Plugin Commands
The DevSpec plugin adds slash commands to Claude Code covering the full action item workflow — from creating and scoping items, to implementing and committing them, to running the full autopilot loop. Every command wraps DevSpec MCP endpoints in a user-friendly interface.
Prerequisites
Before using plugin commands, ensure you have the following set up.
- Claude Code v2.0 or later installed.
- A DevSpec MCP token generated (Settings > MCP Tokens in the DevSpec web UI).
- MCP connection configured in Claude Code (via
claude mcp addor manual.mcp.json). - The autopilot plugin installed (run
/plugin install devspec-autopilot@devspec-autopilot-marketplacein Claude Code).
If you haven't connected DevSpec to Claude Code yet, see the MCP Endpoint docs for setup instructions.
Quick Reference
All available commands at a glance.
| Name | Scope | Description |
|---|---|---|
/autopilot | read-write | The full unattended loop. Polls DevSpec for staged items, picks them up one at a time, implements and commits them in isolated worktrees, then repeats until nothing is staged or you stop it. |
/autopilot.start | read-write | Kicks off the polling loop. Checks that autopilot is enabled in project settings first, then displays a startup banner and begins cycling. Accepts optional flags to narrow what it picks up: --items=ID1,ID2,... (targeted mode — process only these items, in order, then exit), --mine (default: your items + unassigned), --assigned-to=<user_id>, --all (all staged work, no assignee filter), --created-by=<user_id>, and --drain (process everything staged once, then stop instead of polling). Flags can be combined. |
/autopilot.stop | read-write | Gracefully stops the loop. Won't interrupt a mid-flight item — waits for the current cycle to finish, then prints a session summary (cycles run, completed, failed). |
/autopilot.status | read | Snapshot of the current autopilot state: online/offline, staged-item count, how many items are in-progress, and key settings like push/merge mode. |
/autopilot.history | read | Lists recently completed and failed autopilot runs, sorted by most recent, with relative timestamps and branch names. |
/devspec.work | read-write | The main workhorse. Takes an action item name or ID, optionally brainstorms, implements the changes, commits with a tracking tag, and marks it done. Supports --unattended for fully hands-off execution. |
/devspec.remote | read-write | Connect this local agent as a DevSpec remote-control target. Opens a private control channel, mirrors turns into DevSpec, and polls for instructions from the Agents page / phone. Not Claude's built-in /remote-control. Uses packaged poller + token resolve from .mcp.json. |
/devspec.remote-stop | read-write | Disconnect remote control — marks the Agents page offline immediately and clears local ~/.devspec/remote-control.json state. |
/devspec.create | read-write | Create a new action item in DevSpec from the terminal. Accepts a title, description, type, priority, and whether it should be staged for autopilot. |
/devspec.brainstorm | read-write | Interactive scoping session for a specific item. Asks up to 5 targeted questions across scope, approach, data, edge cases, and acceptance criteria, then saves a summary back to DevSpec. |
/devspec.done | read-write | Retrospective logger for work already completed outside the normal flow. Detects recent git commits and affected files automatically, then creates a completed action item record with implementation notes. |
/devspec.commit | read-write | Generates a commit message with a [devspec:<id>] deployment-tracking tag using DevSpec's API, then executes the commit. |
/devspec.link | read-write | Manually associates an existing git commit SHA with a DevSpec action item. Useful for retroactively connecting commits made without the tracking tag. |
Core Workflow
The most common workflow covers creating, implementing, and recording work.
Implement an action item
Run /devspec.work with an action item name or ID. It optionally brainstorms first, then implements the changes, commits with a tracking tag, and marks it done.
Create tasks on the spot
Found a bug while coding? Run /devspec.create with a title and it's logged immediately — no context switching.
Record completed work
Just finished a quick fix? Run /devspec.done and it immediately creates a full action item from your recent commits — including testing notes, usage notes, and a changelog summary.
Run the autopilot loop
Run /autopilot.start to begin the continuous polling loop. It picks up staged items one at a time, implements them in isolated worktrees, and repeats until nothing is staged.
Deploy with tracking
Run /devspec.commit with an action item ID to generate a commit message with a deployment tracking tag. When deployed, DevSpec automatically marks the item as deployed.
Troubleshooting
Common issues and how to resolve them.
Command not recognized
Claude Code doesn't show the /devspec.* commands in the palette.
Ensure the plugin is installed by running /plugin install devspec-autopilot@devspec-autopilot-marketplace in Claude Code, then run /reload-plugins. The commands are auto-discovered from the plugin's commands/ directory.
MCP token not configured
Commands fail with an authentication error.
Generate a token in DevSpec (Settings > MCP Tokens), then add it to Claude Code via claude mcp add or your .mcp.json configuration file.
Read-only token on write commands
Commands like /devspec.create fail with a scope error.
Write commands require a read-write token. Generate a new token with read-write scope in DevSpec Settings > MCP Tokens.
DevSpec server unreachable
Commands timeout or fail with a connection error.
Check that the DevSpec URL in your MCP configuration is correct and the service is running. If using a self-hosted instance, verify the server is accessible.