mohammadhprp/teammate/verify-change
Claude Code and agent skill by mohammadhprp/teammate
- community
Checking that the SKILL.md file is still in the repository.
Self-verify your own change before handing it off: run the project's own checks, exercise the changed behavior at its boundaries, confirm no regression in touched areas, and capture the raw command output as evidence. Use whenever a worker believes its build is complete but has not yet reported it, and for any change about to leave your hands — never hand off a known failure, and treat an unexercised green run as a process failure, not proof.
Install
npx vskill@latest install mohammadhprp/teammate/verify-changeSKILL.md
The same SKILL.md was published earlier as mohammadhprp/teammate/verify-change.
Verify change
A change is ready to hand off only when the project's own checks say so. This is the worker-side application of the shared verify-evidence standard: that skill defines what counts as evidence, and this one turns the standard into the self-check to run before report-result carries the change to the primary.
Verify here because it is the cheapest place to catch a defect: every check skipped now returns later as a review-change finding and a rework cycle.
When to use
- You believe the build is complete and are about to hand it off (see
report-result).
- The change has boundaries a happy-path run never reaches — empty, null, error,
boundary, or concurrent inputs.
- You changed code that other behavior depends on and need to know you did not
break it.
When not to use
- Never skipped for a change that claims to work. If you cannot verify it, say
so; do not report it as done.
- Mid-build, the checks belong to
implement-taskas you go; this is the
deliberate final pass before reporting.
Inputs
- The change: the diff and the files it touches.
- The claim it must satisfy: the acceptance criteria.
- The project's own check commands, from
load-project-context.
Procedure
- Run the project's own checks. Use the test, lint, build, and typecheck
commands the project defines in its AGENTS.md / CONTEXT.md. The project's command is the authority; an ad-hoc substitute only counts when the project has none, and then say so.
- Exercise the changed behavior at its boundaries. Drive the inputs the
happy path avoids — empty, null, boundary, error, and concurrent cases. The defect usually lives where the happy path does not go. Reuse the project's test framework where it fits, or write a focused reproduction.
- Confirm no regression in touched areas. Run the checks nearest the
change, not only the ones you added; a change to shared code is verified by its consumers' checks too.
- Capture the raw output. Record the exact command and its actual result,
not a paraphrase of intent. Paste what the command really printed; do not reconstruct a transcript from memory or expectation — an unrun transcript is fabricated evidence, and review will catch it. Evidence from before your last edit does not count; re-run after every change.
- Confirm nothing you started is still running. Stop every server, watcher,
or long-lived process you launched for the check, or record its PID and port in the report when it must stay up. A stray listener is a failed handoff, not a harmless leftover.
- Separate proven from assumed. Name what you did not check.
verify-evidence's rule applies to your own run: a suite that never touched the change is a pass with zero checks, not proof.
Output
A claim + command + result set ready to become the Verified section of the handoff-report — plus the list of what was not checked.
Failure and escalation
- A check fails → fix it. If it is out of scope or you cannot fix it, hand it to
raise-blocker; never report a known failure as done.
- A check cannot run → say so and mark the claim
inconclusivein the report
rather than assuming a pass.
- The change works only under conditions you had to invent → that is a gap in
the brief; raise it instead of inventing acceptance criteria.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingsrc/skills/verify-change/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/mohammadhprp/teammate/verify-change)