Governed Agent Teams: subagent personas, authority maps and handoffs

Write a Desk Authority Map for these agents: [list with jobs].
Columns: agent, job, does alone, asks first, never does, evidence it owes, checked by, reports to. Every action under "does alone" must be undoable.

If the checked by column names the same agent, the row is wrong.

Read HANDOFF.md. Check out the branch, confirm the commit, run the tests and compare to the packet.
Report MATCH or MISMATCH. Continue only on MATCH.

The receiver checks before it writes a line.

# Governed Agent Teams: subagent personas, authority maps and handoffs

A PaperChaseWebb pack, version 1.0. Written for Claude Code subagents, Codex agents and Hermes Agent. Plain Markdown, so the charters, maps and packets also work with any agent that reads an instruction file.

Most agent team setups are a list of tools with job titles. This pack adds the part those charts leave blank: what each agent may do alone, what needs your yes, what proof it owes when it says "done", and how work moves from one agent to the next without losing the thread. It comes from running our own team of agents every day and from six public breakdowns of agent team setups, all credited by link in `10-sources.md`. Everything here is written in our own words.

## Files

| File | What it is for |
|---|---|
| 01-three-parts.md | An agent is instructions, a model and reach. Which part you own, which part you rent, and why that decides what survives a model change |
| 02-risk-classes.md | Sort any agent role into one of three risk classes before you build it, with the gate each class needs |
| 03-authority-map.md | The Desk Authority Map: per agent, what it does alone, what it asks for, what it never does, and what evidence it owes |
| 04-subagent-personas.md | 7 ready-to-paste personas in Claude Code subagent format, with matching Codex AGENTS.md sections and Hermes skill notes |
| 05-handoff-packet.md | The handoff packet: goal, tree state, attempts, tests, next step, risks. The receiver verifies before it continues |
| 06-evidence-standard.md | What counts as proof for each class of agent work, and who checks it |
| 07-decision-ledger.md | One ledger for every open decision. Silence is not a yes |
| 08-reach-inventory.md | Map every connector and tool as necessary, convenient or leftover, and cut the leftovers |
| 09-tests-and-prompts.md | 12 test scenarios and 14 prompts for writing and auditing a governed team |
| 10-sources.md | Every source we used, linked and credited |