The Agent Skills Directory

[SAFE]: The skill follows professional software engineering practices for debugging and does not contain any instructions that attempt to bypass security filters, exfiltrate sensitive data, or hide malicious intent.
[COMMAND_EXECUTION]: The skill mentions the use of common developer utilities such as curl, git bisect, and testing frameworks (Playwright, Puppeteer) as part of Phase 1 (Build a feedback loop). These commands are standard for the described purpose of software diagnosis.
[PROMPT_INJECTION]: The skill contains instructional guidance for the AI agent (e.g., "Refuse to give up", "Stop and say so explicitly") that is consistent with its intended functional role and does not constitute a malicious override of system instructions.

diagnosing-bugs