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.

2026-07-0638

debug-playbook

toffyui/ccteams

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.

2026-07-0538

django-playbook

toffyui/ccteams

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.

2026-07-0538

frontend-playbook

toffyui/ccteams

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.

2026-07-0538

generalist-playbook

toffyui/ccteams

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.

2026-07-0538

go-api-playbook

toffyui/ccteams

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.

2026-07-0538

name	use-team
description	Apply a named ccteams agent team to the current project by running ccteams use <team-name>.
argument-hint	["team-name"]
allowed-tools	Bash

Apply the agent team specified by the user's argument to the current project.

Steps:

If no team name was provided, run ccteams list via Bash, show the output, and ask the user which team they want to apply before proceeding.
Decide whether to pass --agent-teams:
- Pass --agent-teams if the user's request uses any of these phrases or clear synonyms: "agent-teams mode", "collaborative", "team mode", "teammates working in parallel", "have them work together", "parallel teammates", "member messaging".
- Otherwise, run the plain form.
Run the appropriate command via Bash:
- With flag: ccteams use <team-name> --agent-teams
- Without flag: ccteams use <team-name>
--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. It is opt-in; you never add it unless the user asked.
Relay ccteams's full output to the user — do not truncate or paraphrase it.
Explicitly repeat the session-restart instruction from ccteams's output: agents load at session start only; the switch is NOT instantly active. The user must run /exit then claude to activate the new team.