The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill directs the agent to load and follow instructions from local file paths (e.g., skills/keyword-research/SKILL.md) based on user input. This dynamic loading is restricted by a static routing table that acts as a whitelist, preventing the agent from accessing unauthorized files or resolving arbitrary paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes natural language input from users to select which internal sub-skills to activate. While it handles untrusted data, the vulnerability surface is effectively mitigated by the hardcoded mapping logic, which ensures only pre-approved internal workflows can be triggered. The analysis of the ingestion point (user message) against the capability (local file reading) confirms that strict sanitization via the routing table is in place.

aso-router