AX-skills
AGENTS.md4 starsChanged 54 days agoMITView on GitHub
Profile: public skills and showcase. Published skills and demos are reference artifacts, not the live agent configuration. Validate only the changed skill/script or render the changed demo; keep machine identifiers and credentials out of public examples. Do not install skills or rewrite borrowed upstream sources as incidental cleanup. Curated public collection of AX's Agent Skills + GitHub Pages demo pages and demo videos for selected skills. Live: <https://xingfanxia.github.io/AX-skills/> Every interactive demo page is guizang-ppt-skill Style A (电子杂志) · 🌙 沙丘…
- Installs packages
- Commits and pushes
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository39 other files it gives its agents.
- agentic-repo-scaffoldSkill
- agent-travel-plannerSkill
- ai-workflow-spineSkill
- ana-awardSkill
- analyze-video-evidenceSkill
- apple-pdfSkill
- audit-fix-loopSkill
- autonomous-grindSkill
- award-flightsSkill
- banxian-skillSkill
- big-taskSkill
- clean-room-rewriteSkill
- cloudflare-r2-setupSkill
- codebase-sweepSkill
- context-infrastructureSkill
- deep-researchSkill
- download-web-videoSkill
- dr-sharpSkill
- game-script-creationSkill
- gemini-imageSkill
- gpt-imageSkill
- investigateSkill
- jewelry-marketingSkill
- learning-coachSkill
- mineru-pdf-parserSkill
- mtcSkill
- narrative-researchSkill
- neat-freakSkill
- pet-forgeSkill
- plan-design-reviewSkill
- pr-fix-loopSkill
- proxy-node-setupSkill
- serenity-bottleneck-researchSkill
- task-shaped-agent-routingSkill
- transcribeSkill
- tridentSkill
- tui-engineeringSkill
- wandian-writerSkill
- web-novel-writingSkill
The file
We split this AGENTS.md into the rules it gives an agent
6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use the smallest existing check that proves the change.
- Distinctive: line 9
Keep the existing structure.
- Distinctive: line 26
Every interactive demo page is guizang-ppt-skill Style A (电子杂志) · 🌙 沙丘 / Dune — uniform:
- Very specific: line 37
Add <skill>/SKILL.md — the skill source (genericized.
- Very specific: line 39
Add an <article class="card <skill>"> to docs/index.html: num / role / h2+.id(skill name) / one-line p / .src(源自 <product> or 配套 <blog>) / …
- Very specific: line 60
Pages CDN (Fastly) caches hard: After a push the new path can 404 / serve stale for ~1–2 min.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AX-skills — agent guide
Project scale and verification
Profile: public skills and showcase. Published skills and demos are reference artifacts, not the live agent configuration. Validate only the changed skill/script or render the changed demo; keep machine identifiers and credentials out of public examples. Do not install skills or rewrite borrowed upstream sources as incidental cleanup.
- The requested behavior/questions define completion. Reviews are read-only unless fixes are requested; report unrelated findings briefly without adding tasks or test backfill.
- Use the smallest existing check that proves the change. Add tests for a concrete regression or consequential boundary; do not impose blanket TDD, new coverage targets, full suites, plans or reviewers. Preserve configured CI and actual release gates; reuse still-valid results.
- Keep the existing structure. Internal contract errors should be clear; add retries, fallbacks or compatibility layers only for an observed external failure or supported contract. Keep secrets private and inspect security only at boundaries changed by this task.
Curated public collection of AX's Agent Skills + GitHub Pages demo pages and demo videos for selected skills. Live: https://xingfanxia.github.io/AX-skills/
Layout
<skill-name>/SKILL.md— maintained AX skill source. Runtime links or installed copies consume this source; resolve their ownership before editing or syncing, rather than treating a runtime copy as a competing authority. Preserve upstream ownership and attribution for borrowed skills.docs/— the Pages site (served frommainbranch/docs, legacy branch-build, auto-deploys on push):docs/index.html— showcase landing (card grid; each card has a JS-injected▶ 影片button → R2-streamed demo video in a lightbox).docs/<skill>/index.html— that skill's interactive demo page (+assets/motion.min.js, optionalimages/, optional PDF; carries the shared#promo-linksnippet + a per-skill#source-link(源码 ↗→tree/main/<skill-dir>) after#home-link). Existing*-roadshow.pdffilenames are legacy compatibility names, not current product terminology.docs/.nojekyll.
remotion/— Remotion source for demo videos (committed;node_modules/+out/gitignored).src/terminal/sessions.ts= the per-skill scripted Claude-Code sessions (the demo content + source of truth),src/terminal/TerminalPromo.tsx= the terminal-sim composition engine,scripts/render-all.mjs+scripts/upload-r2.mjs= the pipeline. Rendered.mp4s are NOT in the repo — they live on R2 (ax-blog-mediabucket,ax-skills/prefix) served athttps://media.ax0x.ai/ax-skills/<slug>.mp4. Composition id === demo slug === R2 filename (one source of truth).scripts/validate_skills.py— structural skill linter (stdlib-only). Runpython3 scripts/validate_skills.pyafter adding/editing any skill: enforces SKILL.md frontmattername=== dir name, 300/500 line caps (500+ needs anOVERSIZE_ALLOWLISTentry with a reason), danglingreferences/*.md/scripts/*.<ext>mentions, orphan asset files, and hardcoded/Users/<name>/paths.--skill <name>for one skill,--strictto make warnings fail.
Demo page style — ONE family, do not diverge
Every interactive demo page is guizang-ppt-skill Style A (电子杂志) · 🌙 沙丘 / Dune — uniform:
- theme
--ink:#1f1a14/--paper:#f0e6d2, dark-paper body - fonts: Playfair Display + Source Serif 4 + Noto Serif SC (serif display/headings/numbers) + Noto Sans SC (body) + IBM Plex Mono (kicker/meta/code)
- shared scaffold: WebGL dual-bg, magazine chrome/foot,
#navpill pager, top-left chrome pills (← AX Skillshome link +▶ Demovideo + per-skill源码 ↗source link), ESC overview,Blow-power, Motion One local-first (assets/motion.min.js) + CDN fallback.
The README once described banxian=IKB / jewelry=柠檬黄 (Swiss) — that's historical; all demos were unified to Dune Style A. Match the current files, not old docs.
When building a new demo page: read an existing demo (e.g. docs/jewelry/index.html) as the exact reference and produce a structural sibling. Never introduce a different theme/style.
Add a new skill demo (the flow)
- Add
<skill>/SKILL.md— the skill source (genericized; see rules). - Build
docs/<skill>/index.htmlin Style A Dune, sibling of an existing demo page; copyassets/motion.min.jsalongside. - Add an
<article class="card <skill>">todocs/index.html:num/role/h2+.id(skill name) / one-linep/.src(源自 <product>or配套 <blog>) /.cta=查看 →(<skill>/) +源码 ↗(github.com/xingfanxia/AX-skills/tree/main/<skill>) [+ optionalPDF ↓]. - Add a demo video (optional): write a scripted terminal session for the skill in
remotion/src/terminal/sessions.ts(id=== demo slug) — ground it in the skill's realSKILL.md(actual invocation / workflow phases / output structure / persona, NOT the card blurb — blurb-scripted sessions misrepresent the skill and get rejected). Keep ≤16 visible lines, font-safe glyphs only (no hexagram chars/emoji → tofu), genericize infra. Add the id to the arrays inremotion/scripts/*.mjs, thencd remotion && node scripts/render-all.mjs <slug> && node scripts/upload-r2.mjs <slug>. The landing▶ 影片button + demo page#promo-linkauto-resolvemedia.ax0x.ai/ax-skills/<slug>.mp4from the slug — no per-card HTML edit needed. Visually-verify a near-end still before declaring done. git push origin main→ Pages rebuilds from/docs.- Visual-verify: open the demo + screenshot key views; play the demo-video lightbox; confirm it's a Dune sibling and reads clean.
Demo videos (Remotion)
Each showcased skill may have a ~13–18s 1920×1080 demo video, authored in remotion/ and hosted on R2 (NOT committed). The demo is a scripted Claude-Code terminal session showing the skill actually being used — not a replay of the interactive page. Design: an inverted-Dune terminal window (ink #17130c body / paper #f0e6d2 text, IBM Plex Mono + Noto Sans SC) floating on the paper dot-grid frame, with traffic-light chrome and the brand corners. The user types an invocation, the agent works (spinner → ✓ steps), output streams in, a gold ◆ result banner lands.
- Authoring is pure content:
src/terminal/sessions.tsholds oneSessionper skill — an array of typed events (inprompt /sysdim line /runspinner-then-✓ /outstreamed lines /donebanner). The engine (src/terminal/TerminalPromo.tsx+EventLine.tsx+types.ts) derives ALL timing from the events and sets duration via the composition's computeddurationInFrames— no per-event frames in the data. Keep a session ≲ 16 visible lines (the window height) so it doesn't need to scroll. Genericize per the rules (proxy usestw.example.com/<你的域名>/<token>); avoid Unicode that isn't in the loaded fonts (e.g. Yijing hexagram glyphs render as tofu — use names). - Render:
cd remotion && npm install && node scripts/render-all.mjs [slug…]→out/<slug>.mp4. Preview:npx remotion studio. Sanity-check a frame:npx remotion still src/index.ts <slug> out.png --frame=175. - Host:
node scripts/upload-r2.mjs [slug…]→wrangler r2 object put ax-blog-media/ax-skills/<slug>.mp4(wrangler is already logged in for ax-blog). Served athttps://media.ax0x.ai/ax-skills/<slug>.mp4(range-request enabled → seekable). - Embed: landing JS injects
▶ 影片per card (slug from the demo link); each demo page carries the shared#promo-linklightbox snippet (slug fromlocation.pathname, robust to/<slug>/and/<slug>/index.html). Both stream from R2 — no big files in the repo. Re-rendering a skill replaces the mp4 at the same URL; no showcase edit needed. - The Remotion skill itself (
remotion-best-practices) is installed in~/.claude/skills/(fromremotion-dev/skills), NOT vendored here — it's a third-party skill, not an AX skill.
Rules (non-negotiable)
- Public repo → no real infra secrets. Genericize server/proxy node IPs, real node domains, tokens, subscription URLs to placeholders (
tw.example.com,<你的域名>,<token>, role labels like "US 母机"/"TW 节点"). Real product domains in源自/配套links are fine (panpanmao.ai, shichuan.ax0x.ai, blog.ax0x.ai). Grep new decks/skills for known real identifiers → must be 0. - No "Roadshow" / "路演" / "Funskills" / "Funskills Roadshow" as current terminology. Those are legacy naming bugs. These are AX Skills demos. Keep legacy filenames only when renaming would break an existing public URL.
- Catalog priority: feature and order AX-original skills and methods first, especially those distilled from shipped products or repeated internal use. Keep explicitly adapted/ported skills discoverable with attribution, but place them after original work in Featured lists, landing pages, and demo inventories unless the user deliberately overrides that order.
- Visual-verify demos — grep/validator alone misses layout bugs (invisible labels, footnote-vs-pager collisions). Open it.
- Pages CDN (Fastly) caches hard. After a push the new path can 404 / serve stale for ~1–2 min;
?cb=does NOT bust it — just retry. Build status:gh api repos/xingfanxia/AX-skills/pages/builds.
Style brief (image-set skills convention)
Skills that batch-generate image sets (gpt-image, gemini-image, and any skill built on them) define a style brief once — 风格名 + palette / typography / composition key params in a few sentences — and inject it verbatim into every prompt of the set; only content parameters vary per image. Cover-first: generate the cover to lock the style, verify one content image against it, then batch the rest. Details (incl. the 中文文字排版纪律 verbatim 禁止项 suffix): gpt-image/SKILL.md / gemini-image/SKILL.md.
Current demos
banxian-skill · jewelry-marketing · game-script-creation · proxy-node-setup · serenity-bottleneck-research · web-novel-writing · dr-sharp · trident — all 沙丘 Dune Style A, each with an R2-hosted terminal demo (a scripted Claude-Code session grounded in the skill's real SKILL.md).
dr-sharp+trident(added 2026-06-14) are improved ports of 秒秒Guo (mmguo.dev) prompts — credit the source in the demo/card (改进自 秒秒Guo), keep them after AX-original work in catalog ordering, and do not present them as flagship original methods.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

