The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting data from several untrusted local files.
Ingestion points: The skill reads configuration and context from data/product/strategy.json, data/product/roadmap.json, data/gtm/icp_profiles.json, data/gtm/project_context.json, data/cfo/latest_forecast.json, data/product/competitive_analysis.json, and CLAUDE.md.
Boundary markers: No delimiters or instructions are provided to the agent to distinguish between the skill's core instructions and potentially malicious instructions embedded within these data files.
Capability inventory: The skill is authorized to perform filesystem write operations (specifically writing to data/product/product_scorecard.json and creating directory structures).
Sanitization: There is no evidence of validation or sanitization of the content loaded from the external files before it is processed by the agent.

cpo