multica-skills

Agent skills that let any AI coding agent — Claude Code, Codex, Cursor, OpenCode, Gemini CLI, and 40+ more — drive the Multica CLI.

Once installed, an agent can:

• create, update, archive, and assign Multica agents
• create and assign issues, including search, rerun, and attachments
• schedule recurring tasks via autopilots and cron
• manage projects, sprints, and epics
• create, import, and edit workspace skills
• run and diagnose the local agent daemon, plus inspect runtimes across machines
• bootstrap a fresh machine end-to-end

All seven skills call the official multica CLI — nothing is reimplemented, nothing talks to the Multica API directly.

Install

Uses npx skills (no global install needed):

# Interactive — pick which skills go to which agents
npx skills add Tigatron/multica-skills

# Preview what's available, don't install
npx skills add Tigatron/multica-skills --list

# Install everything to everything, non-interactive
npx skills add Tigatron/multica-skills --all

# Install one skill to one agent, globally
npx skills add Tigatron/multica-skills -s multica-issues -a claude-code -g

# All Multica skills to Claude Code only, project-scoped
npx skills add Tigatron/multica-skills --skill '*' -a claude-code

Scope flags:

• (default) — project-scoped, e.g. .claude/skills/ in the current repo
• -g, --global — user-scoped, e.g. ~/.claude/skills/

See npx skills --help for the full flag reference. Update, remove, and list are all covered:

npx skills list                  # What's installed?
npx skills update                # Pull the latest version
npx skills remove multica-issues # Uninstall one

Skills in this package

Skill	What it does
multica-setup	Install CLI, authenticate (cloud / self-host), bootstrap daemon, manage profiles
multica-issues	Issue CRUD, search, rerun, comments, subscribers, attachments, execution runs
multica-agents	Agent CRUD (create / update / archive / restore), skill assignment, delegation
multica-projects	Sprint / epic / workstream containers
multica-autopilot	Scheduled and on-demand recurring agent tasks
multica-skills	Workspace skill CRUD, import from clawhub.ai / skills.sh, file management
multica-daemon	Local daemon lifecycle, profiles, plus multica runtime cross-machine control

Each skill is a standalone directory under skills/ containing a single SKILL.md file with YAML frontmatter (name, description) followed by Markdown instructions — the Agent Skills format.

Prerequisites

• The multica CLI. If not installed, the multica-setup skill walks through it, or:

  brew install multica-ai/tap/multica                                                          # macOS / Linux
  irm https://raw.githubusercontent.com/multica-ai/multica/main/scripts/install.ps1 | iex     # Windows

• jq is recommended for the --output json patterns in the skills, but not required.

Verifying the install

After installing and restarting your agent, try a prompt like:

• "List open Multica issues assigned to Lambda."
• "Create a Multica issue titled 'stabilize flaky login test', priority high, assigned to Lambda."
• "Set up a weekday 9 AM autopilot that triages todo issues."
• "Why isn't the Multica daemon picking up work?"

The agent should invoke the multica CLI directly, with behavior informed by the relevant skill.

Design notes

• One skill per capability — keeps each SKILL.md small enough that a model can load only what it needs for a given request.
• Descriptions are the router — each description: line is written as a predicate (when to use me), because that is what agent frameworks match against.
• Commands are verbatim — every command shown is what the skill expects an agent to actually run, with the real flags and positional forms from the Multica CLI docs.
• Gotchas sections capture failure modes — wrong timezone strings, missing assignees, silent no-runtime, etc.

Keeping in sync with the CLI

The Multica CLI moves faster than its docs. To stop skills from drifting, the source of truth here is the CLI binary itself, not CLI_AND_DAEMON.md.

Layout:

• cli-snapshot/ — one file per multica <path> --help node, committed.
• bin/snapshot-cli.sh — regenerates the snapshot from the locally installed CLI.
• bin/drift-pr-body.sh — formats a structured drift report from git diff of the snapshot.
• .github/workflows/cli-drift.yml — runs daily, opens a PR on sync/cli-<version> when the snapshot has moved.

Each drift PR ships only the new snapshot. Updating the affected SKILL.md is a manual review step, and that's deliberate:

What the auto-PR touches	What stays human-only
cli-snapshot/*.txt	Gotchas (must be reproduced before adding)
	Common flows (cross-command workflows)
	description: frontmatter (agent routing predicates)

If you only ever re-emit --help, the skills degrade to a --help reformatter. The differentiator is the hand-on knowledge layered on top, so the workflow is explicitly not allowed to touch it.

To run the audit locally:

bash bin/snapshot-cli.sh cli-snapshot      # regenerates cli-snapshot/
git diff cli-snapshot                      # see what moved

To dry-run the action (no push, no PR), trigger cli-drift from the Actions tab with dry-run: true.

Contributing

PRs welcome. To add a new skill, follow the template used by the existing seven, or run npx skills init <name> and adapt.

Local sanity check (no install required):

for f in skills/*/SKILL.md; do
  head -5 "$f"                       # frontmatter visible
  grep -c '^multica ' "$f" || true   # command count
done

License

MIT. See LICENSE.