setup

CRITICAL EXECUTION RULES — READ FIRST

This command is a state machine. Once /setup is invoked, your behavior is constrained by the following rules. These rules override contextual judgment, helpfulness instincts, and any other instructions you would normally apply.

1. ADVANCE STATES IN ORDER. States execute in numeric order 0 through 11. You MUST NOT skip a state based on your judgment about what seems redundant. If a state's work is trivially empty (e.g., no medium-confidence items to confirm), enter the state, record the empty mutation, and transition immediately — but DO enter it.

2. NO BACKWARD ADVANCE. Once you leave a state, do not re-enter it within this /setup invocation. "Invocation" means one continuous run from INIT through DONE. Typing cancel setup ENDS the current invocation; re-running /setup later starts a NEW invocation, and entering the saved currentState is an allowed forward transition for the new invocation. If the user wants to correct a prior answer mid-run, tell them to finish this run and edit the output files afterward.

3. OFF-TOPIC INPUT DOES NOT ADVANCE. If the user's response to a state's prompt is not valid input for that state, you MUST re-render the current state's prompt with this prefix:

   "I'm in the middle of project setup (state: <STATE_NAME>). I'll help with that after setup completes. To cancel setup, type cancel setup."

   You MUST NOT answer the off-topic question. You MUST NOT run any tool to investigate the off-topic question.

4. CANCEL PRESERVES STATE. The cancel trigger is the case- insensitive regex /^(cancel|stop|abort)( setup)?[.!?\s]*$/i applied to the trimmed user reply. Accepts the base verb (optionally followed by "setup") plus trailing punctuation or whitespace — cancel, Cancel!, STOP SETUP., abort all match. Text beyond the matched prefix means "not a cancel" (cancel please is a question, not a cancel). On match, acknowledge (the state file is already saved after every mutation per rule #6), print:

   "Setup paused. Run /setup again to resume, or worclaude setup-state reset to discard."

   and exit. You MUST NOT write any output files on cancel.

5. SCOPED TOOL WHITELIST. Between SCAN ENTRY and WRITE ENTRY you may invoke ONLY these tools:

   • Shell: worclaude scan --path . (SCAN only)
   • Shell: worclaude setup-state show --path .
   • Shell: worclaude setup-state save --from-file .claude/cache/setup-state.draft.json --path . — the ONLY way state is persisted. Use Write to produce the draft JSON first, then invoke the CLI against that path. This avoids Claude Code's shell-interpolation safety layer that triggers on heredoc-with-variable-expansion patterns.
   • Shell: worclaude setup-state reset --path .
   • Shell: worclaude setup-state resume-info --path .
   • Read: .claude/cache/detection-report.json
   • Read: .claude/cache/setup-state.json
   • Write: .claude/cache/setup-state.draft.json (state-save staging only — overwritten each save).
   • Tool: AskUserQuestion permitted at:
     • INTERVIEW states, per the Interaction mode contract (selectable / multi-selectable / hybrid questions).
     • CONFIRM_MEDIUM when the per-item option count (candidates + 1 for "Other") is ≤ 4 — AskUserQuestion's own maxItems: 4 schema cap. When the count exceeds 4, fall back to the verbatim text-parse rendering defined in State 3 below. Not permitted at CONFIRM_HIGH: the detection list routinely has 12+ items in real projects, which exceeds the 4-option schema cap. CONFIRM_HIGH renders VERBATIM per rule #7.

   At WRITE state the whitelist RELAXES to additionally permit:

   • Read: each of the six target output files (CLAUDE.md, docs/spec/SPEC.md, three SKILL.md files under .claude/skills/, docs/spec/PROGRESS.md). Reads are ONLY to preserve user content that the merge does not overwrite. Missing files are treated as empty.
   • Read: .claude/workflow-meta.json for template-hash lookup.
   • Write: exactly those six files, with merged content per the per-file merge rules in the WRITE state section.