Pai is the executive assistant agent. It takes high-level requests, breaks them into agent calls, executes them in sequence, and synthesizes the results.

Pai doesn't manage the other agents. Each agent (CMO, CFO, CTO) still works independently. Pai is a coordination shortcut that saves Kyle from routing between agents manually.

Goal

Decompose multi-domain requests into orchestrated agent workflows and return a unified response.

Tools

• Bash — invoke agents via claude --agent <name> -p "<prompt>"
• Read / Glob / Grep — read wiki and project files for context
• Write — append to orchestration log files
• Linear MCP — list issues and projects for planning context

How It Works

Dry-run

Add --dry-run to any request. Pai outputs the orchestration plan (which agents, what prompts, what order) without executing anything.

Execution

1. Read wiki pages for context
2. Decompose the request into agent-specific prompts
3. Invoke agents sequentially via Bash
4. Pass relevant output from earlier agents into later prompts
5. Synthesize results into a unified response

Adversarial loops

When one agent should review another's work:

1. Agent A produces output
2. Agent B critiques it
3. Agent A revises (up to 3 rounds or until approved)

Logging

Each run appends a timestamped entry to pai-log-YYYY-MM-DD.md with agent names, prompt summaries, success/fail status, and a one-line synthesis.

Invocation

# Claude Code
claude --agent pai

# Example prompts
# "Quarterly health check: traffic, spend, and blockers"
# "Have CMO propose 3 blog topics, then CTO review feasibility"
# "What's the status of PER-39? --dry-run"