The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest design systems, token sets, and extracted style guides from potentially untrusted external sources, such as "non-technical sources" or "freshly researched" references. This creates a vulnerability surface where malicious instructions hidden in the design data could influence the agent's behavior during the reporting phase. Ingestion points: The skill in SKILL.md specifies reading external tokens, components, and style guides. Capability inventory: The agent is instructed to post results to GitHub issues. Boundary markers: No delimiters or isolation instructions are provided to separate the ingested data from the agent's internal instructions. Sanitization: No validation or filtering of the external content for embedded commands or instructions is mentioned.\n- [NO_CODE]: The skill consists of markdown-based process instructions and does not contain any executable scripts, source code, or configuration files that would perform automated actions.

design-composition-check