The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from the local codebase, including CONTEXT.md, ADRs, and source files, to build its 'design tree' and research project facts. This behavior exposes a surface for indirect prompt injection, where malicious instructions embedded in the project files could theoretically influence the agent's behavior or decision-making process. This is considered a standard operational risk for development-focused agents.
Ingestion points: The agent reads from the filesystem root, specifically targeting CONTEXT.md, CONTEXT-MAP.md, docs/adr/*.md, and files within the src/ directory.
Boundary markers: The instructions do not define specific delimiters (e.g., XML tags or backticks) or provide warnings for the agent to ignore instructions found within the data it reads.
Capability inventory: The skill utilizes file reading, file creation/modification (writing CONTEXT.md and ADR files), and tool invocation to explore the repository.
Sanitization: No sanitization or validation of the content retrieved from the codebase is performed before it is processed.

grill-me