메뉴
Pick and apply the best ccteams agent team for a described need using natural language — reads team metadata, selects the closest match, and applies it.
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
Frontier-grade engineering working method — goal compression, ground truth before opinion, hypothesis discipline, execution as evidence, honest reporting, exit checklist. Read in full at the start of any non-trivial task; inject the digest into every delegation prompt.
Operational playbook for systematic debugging — hypothesis ledgers, discriminating experiments, mechanism sentences, minimal fixes, sibling hunts. Read in full at the start of every bug investigation.
Operational playbook for Django + DRF — settings/version detection, ORM loading strategy, migration discipline, serializer field hygiene, timezone rules, and exact verification commands. Read in full at the start of every task on this team.
Operational playbook for framework-agnostic UI/UX/a11y — semantic-element-first markup, keyboard/focus verification, responsive layout checks, reuse of existing tokens over new CSS. Read in full at the start of every task on this team.
Operational playbook for stack-agnostic feature work — fixed-order stack detection, vertical-slice delivery, a catalog of unfamiliar-codebase mistakes (phantom utils, pattern drift, generated files, invented commands), and verification using only the project's own commands. Read in full at the start of every task on this team.
Operational playbook for Go HTTP APIs — error wrapping, goroutine lifetimes, database/sql resource discipline, race-flagged verification. Read in full at the start of every task on this team.
| name | choose-team |
| description | Pick and apply the best ccteams agent team for a described need using natural language — reads team metadata, selects the closest match, and applies it. |
| argument-hint | ["natural language request"] |
| allowed-tools | Bash |
Select and apply the agent team that best matches the user's natural-language request. Follow these steps exactly:
Run ccteams list --json via Bash. This returns a JSON array where each element
has name, description, tags, and requiresAgentTeams.
Read the user's request and compare it against each team's description and
tags. Identify the ONE team that is the best fit.
If NO team is a plausible match for the request, do NOT force a pick. Present the available teams and ask the user to choose directly — do not apply any team without a confident match.
If a good match is found, tell the user which team you chose and give a single sentence explaining why it fits their request.
Decide whether to pass --agent-teams:
Pass --agent-teams if the user's request contains any of these phrases or clear
synonyms: "in agent-teams mode", "as a team", "have them work in parallel",
"collaborative", "team mode", "teammates", "member messaging", "parallel teammates".
Also pass it if the chosen team has requiresAgentTeams: true in the JSON.
Otherwise, run the plain form.
--agent-teams enables Claude Code's experimental agent-teams feature
(CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 in .claude/settings.json), which allows
teammates to message each other directly. It is optional and opt-in — only set it
when the user asked for it or the team itself requires it.
Run the appropriate command via Bash:
ccteams use <chosen-team-name> --agent-teamsccteams use <chosen-team-name>Relay ccteams's full output to the user — do not truncate or paraphrase it.
Explicitly repeat the session-restart instruction: agents load at session
start only; the switch is NOT instantly active. The user must run /exit
then claude to activate the new team.