[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EBADWDANNWGR1XGT0CA934":3,"skill-same:01M7EBADWDANNWGR1XGT0CA934":52},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":28,"repositoryHtmlUrl":29,"repositoryHomepageUrl":6,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":31,"updatedAt":32,"install":33,"resources":45},"01M7EBADWDANNWGR1XGT0CA934","run-skill-generator",null,"Author or improve the run-\u003Cunit> skill - a per-project skill that tells agents how to build, launch, and drive this project's app. Use when the user asks to set up the project, get it running, write run instructions, or verify build\u002Frun steps work from a clean environment.","lza6\u002FArmor-piercing-system-prompts","lza6","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F128880206?v=4","01M7EB3K7WPX5T28EJ2NVHE6EG","Chinese-tingfeng","01M7EB3K97E1KQ16QWKJMFNGX8",11,"MIT","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F02-厂商系统提示词\u002Fsystem_prompts_leaks\u002FAnthropic\u002Fclaude-code\u002Fskills\u002Frun-skill-generator\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: run-skill-generator\ndescription: Author or improve the run-\u003Cunit> skill - a per-project skill that tells agents how to build, launch, and drive this project's app. Use when the user asks to set up the project, get it running, write run instructions, or verify build\u002Frun steps work from a clean environment.\ndisable-model-invocation: true\n---\n\nYour job is to produce a **skill** at `\u003Cunit>\u002F.claude\u002Fskills\u002Frun-\u003Cunit-name>\u002F`\nthat lets a future agent build, launch, and **drive** this project from\na clean machine.\n\nThe skill has two parts that live together:\n\n```\n\u003Cunit>\u002F.claude\u002Fskills\u002Frun-\u003Cunit-name>\u002F\n  SKILL.md      \u003C- agent-facing instructions - SHORT. Points at the driver.\n  driver.mjs    \u003C- (or driver.py, smoke.sh, ... - or none: web apps use\n                   chromium-cli off-the-shelf, and the heredoc in\n                   SKILL.md is the script)\n```\n\nThat almost always means **writing code**, not just prose. If the app\nhas any interactive surface (GUI, TUI, long-running server, REPL), the\nfuture agent needs a programmatic way to poke it. A markdown file by\nitself cannot click a button - but sometimes the button-clicker\nalready exists: for web apps it's `chromium-cli`, for servers it's\n`curl`. You build (or script) that harness now, commit it alongside\nthe skill, and the `SKILL.md` documents how to use it.\n\n## Definition of done\n\nYou are done when **all** of these are true:\n\n1. **You launched the app in this container and interacted with it** -\n   not its test suite, the actual running app. For anything with a GUI,\n   that means you have a screenshot file on disk that you took.\n2. **The interaction harness is committed** next to the skill. A driver\n   script, a REPL wrapper, a smoke test, or the `chromium-cli` heredoc\n   inline in `SKILL.md` - whatever you used to drive the app in step 1.\n   (Graduated into `scripts\u002F`\u002F`e2e\u002F`? - fine, point at it. Web app with\n   `chromium-cli` off-the-shelf? - the inline script is the harness; no\n   separate file.)\n3. **The `SKILL.md` documents the harness** as the primary agent path -\n   the section a future agent reads first is \"run this driver \u002F pipe\n   these commands to `chromium-cli`,\" not \"run `npm start` and a window\n   opens.\"\n4. **Every code block in `SKILL.md` is a command you ran that worked.**\n   This session. This container. Not from the README, not inferred.\n\nIf you're about to write the skill and you don't have (1), **stop.** You\nare about to paraphrase existing docs. That document already exists -\nit's called the README, and the whole reason you're here is that it\nwasn't enough.\n\n## The deliverables are code AND docs\n\nTypical output is a skill directory containing both:\n\n```\n\u003Cunit>\u002F.claude\u002Fskills\u002Frun-\u003Cunit>\u002F\n  SKILL.md         \u003C- SHORT. Points at the driver. Has the frontmatter\n                     that lets Claude auto-load it when someone asks\n                     to \"run \u003Cunit>\" or \"screenshot \u003Cunit>\".\n  driver.mjs       \u003C- (or driver.py, smoke.sh, ... - or none: web apps\n                     use chromium-cli off-the-shelf, and the heredoc\n                     in SKILL.md is the script)\n```\n\nThe driver lives **inside the skill directory** by default. They are a\npair - the skill's instructions and the code that implements them. A\ndriver that lives here is allowed to be a bit messier than production\ncode; it's agent tooling, not product surface.\n\n**Graduation:** if the driver grows into something the project's own\ntest suite wants to reuse - shared launch helpers, a real e2e harness -\nmove it to `scripts\u002F` or `e2e\u002F` and update `SKILL.md` to reference the\nnew path. The skill stays; the driver finds a better home.\n\nThe exact shape depends on the project, but the principle is constant:\n**the driver is the deliverable.** The `SKILL.md` is its man page. For\na web app, the driver already exists - `chromium-cli`\n([examples\u002Fplaywright.md](examples\u002Fplaywright.md)) - and the skill is\nthe script that runs it. For a desktop app\n([examples\u002Felectron.md](examples\u002Felectron.md)), the driver is a custom\nREPL under tmux that exposes `launch`\u002F`ss`\u002F`click`\u002F`eval`. For a server,\nthe driver is `curl`. Whatever shape it takes, without something that\nreaches into the running app, the skill is a description of a window\nnobody can touch.\n\n## Where the skill goes\n\nThe skill lives at `\u003Cunit>\u002F.claude\u002Fskills\u002Frun-\u003Cunit-name>\u002F`, where\n`\u003Cunit>` is the directory for **one deployable thing** - an app, a\nservice, a library.\n\nClaude Code **natively discovers** skills from nested `.claude\u002Fskills\u002F`\ndirectories: an agent working anywhere inside `\u003Cunit>` will see\n`\u002Frun-\u003Cunit-name>` as an available skill, and it auto-loads when the\nrequest matches its description (e.g. \"run the desktop app,\" \"take a\nscreenshot of billing\").\n\n- **Single-project repo:** `.claude\u002Fskills\u002Frun-\u003Crepo-name>\u002F` at repo root.\n- **Large repo with many apps:** one per app, colocated -\n  `apps\u002Fbilling\u002F.claude\u002Fskills\u002Frun-billing\u002F`,\n  `apps\u002Fdesktop\u002F.claude\u002Fskills\u002Frun-desktop\u002F`.\n- **App with multiple binaries:** still **one** skill at the app's\n  root with a section per binary. They share setup. Start from the\n  closest single-binary example and add a `## Run: \u003Cname>` section\n  per binary.\n\nIf you're not sure where the unit boundary is, **ask the user.**\n\nSlugify the directory name: lowercase, dashes for spaces, no slashes\n(`run-billing-api`, not `run-billing\u002Fapi`). The directory name and\nthe frontmatter `name:` should match - that's the slash command.\n\n## Process\n\n### 0. Find any existing skill about running this app\n\nList the project's skills with their descriptions (same probe `\u002Frun`\nuses - users name these variously, so match on description, not name):\n\n```bash\nd=$PWD; while :; do\n  grep -Hm1 '^description:' \"$d\"\u002F.claude\u002Fskills\u002F*\u002FSKILL.md 2>\u002Fdev\u002Fnull\n  [ -e \"$d\u002F.git\" ] || [ \"$d\" = \u002F ] && break\n  d=$(dirname \"$d\")\ndone\n```\n\nIf one is about launching\u002Fdriving this app - whatever it's named -\n**refine, don't rewrite**: verify its claims, fix what's wrong, add\nwhat's missing, preserve what works. Re-run the driver if there is\none. Keep its existing name.\n\n(Also check for a legacy `.claude\u002Frun.md` - earlier versions of this\ntool produced those. If you find one, migrate it: the body becomes\nthe skill's `SKILL.md` content, any referenced scripts move into the\nskill dir, and delete the old file.)\n\nIf none exists, decide where to create it (see above) and continue.\n\n### 1. Discover - and treat every claim as disprovable\n\nFigure out what you're authoring for:\n\n- Manifest right here (`package.json`, `go.mod`, `pyproject.toml`...) and\n  it's one self-contained thing -> this is the unit.\n- Looks like a mega-repo root (`apps\u002F`, `packages\u002F`, `services\u002F`) ->\n  **ask which one.** List candidates, let them pick, `cd` there.\n- Genuinely ambiguous -> ask.\n\nSurvey the usual places: `README.md`, `package.json` scripts,\n`Dockerfile`, `Makefile`, `.github\u002Fworkflows\u002F`, `CONTRIBUTING.md`. CI\nconfigs are often more accurate than READMEs.\n\n**Every claim in existing docs is a hypothesis.** Especially the\nnegative ones:\n\n| When docs say... | What you do |\n|---|---|\n| \"Requires macOS\u002FWindows\" | Launch it on Linux anyway. Apps rarely refuse to start - they crash on a missing `.so`, which `apt-get` fixes. Native modules for *your host's* keychain\u002Fnotifications may no-op; the core usually runs. |\n| \"Requires a GPU\" | Try software rendering. Electron\u002FChrome fall back with `--disable-gpu`. |\n| \"Requires a paid account \u002F feature flag\" | The gate is code you can read. Find it (env var? build define? SSR-embedded JSON?) and patch it for your local run. Document the patch. |\n| \"Run `npm start`\" | That's the human path (spawns a window, waits forever). Find or build the *programmatic* path - `electron-forge start` to build then launch via Playwright, or equivalent. |\n\n\"Not supported on Linux\" in a README written by a macOS developer\nmeans \"I never tried.\" You're about to try. **If you give up here, the\nskill you write is the README with extra steps.**\n\n### 2. Execute - and BUILD the harness you need\n\nYou're in a headless Linux container. The app is going to fight you.\nThat fight is the content of the skill.\n\nKeep a running `NOTES.md` as you go. Every error -> every fix -> every\ncommand that finally worked. This scratchpad becomes the\nTroubleshooting section.\n\n**Work up to a real interaction:**\n\n- **Install + build.** When something's missing, note the exact\n  `apt-get` \u002F `npm install` that fixed it.\n- **Launch the app.** Not the test suite - the app. A desktop GUI\n  (Electron, native) needs `xvfb-run` and a handful of `lib*`\n  packages; a web app driven by `chromium-cli` runs headless and\n  needs neither. Launch timeouts and cryptic crashes are normal at\n  this stage. Read the stack trace, install the missing thing, try\n  again.\n- **Build a harness to drive it.** You need a handle on the running\n  app that lets you send input and observe output programmatically.\n  The shape depends on the project (see table below).\n\n  **Cover the layer(s) PRs actually touch.** A tmux driver that pokes\n  the CLI's user surface is the right handle for UI changes - and the\n  wrong one for a PR that touches one internal function. For the\n  latter an agent wants `NODE_ENV=test bun run script.ts` (or\n  equivalent): import the function, call it, observe. If most PRs\n  here touch internals, that direct-invocation path is the driver's\n  main entry point, and the tmux launch is secondary. Look at recent\n  merged PRs: what layer do they touch? Cover that.\n\n  For a **web** app, `chromium-cli` is the driver - you script it,\n  you don't write it (see [examples\u002Fplaywright.md](examples\u002Fplaywright.md)).\n  For a **desktop** GUI (Electron), write a REPL driver (stdin\n  commands -> click\u002Ftype\u002Fscreenshot), run it inside tmux, and use\n  `send-keys` \u002F `capture-pane`. You will iterate on that driver - it\n  starts minimal (`launch`, `ss`, `quit`) and grows whatever commands\n  you need to reach the interesting part of the app.\n- **Do one real user flow end-to-end.** Click the button. Fill the\n  form. See the result in the DOM. Take a screenshot. **Actually look\n  at the screenshot.** If it's blank or showing an error page, you're\n  not done.\n- **Then run the tests.** Unit tests are a sanity check, not the main\n  event.\n- **Stop cleanly.**\n\n**Obstacles are content.** You will hit weird ones - coordinate systems\nthat don't line up, APIs that return empty on this Electron version,\nfeature gates that hide the thing you need to test. Each of these gets\na bullet in Gotchas and (often) a helper in your driver. The gold\nstandard is a Gotchas section full of things nobody could have guessed.\n\n**The driver script gets committed alongside the skill.** It is not\nscaffolding. It is the way future agents (and humans) will drive this\napp. It defaults to living inside the skill directory (for a web app\nusing `chromium-cli`, that means inline in `SKILL.md` - the heredoc\nis the script). If it outgrows that - if the project's real test\nsuite wants to import from it - move it to `scripts\u002F` or `e2e\u002F` and\nupdate `SKILL.md` to point there.\n\n### 3. Write SKILL.md\n\nShort. Point at the driver. Use [template.md](template.md) as the\nstarting structure - it has the frontmatter shape.\n\n**The frontmatter matters.** The `name:` becomes the slash command\n(`\u002Frun-billing`). The `description:` is what Claude scans to decide\nwhether to auto-load this skill - put the **verbs an agent would\nactually type** in it: \"run,\" \"start,\" \"build,\" \"test,\" \"screenshot.\"\nGeneric descriptions (\"helpful utilities for billing\") won't match.\n\nBody structure:\n\n1. One-paragraph intro: what this app is, how it's driven -\n   `\u003Cdriver-path>` under xvfb\u002Ftmux for desktop, `chromium-cli` for\n   web, `curl` for a server.\n2. **Prerequisites** - the exact `apt-get install` line you ran.\n3. **Build** - the exact commands, in order. Include any patches you\n   had to apply (feature gates, config overrides) with the exact `sed`\n   or edit.\n4. **Run (agent path)** - FIRST. How to launch the driver, what\n   commands it accepts, where screenshots land. If it's a REPL, show\n   the tmux wrapping. This is the section the next agent will actually\n   use.\n5. **Run (human path)** - SECOND, if different. `npm start` -> window\n   opens -> Ctrl-C. Brief. Note that it's useless headless.\n6. **Gotchas** - the battle scars. The things that look like they\n   should work but don't, and the workaround. If this section is\n   generic, you didn't fight hard enough.\n7. **Troubleshooting** - symptom -> fix. Only errors you actually hit.\n\nKeep it **verified** (you ran it), **prescriptive** (one path, not\noptions), **honest** (flaky? slow? say so).\n\n**Paths in SKILL.md are relative to `\u003Cunit>\u002F`,** not to the skill\ndirectory. State this at the top if there's any ambiguity. When the\ndriver lives inside the skill, its path from `\u003Cunit>` is\n`.claude\u002Fskills\u002Frun-\u003Cunit-name>\u002Fdriver.mjs` - it's long, but explicit.\n\n### 4. Verify\n\nFresh shell, `cd` into the unit, follow the skill's `SKILL.md`\nline-by-line without deviating. Any improvisation = a gap. Fix it.\n\n## Project-type patterns\n\nPick a starting shape for your driver. These examples are shared with\nthe `\u002Frun` skill (same per-project-type patterns are used as the\nfallback when no project-specific run skill exists) - if you're\nauthoring a new one, the example is your starting template.\n\n| Project type | Driver shape | Example |\n|---|---|---|\n| Web server \u002F API | Background-launch + `curl`-based smoke script | [examples\u002Fserver.md](examples\u002Fserver.md) |\n| CLI tool | Representative-args smoke script, check exit codes + output | [examples\u002Fcli.md](examples\u002Fcli.md) |\n| TUI \u002F interactive terminal | tmux wrapper: `send-keys` \u002F `capture-pane` | [examples\u002Ftui.md](examples\u002Ftui.md) |\n| Electron \u002F desktop GUI | Playwright `_electron` REPL driver under xvfb, screenshots, tmux-wrapped | [examples\u002Felectron.md](examples\u002Felectron.md) |\n| Browser-driven | dev server + `chromium-cli` script | [examples\u002Fplaywright.md](examples\u002Fplaywright.md) |\n| Library \u002F SDK | Import-and-call smoke script | [examples\u002Flibrary.md](examples\u002Flibrary.md) |\n\nFor a web app, start from [examples\u002Fplaywright.md](examples\u002Fplaywright.md)\n-- drive it with `chromium-cli`, no custom driver needed. For a\ndesktop app, start from [examples\u002Felectron.md](examples\u002Felectron.md)\n-- it has the full `_electron` REPL driver skeleton, the tmux wrapping,\nand the catalog of obstacles you'll hit.\n\n## What to include\n\n- **Prerequisites** - OS packages, runtimes, tools. Ubuntu `apt-get`\n  lines. The exact ones.\n- **Setup** - install deps, configure, any patches.\n- **Build** - compile\u002Fbundle.\n- **Run (agent path)** - the driver. Commands. Screenshot location.\n- **Direct invocation** - if callable: how to import and run internal\n  code without the full app. The env var \u002F flag that bypasses init\n  guards. Many PRs need only this.\n- **Run (human path)** - if meaningfully different.\n- **Test** - the test suite command.\n- **Gotchas** - non-obvious traps you hit.\n- **Troubleshooting** - error -> fix.\n- **The driver itself** - committed in the skill dir (or graduated\n  to `scripts\u002F`\u002F`e2e\u002F`), or inline in `SKILL.md` for `chromium-cli`\n  web apps; referenced from `SKILL.md` either way.\n\n## What to leave out\n\n- **Anything you didn't run.** If the README says `yarn start:prod` and\n  you never ran it, it's not in the skill. Full stop.\n- **Documented happy paths for platforms you're not on.** You're in a\n  Linux container. A macOS-only section you can't verify is\n  speculation. Mention it exists; don't elaborate.\n- **Exhaustive options.** One working path.\n- **Architecture prose.** That's other docs.\n- **Generic troubleshooting.** \"If the build fails, check your Node\n  version\" - useless. Only include errors you actually hit and fixed.\n\n## Red flags - you are about to ship the wrong thing\n\nStop and reconsider if:\n\n- **You haven't taken a screenshot** of a GUI app. You didn't run it.\n- **Your skill has no driver\u002Fsmoke script** to point at, and the app\n  is interactive. The next agent has no way to drive it. (Web app\n  using `chromium-cli`? - the heredoc in `SKILL.md` is the driver;\n  no separate file needed.)\n- **Your skill reads like the README.** Same structure, same\n  commands, same caveats. You paraphrased.\n- **Your Troubleshooting section is generic.** Real execution produces\n  specific, weird errors. Generic errors = you didn't execute.\n- **You wrote \"not supported on this platform\"** without trying to\n  launch it. The README author was on a Mac. You are not. Try.\n- **Everything worked first try.** Either this project is trivially\n  simple, or you ran the test suite and called it done.\n","Your job is to produce a **skill** at `\u003Cunit>\u002F.claude\u002Fskills\u002Frun-\u003Cunit-name>\u002F`\nthat lets a future agent build, launch, and **drive** this project from\na clean machine.\n\nThe skill has two parts that live together:\n\n```\n\u003Cunit>\u002F.claude\u002Fskills\u002Frun-\u003Cunit-name>\u002F\n  SKILL.md      \u003C- agent-facing instructions - SHORT. Points at the driver.\n  driver.mjs    \u003C- (or driver.py, smoke.sh, ... - or none: web apps use\n                   chromium-cli off-the-shelf, and the heredoc in\n                   SKILL.md is the script)\n```\n\nThat almost always means **writing code**, not just prose. If the app\nhas any interactive surface (GUI, TUI, long-running server, REPL), the\nfuture agent needs a programmatic way to poke it. A markdown file by\nitself cannot click a button - but sometimes the button-clicker\nalready exists: for web apps it's `chromium-cli`, for servers it's\n`curl`. You build (or script) that harness now, commit it alongside\nthe skill, and the `SKILL.md` documents how to use it.\n\n## Definition of done\n\nYou are done when **all** of these are true:\n\n1. **You launched the app in this container and interacted with it** -\n   not its test suite, the actual running app. For anything with a GUI,\n   that means you have a screenshot file on disk that you took.\n2. **The interaction harness is committed** next to the skill. A driver\n   script, a REPL wrapper, a smoke test, or the `chromium-cli` heredoc\n   inline in `SKILL.md` - whatever you used to drive the app in step 1.\n   (Graduated into `scripts\u002F`\u002F`e2e\u002F`? - fine, point at it. Web app with\n   `chromium-cli` off-the-shelf? - the inline script is the harness; no\n   separate file.)\n3. **The `SKILL.md` documents the harness** as the primary agent path -\n   the section a future agent reads first is \"run this driver \u002F pipe\n   these commands to `chromium-cli`,\" not \"run `npm start` and a window\n   opens.\"\n4. **Every code block in `SKILL.md` is a command you ran that worked.**\n   This session. This container. Not from the README, not inferred.\n\nIf you're about to write the skill and you don't have (1), **stop.** You\nare about to paraphrase existing docs. That document already exists -\nit's called the README, and the whole reason you're here is that it\nwasn't enough.\n\n## The deliverables are code AND docs\n\nTypical output is a skill directory containing both:\n\n```\n\u003Cunit>\u002F.claude\u002Fskills\u002Frun-\u003Cunit>\u002F\n  SKILL.md         \u003C- SHORT. Points at the driver. Has the frontmatter\n                     that lets Claude auto-load it when someone asks\n                     to \"run \u003Cunit>\" or \"screenshot \u003Cunit>\".\n  driver.mjs       \u003C- (or driver.py, smoke.sh, ... - or none: web apps\n                     use chromium-cli off-the-shelf, and the heredoc\n                     in SKILL.md is the script)\n```\n\nThe driver lives **inside the skill directory** by default. They are a\npair - the skill's instructions and the code that implements them. A\ndriver that lives here is allowed to be a bit messier than production\ncode; it's agent tooling, not product surface.\n\n**Graduation:** if the driver grows into something the project's own\ntest suite wants to reuse - shared launch helpers, a real e2e harness -\nmove it to `scripts\u002F` or `e2e\u002F` and update `SKILL.md` to reference the\nnew path. The skill stays; the driver finds a better home.\n\nThe exact shape depends on the project, but the principle is constant:\n**the driver is the deliverable.** The `SKILL.md` is its man page. For\na web app, the driver already exists - `chromium-cli`\n([examples\u002Fplaywright.md](examples\u002Fplaywright.md)) - and the skill is\nthe script that runs it. For a desktop app\n([examples\u002Felectron.md](examples\u002Felectron.md)), the driver is a custom\nREPL under tmux that exposes `launch`\u002F`ss`\u002F`click`\u002F`eval`. For a server,\nthe driver is `curl`. Whatever shape it takes, without something that\nreaches into the running app, the skill is a description of a window\nnobody can touch.\n\n## Where the skill goes\n\nThe skill lives at `\u003Cunit>\u002F.claude\u002Fskills\u002Frun-\u003Cunit-name>\u002F`, where\n`\u003Cunit>` is the directory for **one deployable thing** - an app, a\nservice, a library.\n\nClaude Code **natively discovers** skills from nested `.claude\u002Fskills\u002F`\ndirectories: an agent working anywhere inside `\u003Cunit>` will see\n`\u002Frun-\u003Cunit-name>` as an available skill, and it auto-loads when the\nrequest matches its description (e.g. \"run the desktop app,\" \"take a\nscreenshot of billing\").\n\n- **Single-project repo:** `.claude\u002Fskills\u002Frun-\u003Crepo-name>\u002F` at repo root.\n- **Large repo with many apps:** one per app, colocated -\n  `apps\u002Fbilling\u002F.claude\u002Fskills\u002Frun-billing\u002F`,\n  `apps\u002Fdesktop\u002F.claude\u002Fskills\u002Frun-desktop\u002F`.\n- **App with multiple binaries:** still **one** skill at the app's\n  root with a section per binary. They share setup. Start from the\n  closest single-binary example and add a `## Run: \u003Cname>` section\n  per binary.\n\nIf you're not sure where the unit boundary is, **ask the user.**\n\nSlugify the directory name: lowercase, dashes for spaces, no slashes\n(`run-billing-api`, not `run-billing\u002Fapi`). The directory name and\nthe frontmatter `name:` should match - that's the slash command.\n\n## Process\n\n### 0. Find any existing skill about running this app\n\nList the project's skills with their descriptions (same probe `\u002Frun`\nuses - users name these variously, so match on description, not name):\n\n```bash\nd=$PWD; while :; do\n  grep -Hm1 '^description:' \"$d\"\u002F.claude\u002Fskills\u002F*\u002FSKILL.md 2>\u002Fdev\u002Fnull\n  [ -e \"$d\u002F.git\" ] || [ \"$d\" = \u002F ] && break\n  d=$(dirname \"$d\")\ndone\n```\n\nIf one is about launching\u002Fdriving this app - whatever it's named -\n**refine, don't rewrite**: verify its claims, fix what's wrong, add\nwhat's missing, preserve what works. Re-run the driver if there is\none. Keep its existing name.\n\n(Also check for a legacy `.claude\u002Frun.md` - earlier versions of this\ntool produced those. If you find one, migrate it: the body becomes\nthe skill's `SKILL.md` content, any referenced scripts move into the\nskill dir, and delete the old file.)\n\nIf none exists, decide where to create it (see above) and continue.\n\n### 1. Discover - and treat every claim as disprovable\n\nFigure out what you're authoring for:\n\n- Manifest right here (`package.json`, `go.mod`, `pyproject.toml`...) and\n  it's one self-contained thing -> this is the unit.\n- Looks like a mega-repo root (`apps\u002F`, `packages\u002F`, `services\u002F`) ->\n  **ask which one.** List candidates, let them pick, `cd` there.\n- Genuinely ambiguous -> ask.\n\nSurvey the usual places: `README.md`, `package.json` scripts,\n`Dockerfile`, `Makefile`, `.github\u002Fworkflows\u002F`, `CONTRIBUTING.md`. CI\nconfigs are often more accurate than READMEs.\n\n**Every claim in existing docs is a hypothesis.** Especially the\nnegative ones:\n\n| When docs say... | What you do |\n|---|---|\n| \"Requires macOS\u002FWindows\" | Launch it on Linux anyway. Apps rarely refuse to start - they crash on a missing `.so`, which `apt-get` fixes. Native modules for *your host's* keychain\u002Fnotifications may no-op; the core usually runs. |\n| \"Requires a GPU\" | Try software rendering. Electron\u002FChrome fall back with `--disable-gpu`. |\n| \"Requires a paid account \u002F feature flag\" | The gate is code you can read. Find it (env var? build define? SSR-embedded JSON?) and patch it for your local run. Document the patch. |\n| \"Run `npm start`\" | That's the human path (spawns a window, waits forever). Find or build the *programmatic* path - `electron-forge start` to build then launch via Playwright, or equivalent. |\n\n\"Not supported on Linux\" in a README written by a macOS developer\nmeans \"I never tried.\" You're about to try. **If you give up here, the\nskill you write is the README with extra steps.**\n\n### 2. Execute - and BUILD the harness you need\n\nYou're in a headless Linux container. The app is going to fight you.\nThat fight is the content of the skill.\n\nKeep a running `NOTES.md` as you go. Every error -> every fix -> every\ncommand that finally worked. This scratchpad becomes the\nTroubleshooting section.\n\n**Work up to a real interaction:**\n\n- **Install + build.** When something's missing, note the exact\n  `apt-get` \u002F `npm install` that fixed it.\n- **Launch the app.** Not the test suite - the app. A desktop GUI\n  (Electron, native) needs `xvfb-run` and a handful of `lib*`\n  packages; a web app driven by `chromium-cli` runs headless and\n  needs neither. Launch timeouts and cryptic crashes are normal at\n  this stage. Read the stack trace, install the missing thing, try\n  again.\n- **Build a harness to drive it.** You need a handle on the running\n  app that lets you send input and observe output programmatically.\n  The shape depends on the project (see table below).\n\n  **Cover the layer(s) PRs actually touch.** A tmux driver that pokes\n  the CLI's user surface is the right handle for UI changes - and the\n  wrong one for a PR that touches one internal function. For the\n  latter an agent wants `NODE_ENV=test bun run script.ts` (or\n  equivalent): import the function, call it, observe. If most PRs\n  here touch internals, that direct-invocation path is the driver's\n  main entry point, and the tmux launch is secondary. Look at recent\n  merged PRs: what layer do they touch? Cover that.\n\n  For a **web** app, `chromium-cli` is the driver - you script it,\n  you don't write it (see [examples\u002Fplaywright.md](examples\u002Fplaywright.md)).\n  For a **desktop** GUI (Electron), write a REPL driver (stdin\n  commands -> click\u002Ftype\u002Fscreenshot), run it inside tmux, and use\n  `send-keys` \u002F `capture-pane`. You will iterate on that driver - it\n  starts minimal (`launch`, `ss`, `quit`) and grows whatever commands\n  you need to reach the interesting part of the app.\n- **Do one real user flow end-to-end.** Click the button. Fill the\n  form. See the result in the DOM. Take a screenshot. **Actually look\n  at the screenshot.** If it's blank or showing an error page, you're\n  not done.\n- **Then run the tests.** Unit tests are a sanity check, not the main\n  event.\n- **Stop cleanly.**\n\n**Obstacles are content.** You will hit weird ones - coordinate systems\nthat don't line up, APIs that return empty on this Electron version,\nfeature gates that hide the thing you need to test. Each of these gets\na bullet in Gotchas and (often) a helper in your driver. The gold\nstandard is a Gotchas section full of things nobody could have guessed.\n\n**The driver script gets committed alongside the skill.** It is not\nscaffolding. It is the way future agents (and humans) will drive this\napp. It defaults to living inside the skill directory (for a web app\nusing `chromium-cli`, that means inline in `SKILL.md` - the heredoc\nis the script). If it outgrows that - if the project's real test\nsuite wants to import from it - move it to `scripts\u002F` or `e2e\u002F` and\nupdate `SKILL.md` to point there.\n\n### 3. Write SKILL.md\n\nShort. Point at the driver. Use [template.md](template.md) as the\nstarting structure - it has the frontmatter shape.\n\n**The frontmatter matters.** The `name:` becomes the slash command\n(`\u002Frun-billing`). The `description:` is what Claude scans to decide\nwhether to auto-load this skill - put the **verbs an agent would\nactually type** in it: \"run,\" \"start,\" \"build,\" \"test,\" \"screenshot.\"\nGeneric descriptions (\"helpful utilities for billing\") won't match.\n\nBody structure:\n\n1. One-paragraph intro: what this app is, how it's driven -\n   `\u003Cdriver-path>` under xvfb\u002Ftmux for desktop, `chromium-cli` for\n   web, `curl` for a server.\n2. **Prerequisites** - the exact `apt-get install` line you ran.\n3. **Build** - the exact commands, in order. Include any patches you\n   had to apply (feature gates, config overrides) with the exact `sed`\n   or edit.\n4. **Run (agent path)** - FIRST. How to launch the driver, what\n   commands it accepts, where screenshots land. If it's a REPL, show\n   the tmux wrapping. This is the section the next agent will actually\n   use.\n5. **Run (human path)** - SECOND, if different. `npm start` -> window\n   opens -> Ctrl-C. Brief. Note that it's useless headless.\n6. **Gotchas** - the battle scars. The things that look like they\n   should work but don't, and the workaround. If this section is\n   generic, you didn't fight hard enough.\n7. **Troubleshooting** - symptom -> fix. Only errors you actually hit.\n\nKeep it **verified** (you ran it), **prescriptive** (one path, not\noptions), **honest** (flaky? slow? say so).\n\n**Paths in SKILL.md are relative to `\u003Cunit>\u002F`,** not to the skill\ndirectory. State this at the top if there's any ambiguity. When the\ndriver lives inside the skill, its path from `\u003Cunit>` is\n`.claude\u002Fskills\u002Frun-\u003Cunit-name>\u002Fdriver.mjs` - it's long, but explicit.\n\n### 4. Verify\n\nFresh shell, `cd` into the unit, follow the skill's `SKILL.md`\nline-by-line without deviating. Any improvisation = a gap. Fix it.\n\n## Project-type patterns\n\nPick a starting shape for your driver. These examples are shared with\nthe `\u002Frun` skill (same per-project-type patterns are used as the\nfallback when no project-specific run skill exists) - if you're\nauthoring a new one, the example is your starting template.\n\n| Project type | Driver shape | Example |\n|---|---|---|\n| Web server \u002F API | Background-launch + `curl`-based smoke script | [examples\u002Fserver.md](examples\u002Fserver.md) |\n| CLI tool | Representative-args smoke script, check exit codes + output | [examples\u002Fcli.md](examples\u002Fcli.md) |\n| TUI \u002F interactive terminal | tmux wrapper: `send-keys` \u002F `capture-pane` | [examples\u002Ftui.md](examples\u002Ftui.md) |\n| Electron \u002F desktop GUI | Playwright `_electron` REPL driver under xvfb, screenshots, tmux-wrapped | [examples\u002Felectron.md](examples\u002Felectron.md) |\n| Browser-driven | dev server + `chromium-cli` script | [examples\u002Fplaywright.md](examples\u002Fplaywright.md) |\n| Library \u002F SDK | Import-and-call smoke script | [examples\u002Flibrary.md](examples\u002Flibrary.md) |\n\nFor a web app, start from [examples\u002Fplaywright.md](examples\u002Fplaywright.md)\n-- drive it with `chromium-cli`, no custom driver needed. For a\ndesktop app, start from [examples\u002Felectron.md](examples\u002Felectron.md)\n-- it has the full `_electron` REPL driver skeleton, the tmux wrapping,\nand the catalog of obstacles you'll hit.\n\n## What to include\n\n- **Prerequisites** - OS packages, runtimes, tools. Ubuntu `apt-get`\n  lines. The exact ones.\n- **Setup** - install deps, configure, any patches.\n- **Build** - compile\u002Fbundle.\n- **Run (agent path)** - the driver. Commands. Screenshot location.\n- **Direct invocation** - if callable: how to import and run internal\n  code without the full app. The env var \u002F flag that bypasses init\n  guards. Many PRs need only this.\n- **Run (human path)** - if meaningfully different.\n- **Test** - the test suite command.\n- **Gotchas** - non-obvious traps you hit.\n- **Troubleshooting** - error -> fix.\n- **The driver itself** - committed in the skill dir (or graduated\n  to `scripts\u002F`\u002F`e2e\u002F`), or inline in `SKILL.md` for `chromium-cli`\n  web apps; referenced from `SKILL.md` either way.\n\n## What to leave out\n\n- **Anything you didn't run.** If the README says `yarn start:prod` and\n  you never ran it, it's not in the skill. Full stop.\n- **Documented happy paths for platforms you're not on.** You're in a\n  Linux container. A macOS-only section you can't verify is\n  speculation. Mention it exists; don't elaborate.\n- **Exhaustive options.** One working path.\n- **Architecture prose.** That's other docs.\n- **Generic troubleshooting.** \"If the build fails, check your Node\n  version\" - useless. Only include errors you actually hit and fixed.\n\n## Red flags - you are about to ship the wrong thing\n\nStop and reconsider if:\n\n- **You haven't taken a screenshot** of a GUI app. You didn't run it.\n- **Your skill has no driver\u002Fsmoke script** to point at, and the app\n  is interactive. The next agent has no way to drive it. (Web app\n  using `chromium-cli`? - the heredoc in `SKILL.md` is the driver;\n  no separate file needed.)\n- **Your skill reads like the README.** Same structure, same\n  commands, same caveats. You paraphrased.\n- **Your Troubleshooting section is generic.** Real execution produces\n  specific, weird errors. Generic errors = you didn't execute.\n- **You wrote \"not supported on this platform\"** without trying to\n  launch it. The README author was on a Mac. You are not. Try.\n- **Everything worked first try.** Either this project is trivially\n  simple, or you ran the test suite and called it done.\n",{"name":5,"description":7,"disable-model-invocation":24},true,"main","d69096e239f2a89aabc2cd078c47664a9d3b1e11","02-厂商系统提示词\u002Fsystem_prompts_leaks\u002FAnthropic\u002Fclaude-code\u002Fskills\u002Frun-skill-generator",3,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts","https:\u002F\u002Fgithub.com\u002Flza6","个人vx：Tf00798\r\n个人✈：@TFGY999","2026-10-08 14:29:22",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":24},[35,38,41],{"tool":36,"command":37},"npx","npx skills add lza6\u002FArmor-piercing-system-prompts --skill run-skill-generator",{"tool":39,"command":40},"bunx","bunx skills add lza6\u002FArmor-piercing-system-prompts --skill run-skill-generator",{"tool":42,"command":43},"pnpm","pnpm dlx skills add lza6\u002FArmor-piercing-system-prompts --skill run-skill-generator","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F02-厂商系统提示词\u002Fsystem_prompts_leaks\u002FAnthropic\u002Fclaude-code\u002Fskills\u002Frun-skill-generator\u002FSKILL.md",[46],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"template.md","REFERENCE","参考文档",4096,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F02-厂商系统提示词\u002Fsystem_prompts_leaks\u002FAnthropic\u002Fclaude-code\u002Fskills\u002Frun-skill-generator\u002Ftemplate.md",[53,61,66,73,80,87],{"skillNo":54,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":60},"01M7FFG485KY83TD6HSC4YPCMC","obsidian-clipper-template-creator","Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator\u002FSKILL.md",[],[],"2026-11-16 10:05:34",{"skillNo":62,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"publishedAt":60},"01M7FFG0T6AZ0STEJPSB3CX84K","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator-antigravity-awesome-skills-main\u002FSKILL.md",[],[],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":60},"01M7FFFY0QEM14SKWPDYGMXM2P","python-api","Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP\u002FmAP semantics, and source-style import diagnostics.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Fpython-api\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":60},"01M7FFFVZQZ2GJ0C0TGP7JHD20","file-evaluation","Folder-based PASCAL VOC AP\u002FmAP evaluation from ground-truth and detection text files.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Ffile-evaluation\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":60},"01M7FFFSFNYSE77SJ6PE1CK2DA","object-detection-metrics","Guides legacy Object-Detection-Metrics PASCAL VOC AP\u002FmAP evaluation from detection text folders and direct Python API classes.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":60},"01M7FFFP0GCK1RVXBCPQ8APBMT","nx-workspace-patterns","Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fnx-workspace-patterns\u002FSKILL.md",[],[]]