🛠️ claude-skills

A curated collection of 5 core production-grade Agent Skills for Claude Code, Goose, and Berd.

Designed for context engineering & multi-agent orchestration — fanning out multi-model panels, offloading heavy coding to worker agents, managing decision UI, and keeping human judgment at high leverage.

---

🚀 Quick Install

Option A: 1-Command Install via npx / skillfish (Recommended)

Run a single command without cloning:

npx skillfish add ArchieeR/claude-skills --all

To install a specific skill (e.g., ask or team):

npx skillfish add ArchieeR/claude-skills ask

---

Option B: Clone & Symlink / Copy

Clone the repository and copy or symlink into your global ~/.claude/skills/ directory:

git clone https://github.com/ArchieeR/claude-skills.git ~/claude-skills
mkdir -p ~/.claude/skills

# Copy all 5 core skills
cp -R ~/claude-skills/skills/* ~/.claude/skills/

# OR symlink individual skills so `git pull` keeps them fresh:
ln -s ~/claude-skills/skills/ask ~/.claude/skills/ask

---

🧰 The 5 Core Skills

Skill	Trigger / Command	Description
ask	/ask	Converts fat text and sprawling sub-agent reports into interactive AskUserQuestion decision chips so you can steer in seconds.
counsel	/counsel <question>	Fires parallel consensus panels across Grok CLI (infra scouting) and Codex CLI (code review) to eliminate blind spots.
team	/team [a|b|c] <task>	Multi-model implement → verify → fix → commit loop with 3 staffing modes to offload heavy investigation, writing, and fixing to Codex workers.
recap	/recap, /table	Dense table-format recap of the current session: work done, status, decisions made, and open follow-up items.
adhd	/adhd	Shapes model output specifically for ADHD focus: leads with immediate action, numbers multi-step work, suppresses tangents, and gives exact time estimates.

---

🖼️ Visual Guides

1. /ask — Decision UI & Choice Chips

Converts sprawling context and fat sub-agent reports into clean AskUserQuestion decision chips with explicit trade-offs.

---

2. /counsel — Multi-Model Consensus Panels

Fires parallel panels across Grok CLI (for infrastructure scouting) and Codex CLI (for code review) to eliminate blind spots in ~35 seconds.

---

3. /team — Multi-Model Execution & Offloading

Offloads heavy investigation, writing, and fixing to Codex worker agents across 3 staffing modes (A, B, C) while enforcing strict verification gates ("Sonnet never leads").

---

4. /recap — Session State at a Glance

Dense, 3-table snapshot of the session: status of work done, decisions locked, and open follow-up items.

---

5. /adhd — Action-First Output Shaping

Shapes model output specifically for ADHD focus: leads with code/commands first, provides exact time estimates, numbers multi-step work, and eliminates prose fluff.

---

🎯 Key Design Principles

1. Model Agnostic & Multi-CLI: Skills referencing external CLIs (codex, agy, grok) degrade gracefully if those binaries aren't installed or authed.
2. Context Efficiency: Heavy transcript reading and deep searches are offloaded to background sub-agents so your primary conversation context stays lightweight and sharp.
3. Scrubbed & Portable: All skills are scrubbed of machine-specific paths, private tokens, and internal project names so they run out of the box on any system.

---

📄 License

Distributed under the MIT License.