The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to have multiple agents independently explore a codebase and process a raw user prompt. This creates a risk where malicious instructions embedded in the codebase (e.g., within comments, documentation, or string literals) could influence the behavior of the solver agents.
Ingestion points: The solver agents ingest the user's raw prompt and read codebase content using Read, Grep, Glob, and LS tools (SKILL.md, Workflow Step 3).
Boundary markers: The skill instructions do not specify any delimiters or instructions for the agents to ignore embedded commands within the files they read.
Capability inventory: The skill allows agents to read files and identify root causes, with the final workflow step involving the implementation of the winning code solution (SKILL.md, Workflow Step 6).
Sanitization: There is no evidence of sanitization or filtering of codebase content before it is processed by the agents.

debug-council