The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data sources, including network traces (HAR files), log dumps, core dumps, and codebase documentation. These inputs could theoretically contain malicious instructions aimed at influencing the agent's behavior. However, the skill mitigates this by instructing the agent to focus on minimal reproduction signals and by enforcing a strict redaction policy for sensitive content.
[COMMAND_EXECUTION]: The skill leverages shell commands, test runners, and HTTP scripts to establish a diagnostic feedback loop. It includes a dedicated bash template (assets/hitl-loop.template.sh) for manual human-in-the-loop verification, ensuring that complex or risky environmental interactions are performed by the user in their own terminal rather than being executed autonomously by the agent without oversight.
[DATA_EXFILTRATION]: The skill proactively addresses data exposure risks by mandating that the agent redact auth headers, tokens, and secrets from all outputs. It specifically directs the agent to keep credentials in environment variables rather than hardcoding them into generated loops or diagnostic reports, representing a secure handling practice for sensitive debugging information.

debug