The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GOAL.md which is then interpolated into the generated IDEA_BRIEF.md and queries.md files.
Ingestion points: The topic is extracted from GOAL.md in scripts/run.py and used to populate various sections of the brief.
Boundary markers: There are no delimiters or warnings to downstream agents to ignore embedded instructions within the interpolated topic.
Capability inventory: The script writes multiple markdown and configuration files to the workspace which are intended to guide the behavior of subsequent agents in the pipeline.
Sanitization: No validation or escaping is performed on the user-provided topic before it is formatted into the output files, allowing potential markdown-based injection.

idea-brief