A personal collection of cross-platform AI agent skills (SKILL.md
format). Each skill is self-contained — install the ones you want and skip the
rest. No per-platform plugin to maintain; the skills work anywhere the Agent
Skills standard is supported.
The shortest install path is the generic skills CLI
(npx skills add), which works across Claude Code, Cursor, Codex, and other
Agent-Skills runtimes; Hermes and OpenClaw have their own first-class
lanes below. Installation is runtime-specific; do not assume one installer
command works everywhere.
npx skills add tmchow/agent-skillsLaunches an interactive picker: choose any (or several) of this repo's skills and which of your agents to install them into (Claude Code, Cursor, Codex, …). To install one skill directly, name it:
npx skills add tmchow/agent-skills --skill clawpatchHermes installs by owner/repo/skill identifier (a direct SKILL.md URL
also works):
hermes skills install tmchow/agent-skills/camofox-cloaked-browser
hermes skills install tmchow/agent-skills/clawpatchFrom an interactive Hermes CLI session, use the slash command equivalent:
/skills install tmchow/agent-skills/camofox-cloaked-browser
/reload-skills
/skill camofox-cloaked-browser
Notes:
/skillsis the Hermes interactive slash command for search/install/inspect/manage./reload-skillsmakes newly installed skills visible in the current session./skill <name>loads an installed skill into the current session.
Install from ClawHub:
openclaw skills install camofox-cloaked-browserClawHub page: https://clawhub.ai/tmchow/camofox-cloaked-browser
Any Agent-Skills-compatible installer can consume this repo. When documenting another runtime, add it as a separate lane rather than replacing the existing instructions.
| Skill | What it does |
|---|---|
camofox-cloaked-browser |
Use Camofox/Camoufox selectively for cloaked browser automation — npm/npx startup, OpenClaw plugin tools, raw REST commands, session/tab workflow, environment variables, process-scoped Hermes CAMOFOX_URL, and snapshot-after-action discipline. |
clawpatch |
Drive the Clawpatch CLI for automated AI code review and per-finding fixes — scanner-only subagent dispatch (parallel) or full-cycle clawpatch fix loops, with wire-format JSON parsing, mode selection, and exit-code/pitfall handling. |
vhs-terminal-recorder |
Drive Charmbracelet VHS for reproducible terminal recordings — .tape authoring, deterministic hidden setup, validation-vs-render discipline, GIF/MP4/WebM outputs, theme selection, and publish-safety guardrails. |
yt-dlp-cli |
Drive yt-dlp for streaming-site media downloads, audio extraction, subtitle pulls, clips, and bounded playlist/channel archives, with format inspection, ffmpeg checks, output verification, and safe YouTube auth failure handling. |
Each skill lives in its own top-level directory with a SKILL.md (agent
instructions) and a README.md (human landing page — what it does,
prerequisites, install) at its root. Deep material can optionally go in a
references/ subdir, but skills should stay thin and defer to the wrapped
tool's own --help/output rather than re-document it:
agent-skills/
├── camofox-cloaked-browser/
│ ├── SKILL.md # agent-facing instructions
│ └── README.md # human landing page
├── clawpatch/
│ ├── SKILL.md # agent-facing instructions
│ └── README.md # human landing page
└── <next-skill>/
├── SKILL.md
└── README.md
Adding a skill? See AGENTS.md for the conventions (frontmatter
rules, the SKILL.md-vs-README split, validation).
Every skill declares the two required fields (name, description) plus
version, so it works across any Agent-Skills-compatible runtime. Unknown
frontmatter fields are ignored per the standard, which lets individual skills
add runtime-specific metadata without breaking others — for example,
clawpatch carries metadata.openclaw (npm install directive, optional env
vars) and metadata.hermes (tags, required toolset) blocks, while
camofox-cloaked-browser carries OpenClaw optional env-var declarations plus
Hermes skill config for the local Camofox base URL and explicitly scoped
Hermes-only CAMOFOX_URL guidance.
MIT © Trevin Chow