# Agent Skills

Personal skills for AI coding agents, following the [Agent Skills](https://agentskills.io/) format.

## Install locally

```sh
bunx skills add /path/to/agent-skills
```

## Install from GitHub

```sh
bunx skills add lunarmoon26/agent-skills
```

## Executable tools

The repository is also an [Agent Plugins 1.0](https://agent-plugins.org/) package. Its `photo_process`, `photocraft`, and `vectorcraft` MCP tools use [`@lunarmoon26/agent-skill-runtime`](https://github.com/lunarmoon26/agent-skill-runtime) with the manifests in `skills/photo-processing/` and `skills/vector-illustration/`.
`photo_process` is limited to batch resizing/ICO export, luminance-to-alpha, and sprite extraction; inspection and general image editing use `photocraft`. Removed Pillow commands and their migration paths are documented in `skills/photo-processing/references/photocraft.md`.
The MCP bridge requires Node.js 22.20 or newer; executable tools use `uv`
and Python 3.11 through 3.14. Native editing additionally requires the corresponding `photocraft-cli` or `vectorcraft-cli` installed on PATH; runtime preparation does not install these binaries.

Prepare the exact PEP 723 dependency once before enabling the MCP server:

```sh
npx @lunarmoon26/agent-skill-runtime@0.1.1 prepare \
  --root . \
  --manifest skills/photo-processing/skill-runtime.json \
  --manifest skills/vector-illustration/skill-runtime.json
```

Normal tool calls run `uv` offline. Agent Plugins clients read `mcp.json`; clients without Agent Plugins support can launch the same stdio server directly with the command declared there.

Runtime manifests complement the Agent Skills format rather than replacing it.
Only deterministic executable tools, currently `photo-processing` and `vector-illustration`, declare a
`skill-runtime.json`. The reasoning and writing skills remain portable
instruction bundles. Their `evals/` directories are development fixtures using
the layout documented by the official Agent Skills evaluation guide.

## Included skills

- `document-driven-development`
- `lightweight-architecture`
- `llm-wiki-review`
- `photo-processing`
- `vector-illustration`
- `research-driven-development`
- `tech-breakdown-in-chinese`
- `privacy-guard` — local privacy checks and approved redaction for code, articles, blogs, and forum payloads; optional offline Rampart detector

Each skill lives in `skills/<skill-name>/`, with scripts, references, tests, or eval fixtures where needed.