create-verification-skill
Generate a project-local skill that drives the real app the way a user does and proves it with saved evidence: any language, framework, or platform. Use for /create-verification-skill, "build a verify skill for this repo", or when a project has no scripted way to prove UI/CLI/service behavior beyond "it compiles".
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- create-verification-skill
- description
- Generate a project-local skill that drives the real app the way a user does and proves it with saved evidence: any language, framework, or platform. Use for /create-verification-skill, "build a verify skill for this repo", or when a project has no scripted way to prove UI/CLI/service behavior beyond "it compiles".
- argument-hint
- [app-or-surface name, optional -- defaults to the repo's primary user-facing surface]
- disable-model-invocation
- true
02 资源文件 5
- 许可证LICENSE1.0 KB
- 参考文档references/feature-map-example/create-note.md2.9 KB
- 参考文档references/feature-map-example/README.md2.8 KB
- 参考文档references/feature-map-example/search.md3.4 KB
- 参考文档UPSTREAM.md16.8 KB
03 SKILL.md 正文
Create a verification skill
Every serious project needs a scripted way to drive the real app and prove behavior: launch it, exercise a feature the way a user would, and capture evidence. This skill generates that as a project-local skill (<repo>/.claude/skills/verify-<app>/) tailored to the repo. Write the generator's output for the next agent, not for a human: it will be read cold, mid-task, by an agent that has never seen the app.
1. Interview the repo, not the user
Answer these from the codebase and only ask the user what you cannot observe:
- Surface: what does a user actually touch? A web UI, a CLI/TUI, a desktop app, an API, a mobile app, a library? A repo can have several; pick the primary one (or the one named by this skill's argument) and note the rest.
- Run: how does the app start locally? Prefer the repo's own documented dev command (package scripts, Makefile, README quickstart). Note ports, env vars, seed data, auth.
- Drive: how can an agent interact with it programmatically? Check the installed skill roster for a skill that already covers this repo's stack (for example,
~/.claude/skills/e2e-testing-for-webappsfor a Next.js, Playwright, and Supabase repo) and reuse its infrastructure rather than deriving a parallel harness. Otherwise, choose the harness in this order: an existing harness already in the repo wins (Playwright/Cypress specs, expect scripts, PTY helpers, curl-able endpoints, a debug port); then the driverce-test-browserselects for this host, per itsSKILL.mddriver-selection lines; headless Playwright only when neither exists, and never installed alongside an existing browser stack; wsl-cdp only when a feature needs the operator's already-authenticated browser, and then read-only inspection unless the operator confirms a mutating drive in that session. A tmux/PTY harness drives CLI/TUI; plain HTTP drives services. The harness choice is made per repo, and the generated skill states which one it uses and why. - Observe: what evidence can be captured? Screenshots, DOM/ARIA text, terminal transcripts, response bodies, logs, exit codes, DB state.
- Isolate: can two instances run side by side (ports, data dirs, profiles)? If not, say so in the generated skill: refusing to double-drive a shared instance beats corrupting the user's session.
If the checkout doesn't build or start as-is, fix that first (or report it precisely) before generating; a skill written against a broken base teaches wrong steps. When an irrelevant missing asset blocks startup (a static dir the API never serves, a sample config), the generated skill may create it, clearly marked as verification scaffolding, and remove it in cleanup. A missing asset the project's own build step would normally generate, such as a gitignored generated file or a compiled worker stub, is a build prerequisite rather than verification scaffolding: create the same placeholder the build produces and leave it in place. Only scaffolding invented for verification is removed in cleanup.
2. Generate the skill
Check the installed roster before choosing verify-<app>; a project skill that shadows an installed one, for example verify-review, gets a different app name.
Write <repo>/.claude/skills/verify-<app>/SKILL.md with YAML frontmatter (name: verify-<app>, disable-model-invocation: true because it starts and kills real processes and must run only when named, and a description that names the app, the surface, and when to reach for it -- without frontmatter the skill never registers) and these sections, each grounded in what the interview actually found (no placeholders left):
- Launch: the exact command that starts the app for verification, and how to tell it's ready (a log line, a prompt). Launch must reach a usable state from an empty data dir, including any migrations and seed data. Ready means usable, not just that a process answers. A database health check can pass while its init scripts are still running. Include teardown. For a short-lived CLI or TUI there is no server to keep alive: launch means build the binary (or install deps) once, then start each drive in its own isolated PTY or tmux session.
- Doctor: one read-only check that answers "is this instance worth driving?": process up, right version/build, port owned by us, auth valid. Never drive an instance the verification run did not start. The doctor check must prove that (a port owned by us, a build/version stamp, a scratch data or config directory named for this run), not just that something is listening. An agent runs this first whenever anything looks off.
- Drive: the harness recipe with real selectors/commands from this repo, not examples. Prefer stable handles (ARIA labels, data attributes, prompt strings, route paths) over coordinates and tab order.
- Evidence: what to capture for a proof and where it goes. Save every artifact behind a pass claim (a screenshot, a DOM/ARIA text dump, or an exit-code log) to a fresh
/tmp/verify-<app>-<run-id>/directory, one file per pass claim, surviving cleanup there, so a pass claim always names a path a later reader can open. When a run's proof is meant for the operator, write a short summary plus key artifacts to<repo>/.claude/deliverables/verify-<app>/<YYYY-MM-DD-HHMM>-<slug>/, runopen-deliverable <path>, and report the path. State the proof standards: exercise the real user path, not internal setters or test-only endpoints; capture the action and the resulting state, not just the final screen; verify side effects (files written, rows inserted, messages sent) alongside what's visible; mocks only where a production boundary already isolates the external system. When the safe path is a dry-run or test mode, verify what it actually skips by observing (files, network, git refs) rather than trusting its name: some dry-runs still touch the network or open a browser. If the app cannot start for a reason outside the checkout (a missing credential, an unavailable service, production-only data), stop and reportenvironment blocked: <what>: a distinct outcome from a mechanism failure in the generated skill itself, and in the harness outcome taxonomy this isNO_EVIDENCE: the gate could not be evaluated (harness/evidence-regime.mdsection 5). Never guess at a secret value to get past a blocked environment. - Cleanup: how to tear down instances the run created. Never kill by process name; kill what you started. Cleanup removes instances and scratch state, never the evidence: proof artifacts under
/tmp/verify-<app>-<run-id>/, and anything already copied to<repo>/.claude/deliverables/verify-<app>/for the operator, survive the teardown. A cleanup that eats the proof fails this step. - Helpers: any script the skill ships is executable and its invocation is shown in the skill body. A helper the reader has to reverse-engineer is not a helper.
3. Seed the feature map
Create <repo>/.claude/skills/verify-<app>/features/README.md plus one file per user-facing feature you can identify (aim for the top 3-5 to start, from routes, commands, menus, or docs). Follow the shape in references/feature-map-example/, with a README index and one file per feature. Each file answers, from the user's point of view: what the feature is, how to reach it, how to drive it with the harness, and what observable end state proves it works. The four H2s are Sub-features, How to get to it from the user's point of view, Driving it with the harness, and Gotchas. If the seed data lacks a precondition that a journey needs, the journey reports a failure and the reason goes in that feature's Gotchas. Never weaken the journey to make it pass. The map is the repo's maintained verification source; a proof that drives one convenient entry point is incomplete when the map lists others. Mark every state assumption in a feature file the proof run did not drive with an inline unverified: prefix, so a later run confirms it before trusting it.
4. Prove the generated skill before handing it over
Run its own instructions end to end once: launch, doctor, drive ONE mapped feature (one is enough; the map exists so later runs can cover the rest), capture evidence under a fresh /tmp/verify-<app>-<run-id>/ directory (and, when it's meant for the operator, copy a summary plus key artifacts to <repo>/.claude/deliverables/verify-<app>/<YYYY-MM-DD-HHMM>-<slug>/ and run open-deliverable <path>), clean up. After cleanup, confirm the evidence still exists at the named location: a cleanup that eats the proof fails this step. Fix what fails, and run the generated cleanup after every failed iteration too, so broken attempts don't strand processes and ports. Never drive an instance the verification run did not start. If the app cannot start for a reason outside the checkout, report environment blocked: <what> instead of forcing a run through it; that is not the same failure as a mechanism bug in the generated skill, and in the harness outcome taxonomy it is NO_EVIDENCE: the gate could not be evaluated (harness/evidence-regime.md section 5). The agent that wrote the skill also proved it, so this one proof run is self-certification: record which feature was proven, and leave every other feature file marked unverified: until a fresh session drives it. A generated skill that was never executed is a draft, not a deliverable. If you proved it in a stand-in for the real target (a container in place of a fresh VM, a local stack in place of a remote one), list every setup step the stand-in did that the target will not. Those steps stay unverified until a run on the real target covers them.
5. Offer the maintenance loop
Point the user at /maintain-verification-skill for keeping the map honest as the app changes. Suggest a cadence only if they ask.
04 同仓库更多 Skills
review-adversarial-first
Operator-invoked adversarial review: 3 refute-by-default lenses plus a deterministic judge against a PR, branch, or path, then a decision on whether /ce-code-review's topic fan-out runs. Off the default merge path since the 2026-10-17 merge-and-end ruling; run only when asked for by name.
memory-mine
Run one deterministic, budgeted directed-capture batch for an explicit Claude Code project store. Use manual workspace mode only when cross-agent cass sessions are intentionally in scope.
papers
Search preprints from arXiv, bioRxiv, and medRxiv without API keys. Supports search, lookup, extraction, citations, enrichment, and paper downloads that bundle PDFs plus linked GitHub code.
papers
Search preprints from arXiv, bioRxiv, and medRxiv without API keys. Supports search, lookup, extraction, citations, enrichment, and paper downloads that bundle PDFs plus linked GitHub code.
wsl-cdp
Drive the user's real, logged-in Windows browser (Chrome, Edge, or Brave) from a Claude Code session running inside WSL2, over the Chrome DevTools Protocol - read authenticated dashboards and consoles, take real screenshots, open and read pages with the user's sessions. Use when a task needs the user's existing web logins and the session runs in WSL2, where the Claude in Chrome extension cannot work. Not for native-Windows or macOS Claude Code (use the official Claude in Chrome extension), not for anonymous scraping or CI (use headless Playwright inside WSL), not for getting past login walls (those are handed to the human).
token-eater
[BETA] Spend a subscription's idle, about-to-expire credits on safe, machine-verified cleanup. You pick the service (Grok / Codex / Claude) and a token-heavy maintenance skill (de-monolithize, de-slop, simplify, dead-code); that service runs the whole skill -> gate -> self-review -> draft-PR loop on its own credits, in an isolated worktree. token-eater verifies the gate and never merges. Manual invocation only during beta.