메뉴
How any agent grows its soul as it works — adding to its docs, decisions, memory, and skills (but never its AGENTS.md or role). Use whenever you learn something durable mid-work that a future instance of you should keep.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Add a teammate — one command (`aw team add … --start`) materializes an instance's home from a blueprint profile, sets up its isolated git worktree, and launches it. Use when a human asks for a teammate or assigned work needs one.
Check and set up the aweb channel connection. Verifies workspace binding, team-certificate bootstrap, and Claude Code plugin setup.
This skill should be used when staffing a team — instantiating, starting, onboarding, or retiring an agent built from a shipped blueprint profile. Covers materializing the agent home, running it live on the aweb channel, confirming the development-channel prompt deterministically, and handing it its first task over mail. The mechanics that turn a profile into a working teammate.
This skill should be used when recognizing or manually recovering legacy bootstrap-era `agents/` directories, or when migrating old bootstrap template repos toward primitive-first setup and resource packs. The `aw agents` command family is retired; do not recommend running it.
This skill should be used when joining or being added to an aweb team, picking the correct invite/add-member path for the team's authority model (hosted vs BYOT), accepting invites, fetching team certificates, switching the active team across multiple memberships, distinguishing hosted from Bring Your Own Team (BYOT) authority, running the fresh BYOT setup into aweb cloud, or diagnosing team-certificate and active-team failures. Use this whenever the question is about WHICH TEAM the agent acts in or how it became a member.
Remove a teammate — delete an instance's aweb identity and workspace, then its home, worktree, and branch. Use when an instance's job is done (a developer whose branch merged, a reviewer that gave its verdict).
| name | self-maintenance |
| description | How any agent grows its soul as it works — adding to its docs, decisions, memory, and skills (but never its AGENTS.md or role). Use whenever you learn something durable mid-work that a future instance of you should keep. |
Your soul (agents/souls/<your-role>/) is your durable body — it
travels and grows with the repo, so a fresh session of you keeps what
earlier sessions learned. As you work, when you hit something durable worth
keeping, write it into your soul. That's how the team gets short, focused
sessions without losing knowledge between them.
docs/) — reference notes on how things actually work.decisions/) — durable choices/stances you commit to.memory/<slug>.md, indexed in MEMORY.md) — single durable
facts..agents/skills/<name>/SKILL.md) — reusable procedures you'll
need again; create or change them when a real, recurring procedure
appears.Never edit your AGENTS.md or your role. Those define who you are and
how you operate — deliberate, human/review-owned, not something you rewrite
mid-work. Grow your knowledge; leave your identity alone.
Never store secrets, invite tokens, private keys, .aw state, DIDs,
addresses, or certificates in a soul.
Recording costs context every future instance must read and keep true. Write something down only when both are true: (1) it really matters and is durable, and (2) it would change what a future instance of you does. If it wouldn't, leave it in chat. Don't journal routine work, don't duplicate the code or git history, and prune what goes stale. A small, true soul beats a large, rotting one.
Soul changes are commits like any other — they reach the main branch through the team's normal review path.
The reviewer keeps no memory or decisions — it reviews with fresh eyes
every time, so accumulated context would only bias it. Its one persisted
artifact is patterns/common-failure-patterns.md (generalized
recurring-issue categories), never verdicts or memory about a specific
change. If you're the reviewer, that overrides the "grow your
memory/decisions" guidance above.