Multica Mentioning & Delegation Skill
Construct correct mention links to delegate work to agents, squads, or broadcast to all.
Evidence shows precise description of mention mechanics with clear permission boundaries, such as the distinction between invoke gate and see gate, the deliberate use of same reason_code to avoid existence leakage, and lineage restrictions on autopilot delegation. However, the skill only provides Bash(multica *) tool, no user confirmation step or data-flow transparency; it may trigger external effects (enqueuing tasks) without rollback or recovery. Deductions for missing explicit confirmation and rollback.
Skill relies on source map with file:line evidence, CLI commands are specific, high self-consistency. Key paths (parsing, triggering, suppression) covered by Go tests (e.g., TestMentioningSkillTeachesTheParserContract), but static review cannot reproduce execution. Error handling well described, but depends on external CLI availability, unverified. Cap at 10 due to static review.
Skill clearly states when to use: when needing to @mention someone; effectively defines non-fit cases (project references, member mentions don't trigger runs). Trigger conditions clear (exact mention syntax) to avoid false triggers. However, environment fit: requires access to Multica platform, possibly unreachable from mainland China; skill written in English only, no Chinese version, Chinese users may need translation. Deductions for limited environment evidence.
Documentation layered: SKILL.md provides overview, source map provides detailed evidence, naming consistent. No version or changelog mentioned, but CI workflow exists. License is NOASSERTION, though custom Multica License exists with terms that may affect usage. Publisher unverified, but maintenance responsibility clear in README. Deductions for non-standard license and unclear versioning.
Skill's goal (producing correct mention links) clear, with correct/incorrect examples, output directly usable. Saves time vs manual UUID lookup and link construction. But static review cannot verify actual effect; no benchmarks or alternative comparisons. Thus cap at 7.
Skill provides source map with file:line references and references multiple test files (e.g., mention_test.go, comment_trigger_preview_test.go), which serve as auditable primary evidence. CI configuration shows test suites exist. However, static review did not execute any tests, cannot independently reproduce; test coverage limited to some key paths. Thus cap at 5.
- This skill is only applicable to the Multica platform; if core functionality depends on overseas services, it may be unreachable from mainland China networks.
- License is NOASSERTION; the custom Multica License has additional conditions that may affect distribution and usage; review carefully before use.
- The skill does not include a user confirmation step; it may automatically trigger agent tasks, so ensure the agent runtime environment is secure.
What does this skill do, and when should you use it?
The Multica "Mentioning & Delegating" skill instructs how to construct mention links to trigger agents or squads in the Multica platform. It emphasizes using real UUIDs instead of display names, and details the behavior of different mention types (agent, squad, member, issue) and the @all broadcast. The skill is backed by source code references and tests for accuracy, and provides step-by-step for looking up UUIDs.
The skill reads Multica CLI commands to look up UUIDs for members, agents, and squads, then constructs mention links in the form [@Label](mention://type/uuid). It differentiates which mentions trigger new tasks (agent and squad), which only render links (member and issue), and explains the @all broadcast and failure behaviors such as coalesced, deferred, and blocked outcomes.
- A project manager wants to assign an issue to a specific agent by mentioning it to trigger its run.
- A squad leader needs to hand work to the whole squad by mentioning the squad so the leader agent can distribute it.
- A teammate wants to notify a human member in a comment without triggering any automated task.
- Someone wants to reference an issue as context without starting any new run.
- A team wants to broadcast @all to notify everyone on the issue while suppressing the automatic trigger for the assignee.
What are this skill's strengths and limitations?
- Detailed and source-backed for accuracy
- Clearly distinguishes mention types that trigger tasks from those that only render links
- Provides detailed explanations of failure modes for debugging
- Enables automation and delegation within teams
- Tightly integrated with the Multica platform for a unified workflow
- Only applicable to the Multica platform; not generalizable
- Requires familiarity with UUID lookups, which may be a learning curve
- Mentions that fail do not return error responses; only trigger_outcomes array reveals issues
- No standalone test suite; relies on upstream Go tests
How do you install this skill?
The skill ships in the Multica repository at server/internal/service/builtin_skills/multica-mentioning/SKILL.md. To use it, you need to install the Multica CLI (via Homebrew or install script) and complete setup. The detailed steps for adding the skill to your agent environment are not explicitly documented in the source.
How do you use this skill?
- Use CLI commands to look up UUIDs for members, agents, or squads, e.g.,
multica workspace member list --output jsonfor a member's user_id. 2. Construct mention links accordingly: [@Name](mention://agent/<uuid>) triggers an agent, [@Name](mention://squad/<uuid>) triggers the squad leader, [@Name](mention://member/<user_id>) only links, [@Name](mention://issue/<uuid>) only references. 3. Embed the link in a comment and be aware of @all usage.