The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves processing external, potentially untrusted data such as error messages, stack traces, and system logs to identify bugs.\n
Ingestion points: Phase 1 of the skill directs the agent to read error messages and gather evidence from multi-component systems.\n
Boundary markers: None explicitly defined for untrusted log data.\n
Capability inventory: The skill suggests command execution and file system inspection for diagnostic purposes.\n
Sanitization: None described for external log content.\n- [COMMAND_EXECUTION]: The skill includes a shell script find-polluter.sh that executes local tests via npm test and includes examples for running diagnostic system commands.\n
Evidence: SKILL.md provides examples for running security list-keychains and codesign. The find-polluter.sh script executes npm test \"$TEST_FILE\".

systematic-debugging