The Agent Skills Directory

[SAFE]: The skill provides procedural guidelines for software debugging. It emphasizes evidence-based diagnosis and discourages making changes without understanding the underlying cause.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as bug reports, code, and test outputs to perform its function. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the code or logs being debugged.
Ingestion points: Bug reports, failing test logs, and existing codebases as described in the description and The procedure section.
Boundary markers: The instructions emphasize reading the "actual error" and "actual code path" which forces the agent to inspect raw data, but it does not specify explicit delimiters for external content.
Capability inventory: The procedure involves command execution (reproduction), code modification (instrumentation and fix), and test execution (regression testing) as noted in SKILL.md steps 1, 5, 6, and 7.
Sanitization: No specific sanitization or filtering of external input is mentioned, though the methodology of minimization and differential hypotheses provides a logical framework that may reduce the impact of injected instructions.

diagnose