Skip to content

CLI Reference ​

Complete reference for the Bear Craft command-line interface.


NPX Installation ​

The recommended way to install Bear Craft is via NPX:

bash
npx @the-bearded-bear/bear-craft [command] [options]

Available Commands ​

CommandDescription
installInstall Bear Craft to a project
initInitialize workflow (shows instructions)
flattenGenerate flattened codebase context
ralphRun Ralph Wiggum continuous loop
checkVerify bear-craft installation structure
listDetailed component inventory
doctorEnvironment diagnostics & health check
updateRefresh an installation: copy the marketplace of the current version, refresh rules, migrate copied files (ours removed, yours kept), re-register the plugins
syncRe-register .bear-craft/marketplace and re-install the plugins recorded in .bear-craft/install.json on this machine (after a clone or a move); idempotent, rules untouched
helpShow help message
(no command)Interactive installation wizard

Interactive Wizard ​

Run without arguments for the interactive wizard:

bash
npx @the-bearded-bear/bear-craft

The wizard will guide you through:

  1. Target directory - Where to install
  2. Technology stack - Which framework(s) to use
  3. Language - Documentation language (en, fr, es, de, pt)
  4. Options - Backup, force overwrite, etc.

Install Command ​

Basic Usage ​

bash
npx @the-bearded-bear/bear-craft install <target-directory> [options]

Options ​

OptionShortDescription
--tech=<technology>-tTechnology stack to install
--lang=<language>-lDocumentation language
--hooks=<mode>Hooks installed into .claude/hooks/ and merged into .claude/settings.json: default (network guard, dangerous commands, pre-commit secrets check, sensitive files, output hints, context re-injection after compaction), full (adds the Stop test gate, post-edit lint and Slack notification) or none; append ,memory (e.g. --hooks=default,memory, --hooks=full,memory) to also install the SQLite cross-session memory add-on into .claude/hooks/memory-lifecycle/ (needs sqlite3 and jq; the installer warns if they are missing). One profile is required, memory alone is rejected. Hook examples are copied to .claude/templates/hooks/ (not with none). Default: default. Existing hooks are never removed or overwritten
--force-fOverwrite existing files
--backup-bCreate backup before install
--dry-run-dSimulate without making changes
--preserve-configKeep existing CLAUDE.md

Technology Options ​

ValueDescription
symfonySymfony/PHP backend
flutterFlutter/Dart mobile
reactReact frontend
reactnativeReact Native mobile
pythonPython backend
angularAngular frontend
csharpC#/.NET backend
laravelLaravel/PHP backend
vuejsVue.js frontend
phpPHP Clean Architecture
commonCommon rules only
allAll technologies

Language Options ​

ValueLanguage
enEnglish (default)
frFrench
esSpanish
deGerman
ptPortuguese

Examples ​

bash
# Install Symfony rules in French
npx @the-bearded-bear/bear-craft install ~/my-project --tech=symfony --lang=fr

# Install multiple technologies
npx @the-bearded-bear/bear-craft install . --tech=react
npx @the-bearded-bear/bear-craft install . --tech=python

# Force reinstall with backup
npx @the-bearded-bear/bear-craft install ~/app --tech=flutter --force --backup

# Dry run to preview changes
npx @the-bearded-bear/bear-craft install . --tech=angular --dry-run

# Install all technologies
npx @the-bearded-bear/bear-craft install ~/project --tech=all --lang=en

Check Command ​

Verify the structure of a bear-craft installation.

Usage ​

bash
npx @the-bearded-bear/bear-craft check [target-directory]

Description ​

Scans the .claude/ directory and reports a summary of installed components: commands (by namespace), agents, references, skills, checklists, and templates.

Examples ​

bash
# Check current directory
npx @the-bearded-bear/bear-craft check

# Check a specific project
npx @the-bearded-bear/bear-craft check ~/my-project

List Command ​

Display a detailed inventory of all installed bear-craft components.

Usage ​

bash
npx @the-bearded-bear/bear-craft list [target-directory]

Description ​

Lists every installed component grouped by category (commands, agents, skills, references, checklists, templates) with counts and names. Useful for verifying what was installed and comparing across projects.

Examples ​

bash
# List components in current directory
npx @the-bearded-bear/bear-craft list

# List components in a project
npx @the-bearded-bear/bear-craft list ~/my-project

Doctor Command ​

Run environment diagnostics and installation health checks.

Usage ​

bash
npx @the-bearded-bear/bear-craft doctor [target-directory]

Description ​

Checks the development environment for required and recommended tools:

CheckStatus
Node.js >= 20[OK] / [FAIL]
npm[OK] / [FAIL]
Claude Code[OK] / [WARN]
git[OK] / [FAIL]
yq[OK] / [WARN]
.claude/ structure[OK] / [WARN]
Shell script permissions[OK] / [WARN]
i18n directories[OK] / [WARN]

Examples ​

bash
# Check current project
npx @the-bearded-bear/bear-craft doctor

# Check a specific project
npx @the-bearded-bear/bear-craft doctor ~/my-project

Update Command ​

Refresh an existing bear-craft installation by re-running install scripts.

Usage ​

bash
npx @the-bearded-bear/bear-craft update [target-directory] [options]

Options ​

OptionDescription
--tech=<technology>Specific technology to update (if omitted, auto-detects from installed references)
--lang=<language>Language override (default: en)

Description ​

Re-runs install scripts with --force to refresh all files. Common rules are always refreshed first, then tech-specific rules. If --tech is omitted, the command auto-detects installed technologies from .claude/references/.

Examples ​

bash
# Update all detected technologies
npx @the-bearded-bear/bear-craft update ~/my-project

# Update only React rules in French
npx @the-bearded-bear/bear-craft update ~/my-project --tech=react --lang=fr

# Update current directory
npx @the-bearded-bear/bear-craft update .

Flatten Command ​

Generate a flattened summary of your codebase for AI assistants.

Usage ​

bash
npx @the-bearded-bear/bear-craft flatten [options]

Options ​

OptionDescription
--output=<file>Output filename (default: CODEBASE.md)
--max-tokens=<n>Maximum tokens before sharding
--exclude=<patterns>Additional patterns to exclude

Examples ​

bash
# Generate flattened codebase
npx @the-bearded-bear/bear-craft flatten

# Custom output file
npx @the-bearded-bear/bear-craft flatten --output=CONTEXT.md

# Limit token count (enables sharding for large codebases)
npx @the-bearded-bear/bear-craft flatten --max-tokens=50000

# Exclude additional directories
npx @the-bearded-bear/bear-craft flatten --exclude="*.test.ts,*.spec.ts"

Output ​

The flatten command generates:

  • File tree structure
  • Priority-ordered file contents
  • Token estimation
  • Automatic sharding for large projects

Makefile Commands ​

When you clone the repository, you can use Make for installation.

Installation Commands ​

bash
# Install specific technology
make install-symfony TARGET=~/project
make install-flutter TARGET=~/project RULES_LANG=fr
make install-react TARGET=~/project OPTIONS="--force"

# Install presets
make install-all TARGET=~/project         # Everything
make install-common TARGET=~/project      # Common rules only
make install-web TARGET=~/project         # React
make install-backend TARGET=~/project     # Symfony + Python
make install-mobile TARGET=~/project      # Flutter + React Native

# Install tools
make install-tools                         # All tools
make install-statusline                    # Custom status line
make install-multiaccount                  # Multi-account manager
make install-projectconfig                 # Project config manager
make install-rtk                           # RTK token optimizer

Dry Run Commands ​

bash
make dry-run-all TARGET=~/project
make dry-run-symfony TARGET=~/project
make dry-run-flutter TARGET=~/project

Configuration Commands ​

bash
make config-list                           # List projects in YAML config
make config-validate                       # Validate YAML config
make config-install PROJECT=my-project     # Install from config
make config-install-all                    # Install all from config
make config-dry-run PROJECT=my-project     # Dry run from config

Utility Commands ​

bash
make help                                  # Show all available commands
make list                                  # List available components
make list-agents                           # List all agents
make list-commands                         # List all commands
make stats                                 # Show statistics
make tree                                  # Show project structure
make fix-permissions                       # Fix script permissions

Migration Commands ​

bash
make migrate-check                         # Check migration status

Plugin Export ​

bash
make plugin-export-all                     # marketplaces/<lang>/plugins/bear-<ns>/ (generated)
make plugin-check                          # marketplaces/ in sync with the installers

Direct Script Execution ​

For advanced control, run installation scripts directly.

Syntax ​

bash
./Dev/scripts/install-{tech}-rules.sh [options] <target-directory>

Available Scripts ​

ScriptTechnology
install-common-rules.shCommon/transversal
install-symfony-rules.shSymfony
install-flutter-rules.shFlutter
install-react-rules.shReact
install-reactnative-rules.shReact Native
install-python-rules.shPython
install-angular-rules.shAngular
install-csharp-rules.shC#/.NET
install-laravel-rules.shLaravel
install-vuejs-rules.shVue.js
install-php-rules.shPHP

Script Options ​

OptionDescription
--installFresh installation (default)
--updateUpdate existing files only
--forceOverwrite all files
--preserve-configKeep CLAUDE.md and project context
--dry-runSimulate without changes
--backupCreate backup before changes
--interactiveGuided installation
--lang=XXSet language (en, fr, es, de, pt)
--agents-onlyInstall only agents
--commands-onlyInstall only commands
--rules-onlyInstall only rules
--templates-onlyInstall only templates
--checklists-onlyInstall only checklists

Examples ​

bash
# Basic installation
./Dev/scripts/install-symfony-rules.sh --lang=fr ~/my-project

# Update existing installation
./Dev/scripts/install-flutter-rules.sh --update ~/my-app

# Force reinstall with backup
./Dev/scripts/install-python-rules.sh --force --backup ~/api

# Interactive mode
./Dev/scripts/install-react-rules.sh --interactive ~/frontend

# Install only agents
./Dev/scripts/install-symfony-rules.sh --agents-only ~/project

Ralph Wiggum CLI ​

Run Claude in a continuous loop until task completion.

Usage ​

bash
npx @the-bearded-bear/bear-craft ralph "task description" [options]

Options ​

OptionDescription
--fullEnable all DoD validators
--max-iterations=<n>Maximum iterations (default: 10)
--dod=<file>Custom DoD configuration file

Examples ​

bash
# Basic task
npx @the-bearded-bear/bear-craft ralph "Implement user authentication"

# With full DoD checks
npx @the-bearded-bear/bear-craft ralph --full "Fix the login bug"

# Custom iteration limit
npx @the-bearded-bear/bear-craft ralph --max-iterations=20 "Refactor the payment module"

Kanban Command ​

Launch a local web UI to view and manage the BMAD v6 backlog and sprints as a Scrum / Kanban board. The server binds to 127.0.0.1 only and never reaches out to the internet.

Usage ​

bash
npx @the-bearded-bear/bear-craft kanban [path] [options]

path defaults to the current working directory. The target must contain a project-management/ subdirectory (generated by /bear-workflow:plan or /bear-workflow:start).

Options ​

OptionDescription
--port=<n>HTTP port, 1024-65535 (default: 3737). The = form is required: --port 4000 (with a space) is rejected with a usage message
--openOpen the default browser automatically
--readonlyDisable all writes: the server answers 403 readonly to every PATCH, PUT, POST and DELETE, and the board offers no move control
--no-watchDisable the file watcher (no live reload on external edits)
--user=<name>Author recorded in the history of changes made from the board (kanban:<name>). Default: BEAR_CRAFT_USER, then git config user.name, then the OS user, then kanban

Views ​

Six views, listed in the navigation bar.

  • Board — 6 columns (backlog, ready-for-dev, in-progress, review, done, blocked). Move a card by drag and drop or with the "Move" menu of the card, which lists only the allowed target columns (the keyboard and single-pointer alternative to dragging). Gates (INVEST or story points, task completion, TDD phase) are enforced server-side. A column header shows its count, or count/limit when a WIP limit is set.
  • Backlog — Epic tree with per-epic progress. Select one or several stories and add them to a sprint, or move them back to the backlog; a capacity bar compares the committed points with the capacity of the target sprint.
  • Sprints — every sprint (planned, active, closed) with its goal, dates, points, stories and expandable tasks, plus its review and retro notes. Create, start and close sprints here.
  • Burndown — ideal vs actual curve for the active sprint, on-track / at-risk / behind indicator. The velocity chart (committed vs delivered points of the closed sprints, average of the last three, forecast of the sprints left) comes with an equivalent table.
  • Dependencies — directed graph of inter-story dependencies with cycles highlighted, and an equivalent table (story, depends on, blocks, status, cycle).
  • Docs — markdown viewer for prd.md, tech-spec.md, personas.md, architecture/*.md and the epic, story and sprint documents. US-XXX identifiers in the text link to the story.

What you can do from the board ​

  • Move a story between columns. Allowed transitions are listed in State machine; blocking a story asks for a reason.
  • Create, edit and delete a story ("New story" on the Board and Backlog views; the form is in the story detail). Editable fields: title, epic, story points, priority, assignee, sprint, TDD phase and dependencies; persona at creation. Deleting asks for confirmation and moves the file to project-management/backlog/user-stories/.trash/ (it can be restored by moving it back). A story in progress or in review needs an explicit confirmation, and a story that other stories depend on cannot be deleted (409 has_dependents).
  • Add, rename, delete tasks and set their status (to-do, in-progress, done, blocked) from the story detail (a task is renamed in place: Enter saves, Escape cancels; deleting asks once); the completed-task count is recomputed. A story goes to review only once it has tasks, all done. For a story that only had a task counter, the first change records a status per task (the first tasks, as many as the counter said, count as done).
  • Plan a sprint: select stories in the Backlog view and add them to a planned or active sprint.
  • Create, start and close a sprint. Creating a sprint (name, start and end dates, goal, capacity) writes sprints/sprint-<N>-<slug>/sprint-goal.md; the sprint is "planned". Starting it makes it the active sprint (refused while another sprint is active). Closing it archives the done stories under archived_sprints and lets you choose where the unfinished stories go: back to the backlog or into a planned sprint.
  • Set WIP limits per column in the Settings panel. They are indicative: exceeding one shows a warning and never blocks a move. A column counts the stories of the active sprint, as the board shows them.
  • Unsaved input is protected: closing a form (Escape, Cancel, the backdrop) or changing the URL (browser Back, a story link) while a story or sprint form holds changes asks before discarding them.
  • Search, filter and sort: the search box (Ctrl+K or /) filters the Board and Backlog; priority, assignee and epic filters and the sort order are kept in the URL, for example #/kanban?q=auth&prio=must,should&who=alice&epic=E4&sort=priority&dir=desc. ?story=US-E4-03 opens a story from any view, and #/docs/<path> opens a document.
  • Interface language: English, French, Spanish, German or Portuguese, chosen automatically from the browser or in the Settings panel. Story titles and documents are shown as written.

The /bear-workflow:* and /bear-project:* commands remain valid and write the same files.

Where data is written ​

WhatFile
Status, sprint membership, tasks, points, assignee, priority, history.bmad/sprint-status.yaml
Story content (description, acceptance criteria, persona)project-management/backlog/user-stories/US-*.md
Sprint goal of a new sprintproject-management/sprints/sprint-<N>-<slug>/sprint-goal.md
WIP limits.bmad/kanban.yaml
Deleted storiesproject-management/backlog/user-stories/.trash/

The markdown files never receive a status line. The optional keys stories.<id>.sprint_id and metadata.closed_date are described in Project/i18n/en/templates/sprint-status.md. A story that exists only as a markdown file (no entry in the YAML) is still moved by rewriting its frontmatter.

Concurrency ​

  • Writes to .bmad/sprint-status.yaml take the same lock as the .bmad/lib/ scripts (status-lock.sh). While a script holds it, the board answers 423 locked; a lock left by a crashed process is taken over after 10 seconds.
  • Conflicts are detected per story: each story carries a revision, sent back as If-Match. A story changed elsewhere in the meantime is refused with 409 conflict and the interface offers to reload it.
  • The file is rewritten with comments, key order and formatting preserved, after revalidation. A file that does not validate is never overwritten (409 sprint_status_invalid).

Security ​

  • Bound to 127.0.0.1 exclusively. No 0.0.0.0, no LAN exposure.
  • The Host header must be the loopback address and the served port (DNS rebinding protection); the Origin is checked on every write, and cross-origin writes are rejected with 403.
  • Request bodies are limited to 64 KB (512 KB for a story body).
  • --readonly blocks every write.
  • Path traversal blocked on the docs endpoint; error responses carry no filesystem path.
  • File writes are atomic (lock, temporary file, rename).
  • Strict CSP (script-src 'self', connect-src 'self') on the served HTML.
  • No eval, no shell execution, no outbound network.

State machine ​

Allowed transitions (all others rejected with 409 invalid_transition, with the list of allowed targets):

backlog       -> ready-for-dev (invest_score >= 6 when set, else story_points > 0) | blocked
ready-for-dev -> in-progress | backlog | blocked
in-progress   -> review (all tasks completed) | blocked
review        -> done (all tasks + tdd_phase in green|refactor|done) | in-progress | blocked
blocked       -> previous_status (unblock only)
done          -> (terminal)

BMAD v6 sprint-status source ​

Stories come from .bmad/sprint-status.yaml (written by /bear-workflow:* and /bear-project:backlog) and from the project-management/ markdown files. When a story is tracked in the YAML, the YAML wins for status, title, points, priority, assignee, epic, tasks and sprint; the markdown keeps the content. The board is live: the watcher reloads it when either source changes. Stories of a closed sprint (archived_sprints) are read-only.

If the board is empty or a story is missing, the diagnostics banner lists the files that could not be read; the same list is returned by GET /api/diagnostics. See Troubleshooting.

HTTP routes ​

Reads (GET): /api/health (ok, readonly, user, wip_limits), /api/diagnostics, /api/stories, /api/stories/:id, /api/dependencies, /api/docs, /api/docs/<path>, /api/sprints, /api/sprints/:id, /api/sprints/current, /api/velocity, /api/events (server-sent events).

Writes:

RoutePurposeSpecific errors
PATCH /api/stories/:id/statusMove a story (status, blocked_reason?, reason?)409 invalid_transition, archived, read_only_source, conflict
PATCH /api/stories/:idEdit title, points, assignee, priority, TDD phase, epic, sprint (null = backlog), dependencies400 invalid_payload; 409 archived, sprint_not_found, conflict
PUT /api/stories/:id/bodyReplace the markdown body (markdown, expected_mtime)409 no_backing_file, conflict; 422 status_line_forbidden
POST /api/storiesCreate a story (title, epic_id?, story_points?, priority?, persona?, sprint_id?)409 sprint_status_missing, sprint_status_invalid, sprint_not_found
DELETE /api/stories/:id[?force=1]Move a story to .trash/409 story_active, has_dependents, archived
PATCH /api/stories/:id/tasks/:taskIdSet a task status and/or rename it (status?, title?, at least one; 1..200 characters)400 invalid_payload; 404 task_not_found; 409 task_exists (name of another task), conflict
DELETE /api/stories/:id/tasks/:taskIdRemove a task (total and completed are recounted)404 task_not_found; 409 conflict
POST /api/stories/:id/tasksAdd a to-do task (title)409 task_exists
POST /api/sprintsCreate a planned sprint409 sprint_exists
POST /api/sprints/:id/startStart a sprint409 sprint_already_active, sprint_closed, conflict
POST /api/sprints/:id/closeClose a sprint (carry_over: backlog or a sprint id)409 not_active, sprint_not_found, conflict
PUT /api/settings/wipSet WIP limits (status → 1..99 or null)400 invalid_payload

:taskId is the current name of the task, URL-encoded (encodeURIComponent): names with /, %, #, spaces or non-ASCII characters are supported.

POST /api/stories, POST /api/stories/:id/tasks and POST /api/sprints accept an Idempotency-Key header (an opaque string of 8 to 128 visible ASCII characters; the board sends a fresh crypto.randomUUID() for each opening of a form). A second request on the same route with the same key within 5 minutes gets the answer of the first (same status and body, plus Idempotent-Replayed: true) and creates nothing; refused requests (status 400 and above) are not remembered, so a retry is evaluated again. The memory is the server process's, bounded to 500 keys. A malformed key is a 400 invalid_idempotency_key.

Errors common to all writes: 400 invalid_json / invalid_payload, 403 forbidden / readonly, 404 not_found, 413 payload_too_large, 423 locked, 500 write_failed. Changes are pushed to open browsers over /api/events.

Keyboard shortcuts (Board) ​

KeyAction
↑ ↓Navigate cards
← →Change column
Alt+MOpen the move menu of the focused card
1-6Move the card to that column (move menu open)
EnterOpen the story details
EscClose the menu or dialog
Ctrl+K (⌘+K) or /Search stories

The shortcuts are optional: every action is reachable with Tab and the native buttons.

Examples ​

bash
# Launch on the current project, open browser
npx @the-bearded-bear/bear-craft kanban --open

# Read-only mode for a demo or code review
npx @the-bearded-bear/bear-craft kanban --readonly --port=4040

# Against another project
npx @the-bearded-bear/bear-craft kanban ~/work/my-project

Claude Code Slash Commands ​

The following slash commands are available in Claude Code (v2.1.105+).

/btw ​

Quick questions without context switching, using minimal context and low effort.

Usage ​

bash
/btw <question>

Description ​

Ask a quick question without polluting the main conversation context. Uses minimal context window and low effort mode for fast responses. Ideal for simple lookups, factual questions, or quick clarifications that don't require full conversation history.

Examples ​

bash
/btw What's the difference between bcrypt and argon2?
/btw How do I check my Node.js version?
/btw What does the --force flag do?

/hooks ​

Interactive hook management interface.

Usage ​

bash
/hooks

Description ​

Launch an interactive TUI for managing Claude Code hooks. Allows you to:

  • View all installed hooks
  • Enable/disable hooks
  • Test hook execution
  • Debug hook issues
  • View hook output and errors

Examples ​

bash
# Open hook management interface
/hooks

/reload-plugins ​

Manually reload all Claude Code plugins.

Usage ​

bash
/reload-plugins

Description ​

Force a reload of all installed plugins without restarting Claude Code. Useful after:

  • Installing new plugins
  • Updating plugin configurations
  • Debugging plugin issues
  • Making changes to plugin files

Examples ​

bash
# Reload all plugins
/reload-plugins

/reload-skills ​

Re-scan all Claude Code skills without restarting the session (v2.1.157+).

Usage ​

bash
/reload-skills

Description ​

Force a re-scan of all installed skills (.claude/skills/) without restarting Claude Code. Distinct from /reload-plugins which reloads code intelligence plugins. Useful after:

  • Adding a new skill to .claude/skills/
  • Modifying an existing skill file
  • Installing Bear Craft updates that include new skills

SessionStart hook: A SessionStart hook can return reloadSkills: true to make newly-installed skills available in the same session without user intervention.

Examples ​

bash
# Reload all skills
/reload-skills

/loop ​

Run a prompt or slash command repeatedly until you stop it — the time-based member of Claude Code's four loop types.

The four loop types (Anthropic taxonomy) ​

TypeTriggerStops whenBest for
Turn-basedyour promptClaude finishes or needs inputshort, one-off tasks
Goal-based (/goal)your promptgoal condition holds or max turnstasks with a verifiable exit criterion
Time-based (/loop, /schedule)an intervalyou cancel / work is donerecurring work, monitoring external systems
Proactiveevent or schedule, no human in the looproutine disabledwell-defined recurring streams (triage, migrations)

Usage ​

bash
/loop [interval] [prompt-or-command]
  • Interval mode: /loop 5m … re-runs every 5 minutes (s/m/h suffixes). Runs locally — the session must stay open.
  • Self-paced (dynamic) mode: omit the interval and Claude schedules its own next wake-up (ScheduleWakeup), picking a delay matched to what it is waiting on. Ideal when the right cadence is not fixed.
  • Customisation: a loop.md file lets you tune loop behaviour per project.
  • Combine with /goal to give the loop a hard completion condition, and monitor cost/turns with /usage, /goal, /workflows.

Examples ​

bash
# Fixed cadence: address review comments and fix CI every 5 min
/loop 5m check my PR, address review comments, and fix failing CI

# Self-paced: watch a deploy, waking when it makes sense
/loop watch the staging deploy and report when it is healthy or failed

/schedule ​

Move a recurring job to the cloud as a Routine — it keeps running on a cron/event trigger after you close the session (unlike /loop, which is local). Use it for unattended nightly work: @qa:recette regression runs, /bear-common:ralph-run-style completion loops, dependency-upgrade sweeps, PR triage. Pair with /goal (exit criteria), skills, and dynamic workflows for fully autonomous operation. Route judgement-heavy steps to capable models and mechanical steps to smaller/faster ones to control token cost.

bash
/schedule            # create / list / manage cloud routines

/proactive ​

Alias for /loop - run a command on a recurring interval.

Usage ​

bash
/proactive [interval] [command]

Description ​

Execute a command or prompt repeatedly on a specified interval. This is an alias for the /loop command. If no interval is provided, the model will self-pace the iterations.

Examples ​

bash
# Check deployment status every 5 minutes
/proactive 5m /status

# Run tests continuously with auto-pacing
/proactive npm test

# Monitor PR reviews every 10 minutes
/proactive 10m Check if there are new PR reviews

/ultrareview ​

Parallel multi-agent code review of the current branch or a GitHub PR (available in Claude Code v2.1.111+).

Usage ​

bash
/ultrareview          # Review current branch
/ultrareview <PR#>    # Review a GitHub PR

Description ​

Spawns multiple reviewer agents (security, architecture, performance, style) in parallel and aggregates findings. Cloud-based, billed separately. Requires a git repository (offer git init if not in one).


/goal (v2.1.139+) ​

Set a persistent session goal that Claude tracks until completion. A Stop hook activates and prevents Claude from stopping until the condition is met.

Usage ​

bash
/goal <description>
/goal clear

Description ​

Defines an objective Claude must complete before the session ends. Useful for long-running tasks (audits, migrations, multi-file refactors) where you want Claude to self-direct until done. The goal is shown in the status line. Use /goal clear to remove it early.

Examples ​

bash
/goal Write unit tests for all files in src/services/
/goal Migrate all Axios calls to the fetch API across the codebase
/goal clear

/usage (v2.1.143+) ​

Display token usage statistics for the current session — input tokens, output tokens, cache reads, and estimated cost.

Usage ​

bash
/usage

Description ​

Shows a breakdown of token consumption since session start: input tokens, cache read tokens, output tokens, and an estimated cost in USD. Since v2.1.149 the breakdown also attributes consumption per component — skill, subagent, plugin, and MCP server — so you can spot the costliest driver before deciding to delegate or /clear. Use before /clear to decide whether compaction is worth it.


/workflows (v2.1.154+) ​

Show the live progress tree for the current Dynamic Workflow run (ultracode mode).

Usage ​

bash
/workflows

Description ​

Displays a real-time view of all agents spawned by an active workflow, their phase assignments, completion status, and timing. Only meaningful when a Workflow tool call is in progress.


/tui ​

Switch the session to the terminal UI mode (available in v2.1.110+).

Usage ​

bash
/tui

Toggle via the tui setting in settings.json.


/recap ​

Summarize the current session progress (v2.1.108+). Useful before /clear to capture state.

Usage ​

bash
/recap

/undo ​

Alias for /rewind (v2.1.108+). Reverts the last change in the session.

Usage ​

bash
/undo

/effort ​

Interactive effort level slider (v2.1.111+). Pick the thinking depth: low, medium, high, xhigh (Opus 5.5, adaptive thinking), ultracode (v2.1.154+, Dynamic Workflows ; separate toggle since v2.1.284).

Usage ​

bash
/effort               # Interactive picker
/effort high          # Set directly
/effort ultracode     # Dynamic Workflows — max code throughput (trigger keyword: ultracode)

Default effort: medium on Opus 5.5 and Haiku 5.5, high on Sonnet 5.5, Fable 5.2 and legacy models. Since v2.1.284 Ultracode is a separate toggle in /effort and no longer forces xhigh.

NiveauModèleUsage
lowHaiku 5.5 (claude-haiku-5-5)Tâches simples, lookups, classification
mediumSonnet 5.5 (claude-sonnet-5-5)Implémentation standard
highOpus 5.5 (claude-opus-5-5)Raisonnement complexe, architecture
xhighOpus 5.5 (adaptive thinking)Décisions critiques, migrations complexes, ADR
ultracodeOpus 5.5 (toggle séparé depuis v2.1.284, Dynamic Workflows)Mode débit code maximal — pipelines automatisés, génération massive

Note Fable 5.2 / Mythos : Fable 5.2 (claude-fable-5-2, $8/$40) est le modèle Fable courant, adapté aux agents créatifs/narratifs ; Fable 5.1 et Fable 5 restent disponibles (legacy). Réflexion toujours active sur Fable 5.2 : pour la réduire, baisser l'effort. Mythos 5 / 5.1 / 5.2 sont en disponibilité limitée (Cyber Verification Program) — ne pas les configurer par défaut.


Claude Code Native Features (v2.1.108+) ​

Native CLI Binary (v2.1.113+) ​

Starting in v2.1.113, Claude Code spawns a native compiled binary instead of the bundled JavaScript. Faster startup, lower RAM footprint. No user action required — automatic when you update.

Forked Subagents (v2.1.117+) ​

Set CLAUDE_CODE_FORK_SUBAGENT=1 to run each subagent in an isolated, forked context. Prevents context pollution between parallel investigations.

bash
export CLAUDE_CODE_FORK_SUBAGENT=1

Prompt Caching Env Vars (v2.1.108+) ​

Fine-tune prompt caching behavior:

VariableEffect
ENABLE_PROMPT_CACHING_1HUse the 1-hour cache tier (vs 5-minute default)
FORCE_PROMPT_CACHING_5MForce the 5-minute tier even if 1h is available

Use ENABLE_PROMPT_CACHING_1H=1 for long-running sessions with a stable context (e.g., architecture reviews that span hours).


Configuration File ​

YAML Configuration ​

For monorepos and multi-project setups, use claude-projects.yaml:

yaml
settings:
  default_lang: "en"

projects:
  - name: "my-monorepo"
    description: "My fullstack application"
    root: "~/Projects/my-monorepo"
    lang: "fr"
    common: true
    modules:
      - path: "frontend"
        tech: react
      - path: "backend"
        tech: symfony
      - path: "mobile"
        tech: flutter
      - path: "api"
        tech: [python, react]  # Multiple technologies

Environment Variables ​

VariableDescriptionDefault
BEAR_CRAFT_LANGDefault languageen
BEAR_CRAFT_TARGETDefault target directory.
BEAR_CRAFT_CONFIGConfig file pathclaude-projects.yaml
CLAUDE_CODE_FORK_SUBAGENTFork each subagent in isolated context (v2.1.117+); on by default in -p, the SDK, desktop and VS Code since 2.1.302, 0 to turn it offunset (on in -p/SDK since 2.1.302)
ENABLE_PROMPT_CACHING_1HUse 1-hour prompt cache tier (v2.1.108+)unset
FORCE_PROMPT_CACHING_5MForce 5-minute cache tier (v2.1.108+)unset
CLAUDE_CODE_SUBAGENT_MODELModel for subagents (e.g. claude-sonnet-5-5)unset

Exit Codes ​

CodeMeaning
0Success
1General error
2Invalid arguments
3Missing prerequisites
4Target directory not found
5Permission denied

Troubleshooting ​

NPX cache issues ​

bash
# Clear NPX cache
npx clear-npx-cache
# or
rm -rf ~/.npm/_npx

Script not executable ​

bash
chmod +x Dev/scripts/*.sh
# or
make fix-permissions

Wrong yq version ​

bash
# Bear Craft requires yq v4 (Mike Farah's version)
yq --version
# Should show: yq (https://github.com/mikefarah/yq/) version v4.x.x

See Also ​