菜单
Set up the microfactory for the current project. Interviews the user, writes .microfactory/config.yaml, authenticates the task backend (Jira via acli, GitHub via gh, or a local TODO.md), and starts the work loop. Use when the user wants to initialize, configure, or reconfigure the factory.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Run one planning iteration of the microfactory — claim the next planning-eligible issue from the configured backend, explore the codebase, write plans/<KEY>.md, push it, and put the issue up for plan review. Intended as a /loop target (e.g. /loop 10m /microfactory:plan-next); also takes an optional issue key to plan a specific issue.
GitHub Issues backend operations for the microfactory — find eligible issues, claim one with race verification, view, comment, and transition (statuses are mapped to labels), all via the gh CLI. Used by plan-next and implement-next when the configured backend is github. Not for general GitHub questions.
Run one implementation iteration of the microfactory — claim the next implementation-eligible issue from the configured backend, implement it (following the approved plan if one exists), run the tests, push (directly or via a feature branch + PR), and update the issue. Intended as a /loop target (e.g. /loop 20m /microfactory:implement-next); also takes an optional issue key to implement a specific issue.
Jira backend operations for the microfactory — find eligible issues, claim one with race verification, view, comment, and transition, all via the acli CLI. Used by plan-next and implement-next when the configured backend is jira. Not for general Jira questions.
Local TODO.md backend operations for the microfactory — find eligible items, claim, comment, and transition by editing checkbox states in a markdown file. Used by plan-next and implement-next when the configured backend is todo. Good for trying out the factory without Jira or GitHub.
| name | init-factory |
| description | Set up the microfactory for the current project. Interviews the user, writes .microfactory/config.yaml, authenticates the task backend (Jira via acli, GitHub via gh, or a local TODO.md), and starts the work loop. Use when the user wants to initialize, configure, or reconfigure the factory. |
Set up the current project so plan-next and implement-next can run continuously. All commands below are examples — adapt them to the user's OS and shell (Windows/macOS/Linux).
Ask (with AskUserQuestion where options fit, free text otherwise):
jira, github, or todo (a local TODO.md file).MYPROJ), GitHub owner/repo, or the path to the TODO file.skip-plan are exempt). If no, only issues labeled needs-plan go through planning.feature/<KEY> branches with a pull request (issues labeled skip-branch are exempt). If no, work lands on the default branch (issues labeled needs-branch still get a branch).acli jira auth status. If not authenticated, the user must log in interactively — suggest they run ! acli jira auth login (they will need their site, email, and an API token). Also ask for the Jira site host (e.g. mycompany.atlassian.net). Determine the user's Atlassian account id (visible in acli jira auth status output, or ask the user); it is needed for self-assignment.gh auth status. If not authenticated, suggest ! gh auth login. Determine the username with gh api user --jq .login.todo-tasks skill for the format).Never write tokens or secrets anywhere — the backend CLIs store their own credentials.
Create .microfactory/config.yaml in the project root (settings only, no secrets — safe to commit):
backend: jira # jira | github | todo
project: MYPROJ # Jira key | owner/repo | path to TODO.md
plan_by_default: false
feature_branches: false
implement_interval: 20m
plan_interval: 10m
# jira backend only:
jira_site: mycompany.atlassian.net
jira_account_id: "712020:abc..."
# github backend only:
github_username: octocat
Include only the keys relevant to the chosen backend. If a config file already exists, show it and ask before overwriting.
For the role the user chose for this session, invoke the loop skill:
/loop <implement_interval> /microfactory:implement-next/loop <plan_interval> /microfactory:plan-nextThen tell the user how to run the other role in a second Claude Code session in the same project, e.g.:
claude
/loop 10m /microfactory:plan-next
If planning is not used (no plan-by-default and no needs-plan issues expected), a single implementer session is enough — say so instead of prescribing two sessions.