pilotso11-skills

pilotso11-skills

Personal collection of Claude Code skills authored by @pilotso11, packaged as a single installable plugin.

Installing in Claude Code

One-line install (recommended)

From within Claude Code, run the slash command:

/plugin install pilotso11-skills@https://github.com/pilotso11/pilotso11-skills

Claude Code will fetch the latest main, register the plugin, and load every skill under skills/.

Local development install

If you want to hack on a skill locally, clone the repo and launch Claude with --plugin-dir pointing at the checkout:

git clone https://github.com/pilotso11/pilotso11-skills.git
cd pilotso11-skills
claude --plugin-dir .

Claude will load the plugin for that session only, picking up local edits to any SKILL.md without reinstalling.

Available skills

Skill	Trigger	What it does
hello	User greets Claude	Confirms the plugin is loaded and active (smoke test)
screen-decomposition	Before building or revising a screen against a locked sample plate	Reads the plate and produces a buildable spec — every visual element decomposed into its constructor call, layout coords, typography role, state binding, and asset source. The before half of the visual-fidelity loop.
screen-comparison	After porting a screen, or "how does this compare to the sample"	Granular side-by-side audit of a rendered screen vs its locked sample plate. Inventories every visual element, marks plate-vs-port match per property (rows are property-level so the output feeds screen-tweaking directly), and produces a categorised score on the full A+→F grade scale + prioritised fix list. The after half of the loop.
screen-tweaking	After screen-comparison returns < A-, or "fix the screen"	Precision repair loop: fixes the screen one property at a time, verifying each change against the plate before moving on. Prevents the failure mode of changing many things at once and losing track of what helped vs regressed.

Publishing a new skill

1. Create the skill directory:

   mkdir -p skills/<skill-name>
   

2. Write skills/<skill-name>/SKILL.md with YAML frontmatter. name and description are required; the description is what Claude uses to decide when to invoke the skill, so be specific about triggering conditions.

   ---
   name: <skill-name>
   description: Use when ... (describe the triggering situation precisely)
   ---
   
   # Skill body
   
   Instructions for Claude when this skill is invoked.
   

3. Validate locally:

   claude plugin validate .
   

   Resolve any errors before opening a PR. Warnings are tolerated but worth addressing.
4. Add your skill to the table in this README under Available skills.
5. Open a pull request. CI will run the same claude plugin validate . check; it must pass before merge.
6. Merge. Once on main, anyone who installed the plugin will pick up the new skill on their next claude plugin update.

CI

Every push to main and every pull request runs claude plugin validate . inside a node:lts-slim container with the official @anthropic-ai/claude-code CLI installed. The job fails if .claude-plugin/plugin.json is missing or malformed, or if any skills/*/SKILL.md has invalid structure.

Branch protection on main requires this check to pass before merge.

License

MIT