Search conversations, projects, playbooks, and more
CLI
Complete command reference for the crewkit CLI.
The crewkit CLI manages your AI coding sessions, team configuration, and project setup from the terminal.
crewkit # Start a coding session — Claude Code runs here, crewkit observingcrewkit code # Start a coding sessioncrewkit auth login # Connect to your account
Bare crewkit is the same as crewkit code — every launch flag works bare (crewkit -c, crewkit -p "...", crewkit --sensitive). A flagless crewkit with piped or captured output shows the setup overview instead, which also lives at crewkit status.
crewkit code --prompt "Fix the bug" # Headless one-shot
crewkit code -c # Continue most recent session
crewkit code --resume # Pick a session to resume
crewkit code --resume <id> # Resume specific session
By default on macOS and Linux, crewkit code launches the native claude — Claude Code owns the terminal exactly as if you'd run it directly — while crewkit loads beside it in the background: session tracking, telemetry, and team context. crewkit adds a statusline to the Claude Code session on every observed launch (opt out with --no-statusline, CREWKIT_STATUSLINE=off, or [statusline] enabled = false in .crewkit/config.toml). If the background observer cannot be set up, the launch fails with the reason rather than silently starting an unobserved session — run crewkit tui if you want the interactive terminal interface. On Windows, the interactive terminal interface remains the default.
Flags:
Flag
Description
--debug
Enable debug logging (per-session log at .crewkit/debug-latest.log)
Max execution time for headless mode (default: 300)
-q, --quiet
Suppress progress output in headless mode
-c, --continue-session
Continue most recent Claude session
-r, --resume [id]
Resume a session by ID, or show picker
--fork-session
Fork the resumed session (new ID, existing context)
--sensitive
Mark session as sensitive (excluded from cross-repo discovery)
--repo-id <id>
Explicit repository ID (also: CREWKIT_REPOSITORY_ID env var)
-y, --yes
Skip interactive prompts (for CI/CD)
--no-auth
Skip authentication (sessions will not be tracked)
--no-statusline
Don't add the crewkit statusline to the Claude Code session
--status
Show the background observers for this directory (same as crewkit sidecar status)
--stop
Stop this terminal's background observer (same as crewkit sidecar stop; --all stops every observer for the directory)
--no-artifacts
Disable artifact context injection into the session
--no-context
Skip injecting save-state recall context (personal + team digests)
--no-title
Disable terminal window title
--disable-lsp
Disable LSP server auto-start
--chrome
Use Chrome for browser automation
--llm-gateway
Enable LLM gateway proxy
--llm-gateway-port <port>
Gateway proxy port (default: ephemeral)
--dangerously-skip-permissions
Skip all permission prompts (CI/CD only — disables safety checks)
--no-tui is deprecated: it named the old opt-out, and its behavior — native Claude Code with crewkit alongside — is now the default. --tui likewise still parses, but the documented spelling is crewkit tui.
Start the same coding session inside crewkit's interactive terminal interface — a live sidebar with session metrics, git context, and AI summaries.
crewkit tui # Coding session with the live sidebarcrewkit tui -c # Session launch flags work here too
crewkit tui takes the session launch flags (-c/--continue-session, --resume, --sensitive, ...) but not headless -p — headless is crewkit -p "...". On Windows, this interface is the default under every spelling.
An alias of crewkit code — the same launch under a second name. Its status and stop subcommands inspect or stop the background observers that ship with the default launch (macOS and Linux).
crewkit sidecar # Launch claude with a background observercrewkit sidecar status # Show observer state for this projectcrewkit sidecar status --json # Machine-readable statecrewkit sidecar stop # Stop the observer in this terminalcrewkit sidecar stop --all # Stop every observer for this directory
Note: crewkit sidecar status reports the background observers, while crewkit status shows the setup overview.
Pin a single-repository mapping locally, or create a multi-repository workspace.
crewkit init # Pin the detected mapping in .agent/config.ymlcrewkit init --yes # The same prompt-free local pin for a single repocrewkit init --workspace # Create and register a workspace project
Single-repository init does not create a project on the server. Use crewkit project create for registration; --yes does not change that boundary. Workspace init still creates the workspace project server-side.
crewkit project list # List projectscrewkit project create "My Project" --org acme # Register without promptscrewkit project info # Show current project
crewkit project create <NAME> [--org <SLUG_OR_ID>] registers a repository project server-side without onboarding or device-flow prompts. Use an explicit organization slug or external ID for deterministic CI and BYOC setup. Inside a git repository, crewkit includes the detected origin remote; outside git, creation still works when --org is supplied. This command does not write local configuration.
crewkit members list # List memberscrewkit members invite user@example.com # Invite (default: member role)crewkit members invite user@example.com -r admin # Invite as admincrewkit members update user@example.com -r admin # Change rolecrewkit members remove user@example.com # Remove membercrewkit members invitations # List pending invitations
View and manage resources (agents, skills, commands, rules).
crewkit resources list # List all resourcescrewkit resources list --type agent # Filter by typecrewkit resources show rails-expert # Show resource detailscrewkit resources inspect rails-expert # Inspect full configcrewkit resources import # Import local resourcescrewkit resources import --dry-run # Preview import
crewkit sessions list # Recent sessions (current project)crewkit sessions list --all-projects # Across the whole organizationcrewkit sessions list --since 7d --user me # Your sessions from the last weekcrewkit sessions show <id> # Full detail (accepts short id prefix)crewkit sessions report --week # Team digest: sessions, cost, top agents
Share a session — create, list, or revoke share links:
crewkit sessions share <id> # Org-visible link, expires in 7 dayscrewkit sessions share <id> --access public # Anyone with the link (requires admin)crewkit sessions share <id> --level summary # Levels: summary, transcript, fullcrewkit sessions share <id> --expires 0 # Never expirescrewkit sessions share <id> --list # List existing sharescrewkit sessions share <id> --revoke <share-id> # Revoke a share
Import history — bring in local Claude Code JSONL transcripts:
crewkit sessions import # Import local Claude conversationscrewkit sessions import --since 2026-01-01 # Import recent onlycrewkit sessions import --dry-run # Preview what would importcrewkit sessions import <transcript-id> # Import specific conversation
crewkit blueprint create "Ship the billing revamp" # Generate a blueprint from a goalcrewkit blueprint list # List blueprints for this projectcrewkit blueprint show <id> # Epics and taskscrewkit blueprint review # Review against the local codebasecrewkit blueprint tasks --status pending # List tasks, filter by statuscrewkit blueprint start <task-id> # Claim a task (sets in_progress)crewkit blueprint complete <task-id> # Mark a task completedcrewkit blueprint export # Export as a plan file to .claude/work/
Manage project save states (personal handoffs + team digests).
crewkit save-state ls # List this project's save statescrewkit save-state show personal # Show your handoffcrewkit save-state show team # Show the most recent team digestcrewkit save-state rm <id> # Delete a save state
MCP server exposing crewkit's project memory to agents. Any MCP-capable agent can pull project context, conventions, prior work, active work, and blueprint state. Sessions launched by crewkit code register it automatically.