The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The file references/github-skill-candidates.md provides a directory of multiple third-party GitHub repositories (e.g., ComposioHQ, addyosmani, Dimillian, Orchestra-Research, shepsci, FrankS-IntelLab) and instructs the agent to facilitate the installation of skills from these sources. Since agent skills consist of executable instructions that modify agent behavior, recommending the installation of code from unverified third-party repositories poses a supply chain risk.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data to make routing decisions for downstream workflows that possess high-privilege capabilities.
Ingestion points: Untrusted data enters the agent context through user-provided intents and task requirements (e.g., SKILL.md, references/runtime.md).
Boundary markers: The skill employs defined output templates for alignment (e.g., ALIGN and LOW Collaborative Calibration in references/output-forms.md) and an 'Intent Lock' mechanism to confirm understanding before execution.
Capability inventory: The skill defines and delegates actions including WRITE, REMOTE_WRITE, and DESTRUCTIVE operations to downstream owners (SKILL.md).
Sanitization: There is no evidence of explicit sanitization or filtering of external input before it is utilized in routing logic or passed to sub-tasks.

briefbound-router