multica-skill

Harness-agnostic skill bundle that turns any Multica board into a full coding pipeline — todo → explore → work → review → qa → done — drivable from Claude Code, Codex, Gemini, OpenCode, Pi, Cursor, Copilot, or Hermes.

One ticket. Many agents. Each phase a fresh context. Failures rewind to work, not into the void.

---

What's inside

Skill	Purpose
multica	Day-to-day Multica CLI guide — login, issues, agents, daemon, autopilot.
multica-workflow	Symphony-style 5-lane pipeline mapped onto Multica's fixed 7 states via phase:* labels. Per-lane prompt templates + a small multica-flow CLI to advance/rewind tickets.
multica-onboarding	First-run bootstrap that registers three battle-tested default skills: obra/superpowers, leweii/atlassian-cli, and Playwright.

The whole bundle is plain SKILL.md + bash, so every harness that reads markdown can consume it.

---

Install

Option 1 — Via Multica (recommended)

Multica has native skill import from GitHub:

multica skill import https://github.com/cskwork/multica-skill
multica skill list | grep multica

Then enable the bundle in any agent:

multica agent update <agent-slug> --skill multica --skill multica-workflow --skill multica-onboarding

Option 2 — Any other harness

git clone https://github.com/cskwork/multica-skill ~/.multica-skill
cd ~/.multica-skill
./install.sh

install.sh detects what's installed (claude, codex, gemini, opencode, pi, cursor, droid) and runs the matching adapter. You can also target one explicitly:

./install.sh claude-code    # ~/.claude/skills/
./install.sh codex          # ~/.codex/commands/
./install.sh gemini         # gemini extensions install
./install.sh opencode       # ~/.config/opencode/commands/
./install.sh pi             # ~/.pi/skills/

After install, run the onboarding skill from inside your harness:

/multica-onboarding

It will:

1. Verify multica CLI is installed and authenticated.
2. Register obra/superpowers (via your harness's plugin/extension system).
3. Drop leweii/atlassian-cli's SKILL.md into your skills dir (no-op if acli itself isn't installed).
4. Install @playwright/test + browsers into the current project, plus the playwright-cli-skill.

---

The pipeline

Multica's states are fixed: backlog | todo | in_progress | in_review | done | blocked | cancelled. multica-workflow layers Symphony's richer lane semantics on top using labels:

backlog ─▶ todo ─▶ in_progress ─▶ in_progress ─▶ in_review ─▶ in_review ─▶ done
          phase:    phase:         phase:         phase:        phase:
          explore   work           review         qa            learn

Severity-based rewinds:

• Review finds CRITICAL / HIGH / MEDIUM → ticket bounces back to in_progress / phase:work with a fresh context. The agent reads ## Review Findings (written into the ticket comments) and fixes only those.
• QA fails → same rewind, agent reads ## QA Failure.
• Three consecutive QA failures on the same ticket → blocked + human ping.

See docs/WORKFLOW.md for the full state diagram and skills/multica-workflow/lanes/ for the per-lane prompts.

---

Quick taste

# Create a ticket and start the pipeline
multica issue create --title "Add CSV export to /reports" --label "phase:explore"
# → returns MUL-142

# Assign the explorer agent — Multica auto-starts it
multica issue assign MUL-142 --agent claude-explorer

# That's it. The agent writes ## Domain Brief into the ticket, swaps the label to
# phase:work, sets state=in_progress, and Multica's daemon picks the next phase.
# Watch progress:
multica issue get MUL-142
multica daemon logs -f

For an end-to-end walkthrough, see examples/csv-export-walkthrough.md.

---

Harness compatibility matrix

Harness	Skill path	Adapter	Status
Multica (native)	multica skill import	n/a	✅ first-class
Claude Code	~/.claude/skills/<name>/SKILL.md	adapters/claude-code.sh	✅
Codex CLI	~/.codex/commands/<name>.md	adapters/codex.sh	✅
Gemini CLI	gemini extensions install	adapters/gemini.sh	✅
OpenCode	~/.config/opencode/commands/<name>.md	adapters/opencode.sh	✅
Pi	~/.pi/skills/<name>/SKILL.md	adapters/pi.sh	✅
Cursor	.cursor/rules/<name>.mdc	(via Multica or manual)	⚠️ partial
Copilot	.github/copilot-instructions.md	(via Multica)	⚠️ partial

Multica's own agent system (Claude Code, Codex, Cursor, Copilot, Gemini, Hermes — 11 total) receives the skill through multica skill import regardless of which adapter you pick locally.

---

Repo layout

multica-skill/
├── skills/
│   ├── multica/              # CLI usage guide
│   ├── multica-workflow/     # pipeline orchestration
│   │   ├── lanes/            # per-phase prompt templates
│   │   └── scripts/          # flow-next.sh, flow-rewind.sh
│   └── multica-onboarding/   # default-skill bootstrap
│       └── scripts/          # install-{superpowers,atlassian,playwright}.sh
├── adapters/                 # per-harness install scripts
├── bin/                      # multica-flow, multica-skill entrypoints
├── templates/                # workflow.yaml, ticket.md
├── docs/                     # WORKFLOW.md, HARNESSES.md, ARCHITECTURE.md
└── examples/                 # end-to-end walkthroughs

---

License

MIT. See LICENSE.

Credits

• Pipeline shape ported from cskwork/symphony-multi-agent.
• Default skill picks: obra/superpowers, leweii/atlassian-cli, Microsoft Playwright.
• Built for Multica's board + agent platform.