The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill mandates careful analysis of error messages and stack traces, which serves as an ingestion point for untrusted data that could contain malicious instructions.
Ingestion points: Error messages, stack traces, and logs from external components or multi-layer systems (SKILL.md, Phase 1).
Boundary markers: The skill does not provide specific delimiters or instructions to treat data in error messages as non-instructional.
Capability inventory: The agent has the ability to execute shell commands, perform file system operations, and interact with development tools like npm and git.
Sanitization: There is no instruction to sanitize or escape data harvested from logs or errors before processing.
[COMMAND_EXECUTION]: The skill includes shell scripts and instructions that execute system commands for diagnostic purposes.
Evidence: The 'find-polluter.sh' script executes 'npm test' on files discovered via pattern matching.
Evidence: SKILL.md provides examples of using 'security list-keychains' and 'security find-identity' to troubleshoot code signing issues, which involves accessing system-level identity information.
Context: These operations are intended for debugging and environment verification, which is the primary purpose of the skill.

systematic-debugging