The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents an orchestration mechanism where specific Markdown link patterns can trigger secondary agent executions or squad assignments. This creates a multi-step execution surface (Indirect Prompt Injection vector 8b) where content processed by the agent can influence subsequent system behavior.
Ingestion points: The skill is designed to handle and respond to issue comments (SKILL.md).
Boundary markers: None specified for the construction of the mention links.
Capability inventory: The skill utilizes Bash(multica *) to query workspace data and can trigger other agents using the mention://agent/ URI scheme.
Sanitization: The backend parser enforces a strict regex for identifiers ([0-9a-fA-F-]+ or all), which limits the potential for arbitrary string injection into the URI, though it does not prevent the logic-based triggering of agents.
[COMMAND_EXECUTION]: The skill uses the multica CLI tool to perform lookups for members, agents, and squads. While these are vendor-provided tools, the skill instructions require the agent to execute shell commands (e.g., multica workspace member list --output json) to obtain necessary UUIDs for constructing mention links.

multica-mentioning