edwardsnjd/subagent

Spawn yourself as a sub-agent via bash to perform a task: $@

Delegating to a subagent

When to use this skill

• Instructed to delegate
• Task has clear boundaries (e.g. "read these 3 files")
• Task needs isolation (e.g. "summarise this file")
• You can pass all required context explicitly
• You have a constrained, specific objective

Critical: Zero context

Subagents see nothing from your session. You must pass:

• Prior findings
• File paths to examine
• Explicit goals

Anti-patterns

• ❌ Pass file contents in prompt — Use @file references instead
• ❌ Delegate the delegation — Coordinate yourself instead
• ❌ Batch processing - Use iterative workflow instead

Size limits

Keep prompts under ~1KB:

• 1 specific task, not "process all items"
• File paths, not file contents

How to delegate

1. Break into tiny tasks (one at a time)
2. Write prompt with Goal, Files, Scope
3. Run pi --print with appropriate arguments (if the user specifies a model, use --provider and --model accordingly)
4. Check response for errors
5. On failure: retry once with simpler scope, or stop and give up

Examples

Find something

pi --print <<-'EOF'
Find all calls to fooBar() in the codebase.
Return a list grouped by file.
EOF

Write something

pi --print <<-'EOF'
Write an email validator function:
- Check format ([email protected])
- Return true/false
EOF

Investigate with context

pi --print <<-'EOF'
You are debugging the auth module.

**Goal:** Find why login fails for users with 2FA.

**Prior:** User reports "works without 2FA, fails with it".

**Files:**
- auth/login.js: main login logic
- auth/totp.js: 2FA handling

**Scope:** Focus on 2FA code path only.

**Output:** List of potential causes ranked by likelihood.
EOF