Skip to content
Back to skills

Jt Verify

ASecurity

Runs the changed feature the way a user would and looks at the real result, so that passing checks are not mistaken for working software. Use when jt-work reaches the See-it-work step, or when the user asks to confirm that a change actually works in the running app, API, or command-line tool.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 18, 2026
ai-agentsgoapidatabase

Works with

  • api

Security analysis

A100/100

Scanned November 18, 2026

npx -y skills add thitinan147/jtek-skill --skill jt-verify --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Jt Verify?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Jt Verify
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/thitinan147-jt-verify/badge)](https://www.skillsdirectory.com/skills/thitinan147-jt-verify)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: jt-verify
description: Runs the changed feature the way a user would and looks at the real result, so that passing checks are not mistaken for working software. Use when jt-work reaches the See-it-work step, or when the user asks to confirm that a change actually works in the running app, API, or command-line tool.
---

# jt-verify

Passing checks do not replace this step. Do it after `jtek.py check` passes.

## 1. Run the real thing

If the repository has `.jtek/verify.md`, follow it to launch and drive the app, and read the file in `.jtek/features/` for the feature you changed. If it has none, use the list below.

For each `Done when` item, do what you named before you built it:

- **Command-line tool or library:** run it on real input, including the exact example in the ticket.
- **API:** start the service and call the endpoint. Read the response body and the status.
- **Screen:** open it in a browser and do the steps. Read the console and the failed network requests.

## 2. Look at the result

1. Compare what you see with the `Done when` item, word for word.
2. Try the edges the ticket implies: empty input, very long input, the same action twice, the wrong order.
3. For a screen, look at each state: empty, loading, error, and text that is too long.
4. Check that behaviour under `Out of scope` is the same as on the base branch.
5. For a change to a database schema or to a response other code reads: apply the migration, undo it, and apply it again. Then run the base branch's code against the new schema or response.

## 3. What does not count

- The code compiles, or the checks pass.
- A file exists or changed.
- Output you saved earlier, before the last commit.
- Your own summary of what the code does.
- A different path than the one the ticket names. A skipped entry point is not verified through another one.
- An unclear result. Inconclusive is not a pass. Say it is inconclusive.

Use the real user path, not an internal function or a test-only endpoint. Capture the action and the state it produced, not only the last screen. When the change writes data, look at the stored value a second way.

## 4. Say only what you saw

- Write down the command or steps and the result for each item. Keep the screenshot, recording, or output file that shows it. This is the evidence. It goes into the pull request, or into the evidence file when there is no pull request.
- Never write "should work" or "probably fixed". Either you saw it, or you did not.
- If you could not run it, say that in the evidence, and say why.

## 5. When it does not work

The change is not done. Find the cause, fix it, commit, and run `jtek.py check <id>` again. Then repeat this skill.

If the project has no way to run the real thing, say so in the evidence and say that the repository needs a verification harness.

If you found a new way to drive a feature, or a gotcha, add it to that feature's file in `.jtek/features/` in the same change.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.