Pi Coding Specialist

Manage Pi Agent as sub-agent for complex coding tasks involving multiple files, extensive refactoring, or systemic debugging.

Security Scan

The skill's instructions align with its stated purpose (running a Pi sub-agent) but rely on an unverified global npm package, assume system paths, and instruct editing agent configs without installation provenance — these gaps merit caution.

Purpose & Capability

The name/description match the SKILL.md: it instructs how to spawn a Pi coding sub-agent, when to use it, and how to monitor it. Requiring a Pi CLI and enabling OpenClaw subagents is coherent with the stated purpose. Minor mismatch: it assumes a macOS-style binary path (/opt/homebrew/bin/pi) and references specific model names (qwen3.5:397b-cloud) without explaining credential requirements for cloud models.

Instruction Scope

The instructions tell the operator/agent to install a global npm package, run the Pi CLI, and read/modify ~/.openclaw/agents/*/config.json and other local files. Those filesystem and config modifications are within the general scope of enabling a sub-agent, but they are sensitive operations (editing agent configs, installing global packages) and are presented with no safety checks, version pinning, or provenance for the external package.

Install Mechanism

There is no formal install spec in the registry entry, but the SKILL.md recommends 'npm install -g @mariozechner/pi-coding-agent' — a global npm install from an unverified package author. That is a moderate-to-high-risk install pattern because it pulls and runs code from the public registry with no version pin, no checksum, and no trusted source indicated. The instructions also assume the existence of a 'pi' binary at a specific path without cross-platform checks.

Credentials

The skill requests no environment variables or credentials in metadata (good). However, it references cloud model identifiers and spawning cloud-backed sessions (e.g., qwen3.5:397b-cloud) which implicitly require platform/model credentials managed elsewhere; the skill does not request or document these, which is a gap but not necessarily malicious.

Persistence & Privilege

always:false and normal invocation behavior are fine. However, the instructions advise enabling 'subagents' in ~/.openclaw/agents/*/config.json (potentially across multiple agents), which writes user config and could affect other agents. Changing other agents' configs without clear scoping or rollback instructions increases risk.

What to consider before installing

This skill looks like a legitimate orchestrator for running a Pi coding sub-agent, but exercise caution before using it: 1) Do not run the suggested 'npm install -g @mariozechner/pi-coding-agent' on production systems until you can verify the package author, package contents, and a pinned version/checksum (install in a sandbox first). 2) Back up ~/.openclaw/agents/*/config.json before changing subagents.enabled and prefer editing only the specific agent you intend to modify. 3) Confirm the origin and integrity of the 'pi' CLI binary and whether the path /opt/homebrew/bin/pi is appropriate for your OS. 4) Expect that cloud model usage will rely on external credentials managed by your platform — verify those separately. If the package author, source repository, signed release, or an official OpenClaw install spec is provided, re-evaluate (that information would move this toward benign).

PI-CODING-SPECIALIST SKILL

Name: pi-coding-specialist
Description: Orchestrate Pi Agent as coding specialist for complex tasks (>5 files)
Trigger: Coding tasks requiring multi-file changes, refactoring, or complex implementations

---

🎯 MISIÓN

Integrar Pi Agent como sub-agent preferido para tareas de coding complejas desde OpenClaw.

---

🔧 CONFIGURACIÓN

Model Routing Integration

{
  "subagents": {
    "enabled": true,
    "coding_specialist": "pi",
    "backend": "pi-subagents"
  }
}

Trigger Conditions

Usar Pi cuando:

• ✅ Coding task >5 archivos
• ✅ Refactoring complejo
• ✅ Nueva feature multi-file
• ✅ Code review profundo
• ✅ Debugging sistémico

NO usar Pi cuando:

• ❌ One-liner fixes
• ❌ Simple file edits
• ❌ Lectura de código

---

🛠️ COMANDO DE SPAWN

# Spawn Pi como sub-agent
openclaw sessions spawn \
  --name="pi-coding-task" \
  --model="qwen3.5:397b-cloud" \
  --workdir="/path/to/project" \
  --prompt="Pi, implementa [task description]. Usa pi CLI para coding."

Pi CLI Direct Execution

# Install (si no existe)
npm install -g @mariozechner/pi-coding-agent

# Execute con PTY
bash pty:true workdir:/path/to/project command:"pi 'Task description'"

# Background mode para tareas largas
bash pty:true workdir:/path/to/project background:true command:"pi 'Complex implementation'"

---

📊 HANDOFF PATTERNS

C-Level → Pi Handoff

1. Nexus/Forge detectan tarea compleja (>5 files)
2. Spawn Pi sub-agent con contexto específico
3. Monitor progreso via process:log
4. Receive results push-based
5. Integrate changes al proyecto principal

Example Workflow

# 1. Detectar necesidad
if files_changed > 5; then
  # 2. Spawn Pi
  bash pty:true workdir:$PROJECT background:true command:"pi '$TASK'"
  
  # 3. Monitor
  process action:log sessionId:$ID
  
  # 4. Complete
  # Pi auto-notifies on completion
fi

---

🔗 INTEGRATION POINTS

Con C-Levels

Agent	Handoff Scenario
Nexus (CTO)	Architecture implementations
Forge (CDO)	Feature development
Spark (CAO)	Complex automations
Hybrid Architect	System integrations

Con Model Routing

• Simple tasks: qwen3.5:9b (inline)
• Medium tasks: hermes3:8b (inline)
• Complex coding: Pi Agent (sub-agent)
• Complex strategy: qwen3.5:397b-cloud (inline)

---

📁 FILES

• ~/.openclaw/skills/pi-coding-specialist/SKILL.md (este archivo)
• ~/.openclaw/agents/*/config.json (subagents.enabled: true)
• /opt/homebrew/bin/pi (Pi CLI binary)

---

✅ VERIFICATION

# Check Pi installed
which pi

# Test Pi execution
pi -p "Hello world"

# Verify subagent config
cat ~/.openclaw/agents/nexus/agent/config.json | grep -A3 subagents

---

Version: 1.0
Status: ACTIVE
Created: 2026-03-31