The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external data, including error messages, stack traces, and source code files, which creates a surface for indirect prompt injection.
Ingestion points: The agent gathers data using read_file, search_files, web_search, and terminal output (e.g., from pytest or git).
Boundary markers: The skill does not define specific delimiters or instructions to ignore malicious commands that might be embedded in analyzed files or error logs.
Capability inventory: The agent is granted access to the terminal for executing commands, as well as file read/write tools.
Sanitization: There is no mention of sanitizing, escaping, or validating the external content before it is processed by the agent's logic.

systematic-debugging