AGENTS.md · diff
git:20260329.893209f to git:20260330.5da680d
40 added, 32 removed. Audit A to A.
- # Agent Guidelines
+ # iOS Skills Collection
- This repo contains ASO & App Marketing skills following the [Agent Skills](https://agentskills.io) open standard.
+ 210 iOS/Swift/Xcode agent skills from 28 authors with phase-aware routing.
## Structure
```
- skills/<name>/SKILL.md — Skill instructions (required, <500 lines)
- skills/<name>/references/ — Detailed docs loaded on demand (optional)
- tools/integrations/ — API integration guides (Appeeky, Firebase, etc.)
- tools/REGISTRY.md — Capability matrix: which skill uses which tool
- ```
-
- Compatible directories: `.cursor/skills/`, `.agents/skills/`, `.claude/skills/`, `.codex/skills/`
-
- ## SKILL.md Format
-
- ```yaml
- ---
- name: skill-name # Must match directory, lowercase+hyphens, 1-64 chars
- description: ... # Trigger phrases + scope boundaries, 1-1024 chars
- metadata:
- version: 1.0.0
- ---
+ .claude-plugin/plugin.json — Claude Code plugin manifest
+ .codex-plugin/plugin.json — Codex CLI plugin manifest
+ hooks/
+ hooks.json — SessionStart + PreToolUse hook config
+ inject-router.mjs — Injects router skill at session start
+ route-skills.mjs — Pattern-matches files/commands → skills
+ skills/
+ _router/SKILL.md — Meta-skill: phase map, knowledge corrections, decision matrix
+ {author}--{name}/SKILL.md — Individual skills (210 total)
+ {author}--{name}/references/ — Detailed docs loaded on demand
+ skills-index.json — Routing rules + phase assignments
+ sources.json — Upstream repo mapping for sync
+ sync.sh — Pull latest from all upstream repos
+ install.sh — Install for Claude Code + Codex CLI
```
- Description drives discovery — agent reads all descriptions to decide which skill to load. Include "When the user wants to...", trigger keywords, and "For X, see Y skill" boundaries.
+ ## How Skills Load
- ## Writing Style
+ 1. **Session start**: Router injects with knowledge corrections table + decision matrix
+ 2. **File edit/read**: Hook matches file patterns → injects 1-3 relevant skills
+ 3. **Bash command**: Hook matches command patterns → injects relevant skills
+ 4. **Import detection**: Writing `import HealthKit` → injects dpearson2699--healthkit
+ 5. **Manual**: Agent reads any skill from `skills/{id}/SKILL.md`
- - Direct, actionable, second person
- - Tables for comparisons, numbered lists for steps
- - Bold for key terms, code blocks for examples
- - No filler, no hedging
+ Max 3 skills per injection. 24KB byte budget. Each skill fires once per session (dedup).
- ## App Store Reference
+ ## Priority
- **iOS limits:** Title 30 chars, Subtitle 30 chars, Keyword field 100 chars (comma-separated, no spaces), Description 4000 chars (not indexed for search), Promotional text 170 chars.
+ Paul Hudson (twostraws) skills are primary for all overlapping domains.
+ When multiple skills cover the same topic, use the one with highest priority in skills-index.json.
- **Google Play limits:** Title 30 chars, Short description 80 chars, Full description 4000 chars (indexed for search).
+ ## Skill Format
- **Key rules:** Title has highest keyword weight. Don't repeat keywords across fields. Use singular forms. Don't include "app" or category names in keyword field.
+ ```yaml
+ ---
+ name: skill-name
+ description: When to use this skill (1-1024 chars)
+ ---
- ## Commits
+ # Skill content (markdown)
+ ```
- `feat(skill-name): ...` / `fix(skill-name): ...` / `docs: ...`
+ ## Contributing
- Branches: `feature/skill-name`, `fix/skill-name-desc`, `docs/desc`
+ Each skill directory is prefixed with the source author. To add a skill from a new upstream repo:
+ 1. Add the repo to `sources.json`
+ 2. Run `./sync.sh` to pull the SKILL.md files
+ 3. Add routing rules to `skills-index.json` if needed
+ 4. Add to the appropriate phase in `skills-index.json`