The Agent Skills Directory

[SAFE]: The skill primarily consists of educational content and best-practice guidelines for software debugging. It emphasizes methodology and rigorous investigation over automated script execution.
[COMMAND_EXECUTION]: The skill utilizes standard, benign development commands for code investigation, including git log, git diff, git bisect, and grep. These tools are used appropriately within the context of their intended purpose for debugging and tracing code changes.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill instructions in SKILL.md (Phase 1) require the agent to ingest external data such as error messages, stack traces, and application logs.
Boundary markers: There are no specific delimiters or markers defined to separate untrusted external data from the agent's instructions.
Capability inventory: The skill utilizes git and grep for file system and history analysis, but does not provide advanced capabilities like network writing or arbitrary code execution.
Sanitization: No explicit sanitization or filtering of external data (logs/errors) is specified. Given the limited capability of the tools involved, this is assessed as a low-risk surface.

systematic-debugging