Skip to main content

verification-skill

Use when an app lacks repeatable behavioral verification or its project-local verification skill needs an audit.

Source facts

Repository
anntnzrb/agents
Last source activity
October 4, 2026 at 16:52
Detected SKILL.md language
English
Stars
1
Forks
0

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer
8 files

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
name
verification-skill
description
Use when an app lacks repeatable behavioral verification or its project-local verification skill needs an audit.
license
AGPL-3.0-or-later
# Verification Skill A verification skill is a project-local skill that lets an agent launch the real app, drive it the way a user does, and capture evidence. It is the loop that lets an agent prove its own work instead of waiting for a human to check it. This skill generates one and keeps it honest. Ported from pstack `create-verification-skill` and `maintain-verification-skill` (Lauren Tan, MIT). See [NOTICE.md](NOTICE.md). ## Pick a mode | Mode | Use when | Read | | --- | --- | --- | | `create` | The project has no scripted way to prove UI, CLI, or service behavior | [references/create.md](references/create.md) | | `maintain` | A `verify-<app>` skill exists and the app has changed, or the user asks to audit it | [references/maintain.md](references/maintain.md) | Read the mode file in full before acting. Applying a mode from this table alone is not allowed. ## Where the generated skill lives - Write the skill to `<repo>/.agents/skills/verify-<app>/`. Codex, pi, and omp discover project skills there. - Claude Code discovers project skills in `.claude/skills/`. Link it instead of copying: `ln -s ../../.agents/skills/verify-<app> .claude/skills/verify-<app>`. One copy per harness drifts. - The generated skill belongs to the app repository. Never put it in the shared skills SSOT. - Helper scripts use the app's own stack and toolchain. The shared-skill `scripts/cli.py` rule does not apply to them. ## Non-negotiables - A generated skill that was never executed end to end is a draft, not a deliverable. - Evidence survives every cleanup. Confirm it at its named location after teardown. - Drive only instances this run started. Never kill by process name. - Never edit product code in `maintain` mode. A broken behavior is a product gap to report, not doc drift to paper over. ## Required follow-up reads | Need | Read | When | | --- | --- | --- | | Generation steps and required sections | [references/create.md](references/create.md) | `create` mode | | Audit pass and outcomes | [references/maintain.md](references/maintain.md) | `maintain` mode | | Feature map shape | [references/feature-map-example/README.md](references/feature-map-example/README.md) | Seeding or auditing a feature map |
View on GitHub