# Init > Initialize or update the project using the `specify` CLI (--refresh for templates, --update to upgrade CLI, --clear to reset status line). Do NOT search for speckit or spec-kit binaries. Use when this capability is needed. - Skill: `tomevault-io/init-4` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/init-4` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/init-4/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/init-4 --- You MUST complete ALL steps below. Do not stop after Step 1. ## Step 1: Run init script Run the command from `` in the `` system reminder. This is your first and only Bash call. Do not run anything else before it. - If output contains `NEED_INSTALL`: show output, STOP. - If output contains `ERROR`: show error, STOP. - If the command was run with `--refresh` or `--update` and output contains `RESTART_REQUIRED`: **SKIP Steps 2 and 3.** Templates and extensions were refreshed but the existing configuration (enabled extensions, auto-commit, permissions) is preserved. Go directly to Step 4 and report that the refresh completed. Tell the user to restart Claude Code. - If output contains `READY` or `RESTART_REQUIRED`: **do not summarize yet**, go to Step 2. ## Step 2: Ask about extensions, auto-commit, and permissions You MUST ask all 4 questions below in a SINGLE AskUserQuestion call. Do NOT split, merge, reword, or reorder them. Pass them exactly as specified: 1. (`multiSelect: true`, header: "Quality"): "Which quality & review extensions do you want to enable?" - "spex-gates": "Quality gates on speckit commands (review-spec, review-code, verification)" - "spex-deep-review": "Multi-perspective code review with autonomous fix loop (5 agents)" - "spex-teams": "Parallel implementation with spec guardian review via Agent Teams (experimental, requires: spex-gates)" 2. (`multiSelect: true`, header: "Workflow"): "Which workflow extensions do you want to enable?" - "spex-worktrees": "Git worktree isolation after speckit-specify (creates sibling worktree)" - "spex-collab": "Phase-split collaboration with REVIEWERS.md for team PRs" 3. (`multiSelect: true`, header: "Auto-commit"): "When should spec-kit auto-commit? Select all that apply." - "After spec": "Auto-commit after specify completes" - "After plan+tasks": "Auto-commit after plan and tasks complete" - "After implement": "Auto-commit after implement completes" 4. (`multiSelect: false`, header: "Permissions"): "How should spex commands handle permission prompts?" - "Standard (Recommended)": "Auto-approve spex plugin scripts (spex-init.sh, specify CLI)" - "YOLO": "Auto-approve everything + enable auto mode for unattended workflows (requires Max/Team/Enterprise)" - "None": "Confirm every spex command before execution" Then apply the selections: **Dependency resolution**: If the user selected `spex-teams` but NOT `spex-gates`, auto-enable `spex-gates` and warn: > **Note:** spex-gates was auto-enabled because spex-teams depends on it for spec guardian review. **Extensions**: Extensions are already installed and enabled by the init script. For any extensions the user did NOT select (after dependency resolution), disable them: ```bash # Disable unselected extensions (ignore errors if already in desired state) specify extension disable 2>/dev/null || true ``` If the user selected all extensions, no action needed (all are enabled by default after init). **Auto-commit**: Update `.specify/extensions/git/git-config.yml` based on the user's selections. Only modify if the file exists (git extension is installed by spec-kit core). Enable each selected option with `yq`. If the user selected nothing, no changes needed (default is already false). - **After spec**: `yq -i '.auto_commit.after_specify.enabled = true' .specify/extensions/git/git-config.yml` - **After plan+tasks**: `yq -i '.auto_commit.after_plan.enabled = true | .auto_commit.after_tasks.enabled = true' .specify/extensions/git/git-config.yml` - **After implement**: `yq -i '.auto_commit.after_implement.enabled = true' .specify/extensions/git/git-config.yml` Combine all selected options into a single `yq` call when multiple are chosen. **Permissions**: The `specify` CLI does not manage permissions. Instead, write permission allowlists directly to `.claude/settings.json` based on the user's choice. Use the EXACT allow arrays below (copy verbatim, do not modify or rephrase the permission strings): - **Standard**: ```json {"permissions": {"allow": ["Skill", "Bash(specify *)", "Bash(*spex-init.sh*)", "Bash(*spex-ship-statusline.sh*)"]}} ``` - **YOLO**: ```json {"permissions": {"defaultMode": "auto", "allow": ["Bash(*)", "Read(*)", "Edit(*)", "Write(*)", "WebFetch", "WebSearch", "Skill", "Bash(specify *)", "Bash(*spex-init.sh*)", "Bash(*spex-ship-statusline.sh*)"]}} ``` Note: `defaultMode: "auto"` enables the classifier-based auto mode that replaces `bypassPermissions`. Auto mode auto-approves routine operations (file edits, skill invocations, local commands) while blocking genuinely dangerous actions. It requires Max, Team, Enterprise, or API plan. The broad `allow` rules serve as fallback for environments where auto mode is unavailable. **MCP tools**: Claude Code does not support wildcard MCP server names in allow rules (e.g., `mcp__*__*` is invalid). To allow MCP tools, add server-specific entries like `mcp__playwright__*` or `mcp__slack__*` for each MCP server the project uses. Since available MCP servers vary by project, do NOT add MCP allow rules by default. The user can add them manually if needed. - **None**: Do not modify permissions (leave defaults) Use the existing project `.claude/settings.json` (create if missing). Merge permission entries without overwriting existing settings. ## Step 3: Detect companion plugins and seed memory Scan the available skills list from the system reminder for companion plugins: ### 3a: Superpowers plugin Check for any of these upstream superpowers skills: `test-driven-development`, `systematic-debugging`, `brainstorming`, `writing-plans`. These are skills from [obra/superpowers](https://github.com/obra/superpowers) that complement spex but are NOT bundled with it. - If **found**: record "superpowers" as a detected companion plugin in the Step 4 report. No further action needed. - If **not found** and the user enabled the `spex-gates` extension in Step 2: show a recommendation in the Step 4 report (see below). Do NOT block init or make it an error. ## Step 4: Report Summarize: extensions enabled, auto-commit level, permission level, and companion plugins detected. If Step 1 said RESTART_REQUIRED or Step 2 permissions said CHANGED, tell user to restart Claude Code. If superpowers companion plugin was NOT detected (Step 3b), append this to the report: > **Recommended companion:** The [Superpowers](https://github.com/obra/superpowers) plugin by Jesse Vincent adds TDD discipline and systematic debugging skills that complement spex's spec-first workflow. spex absorbs superpowers' quality gates and anti-rationalization patterns, but does not bundle these standalone skills: > - **test-driven-development**: strict RED-GREEN-REFACTOR, no production code without failing test > - **systematic-debugging**: 4-phase root cause analysis with defense-in-depth > > Install with: `/plugin install superpowers@claude-plugins-official` (or `claude plugin install superpowers@claude-plugins-official` from the terminal) --- > Source: [rhuss/cc-spex](https://github.com/rhuss/cc-spex) — distributed by [TomeVault](https://tomevault.io).