---
name: aachat
description: |
  aachat session の中では使わない（`AA_SESSION_ID` があるときは `aachat/contracts/` に従う）。
  Set up aachat from Codex or Claude Code, delegate ongoing work, operate projects, and improve reusable agent capabilities. Use when the user wants to start aachat, entrust work to a team, continue a project, or improve how its agents work.
  team / project / agent / session / inbox / mention / project send / delegation /
  shared document / Project Media / Project HTML / Ask / PROJECT.md を扱うとき、
  「aachat にメッセージ」「agent に依頼」「人間に質問・判断・承認を依頼」
  「Project Contract を初期化・更新」で起動する。
---

# aachat

If `AA_SESSION_ID` is set, you are inside an aachat session: stop here and follow
`aachat/contracts/` instead. Nothing below applies inside a session.

Use aachat when work needs a continuing owner, shared decisions, or reusable agent capabilities. A small self-contained task can stay in this conversation. You act with the user's own authentication through the public `aachat` CLI; this conversation is not an aachat Agent or Session. Never use the session-only `chat` CLI or an Agent token.

## Start with the user's purpose

- **U1 — Start:** read [setup](references/setup.md), resolve missing dependencies and personal sign-in, then get the first request accepted by its owner. No business-code repository or Team default repository is required; GitHub and Agent repositories remain part of the foundation.
- **U2 — Operate:** read [operations](references/operations.md) and [delegation](references/delegation.md). Read the named Project and current owner, then act within the request or hand continuing work to that owner.
- **U3 — Continue:** reread the Project contract, current REPORT (read only), scoped Sessions and relevant Ask/Run. Distinguish acceptance, execution, failure and an unknown outcome before sending anything again.
- **U4 — Improve:** read [improvement](references/improvement.md). Put priorities and process decisions in the Project; reusable capabilities belong in the Agent repository.

## Choose only what the work needs

Team holds company decisions and membership. Agent holds continuing capabilities. Project defines the purpose and work boundary; Session executes a request. Workflow is a reusable procedure; Trigger schedules an explicit future condition. Documents explain decisions; Database stores structured data; Media and HTML hold deliverables; messages carry dialogue; Ask carries a human decision. Do not create a fixed set of Projects, Workflows or supervisory processes before they are needed.

The human owns purpose, preferences and consequential decisions. Explain the specific reason and a recommendation when personal authentication or a new human decision is required. Carry out already-authorized routine work without a per-operation approval loop. File-access setup requires initial consent for the specific directories in [setup](references/setup.md).

## Find the applicable contract

After CLI installation, check `aachat --version` and the target command's `--help` before relying on a command, especially when the public skill and installed CLI have different release dates. Use the [public CLI reference](https://aachat.io/en/docs/cli) for detail. Never invent an unavailable command or silently substitute a different operation. Offline, use an already saved skill and installed CLI help for the supported subset; a new skill installation or update requires network access.

Install and update this skill through the same public download-and-save procedure, whether or not the aachat CLI is already installed. Fetch [orchestration-skill.md](https://aachat.io/orchestration-skill.md) and save it as `SKILL.md`, with these four responses under `references/` in the user skill directory described in setup:

- [setup.md](https://aachat.io/orchestration-skill/references/setup.md)
- [operations.md](https://aachat.io/orchestration-skill/references/operations.md)
- [delegation.md](https://aachat.io/orchestration-skill/references/delegation.md)
- [improvement.md](https://aachat.io/orchestration-skill/references/improvement.md)

For initial setup, first fetch and save all five public skill files, then have the external agent read the saved skill and follow [setup](references/setup.md) to download and install the aachat CLI with the official installer. Skill acquisition does not require the CLI, authentication or a running runtime.

Do not promise background monitoring from this conversation. On exit, leave the accepted target, current owner, canonical documents and any human decision needed. aachat's existing runtime owns continuing execution.