fresh output from results produced before the latest changes.

Prepare an evidence-oriented handoff

For a feature, collect the relevant tests, the production build result, and any manual behavior checks required by the task. For a bug, include the original reproduction and the regression result. For an external dependency that could not be reached, report that limitation directly.

The workflow relies on commands appropriate to the repository. It does not prescribe one test runner or install missing environments. A failed or unavailable check is still useful information and should not disappear from the final report.

Keep the scope honest

Use Systematic Debugging to find the cause of a failure and Test-Driven Development to protect a known behavior. When the change is ready for review, give Request Code Review the commit range, the checks you ran and any behavior that still needs a reviewer’s attention.

SkillDex reviewed the upstream instruction file. It has not tested whether installing it changes a particular model’s reliability, and no such effectiveness claim is implied by this listing.

FAQ

Is a successful build enough to call a bug fixed?
Only if the build itself reproduces the bug. Otherwise, verify the original failing behavior and relevant regressions.
Does this skill execute tests automatically?
It supplies an evidence workflow. The agent still needs the project's commands and an environment capable of running them.