The Agent Skills Directory

[SAFE]: The skill provides a rigorous debugging framework across six phases: building feedback loops, reproducing/minimizing, hypothesizing, instrumenting, fixing/testing, and cleaning up. All instructions focus on legitimate engineering practices.
[COMMAND_EXECUTION]: The skill includes a shell script template (scripts/hitl-loop.template.sh) designed for Human-In-The-Loop (HITL) debugging. The script contains benign helpers (step and capture) to guide a user through manual reproduction steps and collect feedback. It uses standard bash features (set -euo pipefail, read, printf) and does not perform any unauthorized network or filesystem operations.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill explicitly instructs the agent to redact secrets and credentials from outputs and artifacts before showing them to the user. It recommends using environment variables for credentials to prevent them from appearing in logs or captured artifacts, which aligns with security best practices.
[PROMPT_INJECTION]: The skill uses instructional language to guide the agent's behavior (e.g., "Be aggressive. Be creative. Refuse to give up."), but these are framed within the context of effective debugging and do not attempt to bypass platform safety guidelines or override core system instructions.
[EXTERNAL_DOWNLOADS]: No external downloads, package installations, or remote code execution patterns were detected. The skill relies on local environment tools (tests, curl, CLI, etc.) as determined by the user's specific project context.

diagnosing-bugs