The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external evidence including error messages, warnings, stack traces, and logs. These inputs represent an attack surface where malicious instructions could be embedded in data processed by the agent.
Ingestion points: Error logs, stack traces, and component boundary data (SKILL.md, Section: 'The Brief').
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore instructions embedded within the ingested logs/evidence.
Capability inventory: The skill requires the agent to modify source code (adding instrumentation) and execute commands to gather evidence or run tests.
Sanitization: No mention of sanitizing or escaping the content of logs or errors before processing.
[COMMAND_EXECUTION]: The process requires the agent to add diagnostic instrumentation and execute the system to gather evidence. It also relies on related tools like gangsta:drill-tdd and gangsta:sweep-verification for test execution and verification, which involve high-privilege interactions with the environment.

interrogation-debugging