feat(skill): orchestrate workspace business goals

[ShirleyGodzilla]

Summary

Add workspace-aware business orchestration to the existing official Multica CLI Skill.

For an open-ended goal, the Skill now:

• performs targeted read-only discovery across relevant Issues, Projects, Agents, Squads, Skills, and Autopilots;
• synthesizes existing business context, prior decisions, ownership, and available capabilities;
• matches capabilities to the goal and prefers reuse over duplicate resources;
• treats unknown sharing as shared and isolates changes that could affect other workflows;
• presents one complete in-chat execution plan, then executes it in dependency order after confirmation;
• requires a separate confirmation immediately before Agent or Skill mutations;
• pauses on material deviations and routes unsupported operations to Multica Web.

The in-chat orchestration design is the execution plan. This does not introduce a repository design/specification cycle for business operations.

Why

The existing Skill is strong for direct Issue and CLI operations, but users still need to know which workspace resources contain the relevant knowledge and capabilities. This addition lets technical and non-technical users start from a business outcome, discover what the team already knows and can do, and reuse that work instead of creating parallel resources or acting on stale assumptions.

This is the focused migration of the plan-then-execute gap discussed in multica-ai/multica#6968. It extends the official Skill rather than introducing a competing Skill in the main Multica repository.

Implementation

• Add skills/multica-cli/references/orchestration.md for the detailed workflow.
• Route open-ended goals from SKILL.md while keeping known-target actions in the existing direct flow.
• Update Claude, Codex, Cursor, English, Chinese, and example discovery surfaces.
• Add a structural orchestration contract test.
• Include the new reference in the existing real-CLI command linter.

No Multica CLI or API changes are required.

Verification

• python3 scripts/test-orchestration-contract.py
• scripts/test-lint.py
• scripts/lint-skill-commands.py --verbose
• python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py skills/multica-cli
• JSON validation for both Claude manifests
• git diff --check

The command linter resolves all 120 documented invocations against multica 0.4.26 and confirms coverage for all 42 required command paths.

AI Disclosure

OpenAI Codex assisted with requirement refinement, implementation, documentation, and verification. The design was constrained to use only the authenticated community CLI, perform targeted rather than full-workspace discovery, avoid API bypasses, and preserve explicit confirmation for side effects.