Licensed to prompt. A sandbox repo for Claude Code sub-agents — one field agent deployed so far.

---

Mission

007 is a playground for building and running Claude Code sub-agents — small, purpose-built agents with their own instructions and tool access, dispatched automatically based on what you ask for. The codename fits: each agent operates independently, reports back with receipts, and never blows its cover outside the task it was briefed for.

One agent is currently active in the field.

Agent Dossier

Codename	Objective	Tools	Status
matt-wood-fyi	Infiltrate mattwood.fyi — the public FYI knowledge graph of AWS Chief AI & Technology Officer Matt Wood — and report back on his current thinking, tensions, and reading	Bash	🟢 Active
classified	awaiting deployment	—	⚪ Standby

Briefing: matt-wood-fyi

Matt Wood runs mattwood.fyi as a live, public, agent-friendly feed of riffs, links, and essay pointers — no auth, no rate limits, and a REST API purpose-built for exactly this kind of querying (search, semantic, edges, summary).

The agent doesn't hit that API directly with raw curl calls — Windows shell quoting on query strings gets ugly fast. Instead it shells out to fyi_client.py, a small wrapper that URL-encodes arguments and returns pretty-printed JSON:

python fyi_client.py summary
python fyi_client.py search "KEYWORD"
python fyi_client.py semantic "natural language question"
python fyi_client.py edges "SHORT_ID"

Ask Claude Code naturally and it delegates based on the agent's description field — no explicit invocation syntax needed:

"What are the latest tensions or challenges Matt is looking at?"

How a mission runs

sequenceDiagram
    participant You
    participant Claude Code
    participant matt-wood-fyi agent
    participant fyi_client.py
    participant mattwood.fyi API

    You->>Claude Code: "What is Matt thinking about X?"
    Claude Code->>matt-wood-fyi agent: delegate (description match)
    matt-wood-fyi agent->>fyi_client.py: python fyi_client.py semantic "X"
    fyi_client.py->>mattwood.fyi API: GET /api/fyi/q/semantic/X
    mattwood.fyi API-->>fyi_client.py: JSON payload
    fyi_client.py-->>matt-wood-fyi agent: formatted JSON
    matt-wood-fyi agent-->>You: cited answer + permalinks

Loading

Quickstart

git clone <this-repo>
cd 007

Open the directory in Claude Code and just ask a question — the agent is already wired up:

"Use the matt-wood-fyi agent — what does Matt think about the middle class of software engineering?"

Note

Claude Code reads .claude/agents/ once at session start. If you edit matt_wood_fyi.md, restart your session before the changes take effect.

Deploying this agent to a different project

1. Create the agent directory:

   mkdir -p .claude/agents

2. Add the agent file — copy matt_wood_fyi.md into .claude/agents/ in your target project. The front matter (name / description / tools) must start on line 1 with ---, or Claude Code won't recognize it.
3. Add the helper script — copy fyi_client.py into your project root.
4. Start a new Claude Code session in that directory and ask naturally — Claude auto-delegates based on the agent's description field.

---

This repository will self-destruct in... never. It's just a sandbox.