🎴 pi-skill-deck

Two-pane categorized skill browser for Pi — replaces the flat alphabetical wall of 150+ skills with a navigable, categorized TUI overlay.

✨ Features

• Two-pane browser — sections on the left, skills + detail on the right
• 8 group-by modes — g cycles, G opens a picker: Category · Source · Framework · Creator · Location · Tag · Usage Tier · Flat
• Source attribution — every skill carries source (pi-agent / claude / codex / npm / agents-pack / project-local), framework (SuperClaude / Baoyu / Marketing pack / npm:<pkg> / …), and creator (baoyu / Anthropic / Corey Haines / Mario Zechner / Ben / …)
• Inline short summaries — each row shows skill name + a one-line summary (derived from the SKILL.md description) so you can scan without selecting
• Expanded DETAILS box — boxed pane shows category · usage · source · location · framework · creator · tags · full Description block · "What it does" body excerpt
• ⚠ Thin-body flag — surfaces skills whose SKILL.md body is missing / sparse so you know which ones still need an explanation written
• Per-skill tags — t (inline editor) or T (modal editor) to apply free-form tags; can be used as a group-by mode
• ★ Top 10 most used — pinned at the top, ranked by frecency (frequency × recency)
• 📌 Bookmarks — Ctrl+B to save skills for quick access
• 💡 Daily suggestions — 1–3 underused skills that match your activity patterns, refreshed daily
• Search — press / to filter skills by name or description in real time
• Skill queueing — selected skill is injected alongside your next message
• Zero config — scans all standard skill locations automatically

📦 Installation

Requires: Pi ≥ 0.1 and Node.js ≥ 18.

The same pi install command works in any shell (bash, zsh, PowerShell, cmd.exe, fish, Git Bash, WSL):

Option 1 — from npm (recommended)

pi install pi-skill-deck

Pi resolves the package name against the npm registry, downloads it, and wires the /skills command automatically.

Option 2 — from GitHub directly

For the very latest commit (may be ahead of npm):

pi install CymatiStatic/pi-skill-deck

Option 3 — manual settings

If you prefer to edit ~/.pi/agent/settings.json yourself, add one of these to the packages array and restart Pi:

{
  "packages": [
    "npm:pi-skill-deck"
  ]
}

or

{
  "packages": [
    "github:CymatiStatic/pi-skill-deck"
  ]
}

Verify the install

Shell	Command
bash / zsh / Git Bash / WSL	pi --version && pi extensions list | grep pi-skill-deck
PowerShell / pwsh	pi --version; pi extensions list | Select-String pi-skill-deck
cmd.exe	pi --version && pi extensions list | findstr pi-skill-deck

Then start a new Pi session and type /skills. You should see a one-line 🎴 Skill Deck: N skills loaded notice at session start.

Uninstall

pi uninstall pi-skill-deck

or remove the corresponding entry from ~/.pi/agent/settings.json packages array.

Updating

pi install pi-skill-deck@latest

or pull a specific version (e.g., pi install pi-skill-deck@0.1.1).

🚀 Usage

Type /skills inside any Pi session to open the browser.

Keyboard shortcuts (inside overlay)

Key hints are always visible in three places so you never have to remember them:

• Inline in the header — the current group-by mode shows [g]/[G] change [?] help right next to it
• Inline in the DETAILS box — the tags row shows [t] inline [T] modal
• In the footer status bar — every key is in a cyan [...] chip

And ? opens a full Keyboard reference panel from anywhere.

Key	Action
Tab / ← →	Switch focus between panes
↑ ↓	Navigate within the focused pane
Enter	Queue the selected skill for your next message
Ctrl+B	Toggle bookmark on highlighted skill
g	Cycle group-by mode (Category → Source → Framework → Creator → Location → Tag → Usage Tier → Flat)
G	Open group-by picker menu
t	Inline tag editor (replaces footer bar)
T (shift+t)	Modal tag editor (floating centered box)
/	Start search (filters by name or description)
?	Toggle the full keyboard reference panel
Esc	Close overlay (or exit search / tag editor / picker / help)
Backspace	Delete search / tag-editor character

Group-by modes

Mode	What it groups on
Category (default)	Semantic buckets (Design & UI, Marketing & GTM, Product Hunt, …)
Source	Where the skill is installed from — Pi Agent / Pi User / Claude / Codex / npm / Agents Pack / Project-local
Framework	Skill library — SuperClaude / Baoyu Skills / Marketing Sub-Library / ProductHunt Sub-Library / context-mode / npm:<pkg>
Creator	Author — baoyu / Anthropic / Corey Haines / yoanbernabeu / gingiris / Mario Zechner / CymatiStatic / Ben (local)
Location	Physical install path (~/.pi/agent/skills, ~/.claude/skills, ~/.agents/skills/marketing, etc.)
Tag	Your applied tags
Usage Tier	Power (10+ uses) · Active (3–9) · Tried (1–2) · Unused
Flat	Single sorted list, no grouping

The selected mode is persisted to ~/.pi/agent/skill-deck-prefs.json and restored next session.

Tag editor — two styles

v0.2.0 ships both editor styles simultaneously so you can pick what feels best:

• t — Inline · a text input replaces the footer hint bar. Fastest to type. Comma- or space-separated tags. ↵ saves, Esc cancels.
• T — Modal · a floating box centered over the right pane shows the skill name + input field. More visible / less ambiguous. Same input format and controls.

Tags are stored in ~/.pi/agent/skill-tags.json and survive restarts.

🔍 How it works

Skill scanning

Automatically discovers skills from all standard Pi skill locations:

Location	Format
~/.pi/agent/skills/	Recursive (SKILL.md)
~/.pi/skills/	Recursive
./.pi/skills/	Project-local
~/.codex/skills/	Codex-compatible
~/.claude/skills/	Claude-compatible
./.claude/skills/	Project-local Claude
~/.agents/skills/	Shared agent skills
npm global node_modules/*/skills/	Installed packages

Categorization

Skills are categorized using a multi-layer strategy (applied in order):

1. Parent directory — e.g., skills under marketing/ → Marketing & GTM
2. Name prefix — baoyu-* → Baoyu Tools, ph-* → Product Hunt, ctx-* → Context-Mode
3. Explicit overrides — hardcoded map for non-obvious skills
4. Description keywords — fallback keyword matching
5. "Other" — last resort

Source detection

Every skill is also tagged with { origin, location, framework, creator }:

1. npm package — path matches node_modules/<pkg>/skills/…, framework = npm:<pkg>, creator pulled from package.json author field
2. Known path root — ~/.pi/agent, ~/.claude, ~/.codex, ~/.agents, etc.
3. Agents Pack sub-library — ~/.agents/skills/marketing/ → Marketing Sub-Library (Corey Haines), producthunt/ → yoanbernabeu, oss-launch/ → gingiris
4. Name-prefix hint — baoyu-* → creator: 宝玉 (baoyu); ctx-* → creator: Mario Zechner / framework: context-mode
5. Fallback — framework: —, creator: —

Frecency tracking

Usage is tracked per-skill with a frecency score: count × 0.5^(age_days / 7). This means a skill used 10 times last week ranks higher than one used 50 times last month. The Top 10 section reflects this ranking.

Daily suggestions

Each day, 1–3 skills are suggested from categories you use but haven't fully explored. The algorithm picks underused skills from your most active categories — no AI needed, just simple heuristics.

📁 State files

All persisted in ~/.pi/agent/:

File	Purpose
skill-usage.json	Per-skill { count, lastUsedAt }
skill-bookmarks.json	Array of bookmarked skill names
skill-suggestion.json	{ date, picks: [{ name, reason }] }
skill-tags.json	Per-skill tag arrays ({ "baoyu-imagine": ["fav", "image"], … })
skill-deck-prefs.json	Group-by mode + last-used tag editor style

📖 DETAILS box — anatomy

┌─ DETAILS: baoyu-imagine ────────────────────────────────────┐
│ category: Baoyu Tools · used 3× · last: 2d ago             │
│ source: claude-user · ~/.claude/skills                     │
│ framework: Baoyu Skills    creator: 宝玉 (baoyu)             │
│ tags:     #fav #image                                      │
│ ───────────────────────────────────────────────────────────── │
│ Description:                                               │
│ AI image generation with OpenAI GPT Image 2, Azure         │
│ OpenAI, Google, OpenRouter, DashScope, …                   │
│ What it does:                                              │
│ Generates images from text prompts using multiple AI       │
│ providers. Supports reference images, aspect ratios, …     │
└───────────────────────────────────────────────────────────────┐

When a skill's SKILL.md body is empty or too sparse, the "What it does" section shows:

│ What it does:                                              │
│ ⚠ No body content in SKILL.md — frontmatter description    │
│   only.                                                    │

The ⚠ indicator also appears next to skills in the list, so you can spot at a glance which skills still need their bodies written.

⚙️ Configuration

Custom category overrides

Edit categories.ts to add your own categories or remap skills. The EXPLICIT_MAP object maps skill names to category labels.

🤝 Contributing

PRs welcome! If you have skills that don't categorize well, open an issue or add an entry to the EXPLICIT_MAP in categories.ts.

📜 Changelog

See CHANGELOG.md for the full version history.

🌐 Distribution

Channel	Identifier
npm	pi-skill-deck
GitHub Packages	@cymatistatic/pi-skill-deck
Source	github.com/CymatiStatic/pi-skill-deck
Releases	GitHub Releases

📄 License

MIT — built by @CymatiStatic