init-factory

Initialize the Microfactory

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).

1. Interview the user

Ask (with AskUserQuestion where options fit, free text otherwise):

1. Task backend: jira, github, or todo (a local TODO.md file).
2. Project identifier: Jira project key (e.g. MYPROJ), GitHub owner/repo, or the path to the TODO file.
3. Plan by default? If yes, every issue needs an approved plan before implementation (issues labeled skip-plan are exempt). If no, only issues labeled needs-plan go through planning.
4. Feature branches? If yes, implementation happens on 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).
5. Loop intervals: how often to look for work. Defaults: implementation every 20 minutes, planning every 10 minutes.
6. This session's role: planner or implementer (the other role runs in a second session).

2. Authenticate the backend

• jira: check 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.
• github: check gh auth status. If not authenticated, suggest ! gh auth login. Determine the username with gh api user --jq .login.
• todo: no authentication. Verify the TODO file exists; if not, offer to create it with a couple of example items (see the todo-tasks skill for the format).

Never write tokens or secrets anywhere — the backend CLIs store their own credentials.

3. Write the configuration file

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.

4. Start the loop

For the role the user chose for this session, invoke the loop skill:

• implementer: /loop <implement_interval> /microfactory:implement-next
• planner: /loop <plan_interval> /microfactory:plan-next

Then 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.