The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by design, as it involves reading and processing data from local planning files that may contain untrusted content.
Ingestion points: The skill is designed to ingest data from project artifacts including docs/planning/overview/ideas-issues.md, package.json, and various source files within the src/ directory.
Boundary markers: The instructions and implementation lack explicit delimiters or specific instructions to the agent to disregard potential natural language instructions or commands embedded within the ingested planning data.
Capability inventory: The implementation includes file system read capabilities (fs.readFileSync, fs.readdirSync) and describes workflows involving file creation and modification (fs.copyFile, fs.mkdir) for updating the MASTER_PLAN.md and archiving project ideas.
Sanitization: No validation or filtering mechanisms are implemented to detect or handle potentially malicious instructions embedded in the raw text of the processed ideas before they are integrated into core project documentation.

chief-architect