Setup

Writes and checks ~/.config/rabbit/profile.md. The format is described in ${CLAUDE_PLUGIN_ROOT}/docs/profile.md, workflow files in ${CLAUDE_PLUGIN_ROOT}/docs/workflow.md; read them only if you need a detail this file lacks.

Rules

• LANG: the profile's Chat language if set and not auto; otherwise the language of the user's last message; English if there is none. Send every fixed message from references/messages.md by its id, translated into LANG; keep Markdown, {{...}} values, paths, URLs, commands and keys exactly as they are.
• Ask questions from references/questions.md one at a time, with AskUserQuestion (load it with ToolSearch if deferred): one question per call, the question, an empty line and its purpose in one line, its Header and Options, the default first with (Recommended); the user types any other answer under Other. A question asked again (its answer did not work) uses the same tool and options, the reason first. Without that tool: one question per message with its purpose, example and default, then end the turn and wait; no list unless the question is a pick list.
• If the user already gave answers (in the request or earlier), use them and do not ask again. An empty reply or default means the default.
• PY = python3, or python when python3 is missing or is not Python 3 (the Windows Store alias). Expand ~ before writing with PY -I -c "import os;print(os.path.expanduser('~'))", or without Python $HOME (%USERPROFILE% on Windows). Never write a token or password.
• Never write to Jira. Only read: the two searches below.
• Run commands only as written here; do not install anything for the user.

1. Existing profile → doctor mode

If ~/.config/rabbit/profile.md exists, read it.

1. Check: Basics has Workspace, Repositories folder, Chat language, Document language; Tracker has Site and Tools; the folders exist; the Tracker tools are available; the Workflow: file, if set, passes the step 2 check. An empty Site or Tools: run step 4. An empty Sprint field: run step 4.3. A Tracker tool name the server no longer offers: replace it with the server's tool for the same job (step 4.1) and say so; Tracker is found, never asked. Write what these checks found or replaced at once (step 5, changed lines only). Send setup.doctor: a | Setting | Value | table of Workflow (or none), the Basics lines and the Tracker lines, then one line per check (✓ or ✗ and why).
2. Ask doctor-next. Cancel: say in one line that nothing else changed, and stop. Set up again: run steps 2–6; in each question the current value comes first with (Recommended), then the other options without it. Any other answer names what to change: change only that, then go to step 5.

No profile: send setup.welcome and start at step 2.

2. Workflow file

Ask workflow. On a path, run PY -I "${CLAUDE_PLUGIN_ROOT}/skills/task/scripts/workflow.py" <path>. If it can't run, read the file instead.

• Exit 3 or unreadable: ask workflow again with setup.workflow-missing as its reason. Never write the profile with a path that can't be read.
• Otherwise send setup.workflow-read with each part that has instructions, named as its heading (Environment when environment is not empty, Plan:After). If unknown is not empty, send setup.workflow-unknown with those headings; if placeholders is not empty, send setup.workflow-placeholders with them.

3. Basics

Ask workspace, repositories, chat-language, doc-language in turn.

• Workspace folder missing: ask setup.folder-create with AskUserQuestion; on Create it, mkdir -p it; else ask again.
• Repositories folder: count clones with find "<dir>" -maxdepth 3 -name .git | wc -l; send setup.repos-found and continue, or setup.repos-none and ask again (unless the user gave this folder in the request: then keep it and continue).

4. Tracker (found, not asked)

1. Tools. A Jira tool set is one server's tools for searching issues (a name ending in searchJiraIssuesUsingJql, or a Jira search tool) and reading an issue, plus its tool for reading remote links when it has one. Use ToolSearch if tools are deferred. None: write the profile with Tracker empty (step 5), send setup.no-jira-tools and stop. More than one set: ask tracker-tools once, listing the sets found.
2. Site. From the tools: getAccessibleAtlassianResources (its url) when the set has it; otherwise the site the server reports, or the host of the self link of the issue step 3's search returns (run it first). More than one site: ask site, listing them.
3. Sprint field. Search assignee = currentUser() AND sprint in openSprints() ORDER BY key, maxResults: 1, fields: ["*all"] (cloudId = Site when the tool takes one). The sprint field is the customfield_* whose value is a list of objects with id, name, state and boardId, or (Jira Server/Data Center) of strings such as …Sprint@x[id=12,rapidViewId=3,state=ACTIVE,name=…]. Found: send setup.sprint-field. No issue: send setup.sprint-field-missing, leave it empty and continue.

5. Write the profile

mkdir -p ~/.config/rabbit, then write ~/.config/rabbit/profile.md:

# Profile

Workflow: <path, only if given>

## Basics
- Workspace: <path>
- Repositories folder: <path>
- Chat language: <value>
- Document language: <value>

## Tracker
- Site: <url>
- Tools: <server> (<search tool>, <issue tool>[, <remote links tool>])
- Sprint field: <customfield_...>

In doctor mode, edit only the lines that change. Send setup.saved with the path.

6. Test run

If Site, Tools and Sprint field are set, run the active-sprint search once (fields: ["summary", "<sprint field>"], maxResults: 50) and send setup.try with the active sprint's name and the issue count, then setup.done; or setup.try-failed with the error.

Whenever setup ends or stops, if another Rabbit command started it, go back to that command.