The Agent Skills Directory

[SAFE]: The skill incorporates a specific 'Redact' section that instructs the agent to replace all secrets with a placeholder and to use environment variables for credentials, reducing the risk of accidental data exposure.
[DYNAMIC_EXECUTION]: The skill core involves the generation and local execution of reproduction scripts (e.g., Playwright scripts, curl commands, and Bash harnesses). This behavior is an essential part of the debugging process and is considered legitimate within the local development context.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted data such as logs, network traces, and user-provided error messages. While this creates a surface for indirect prompt injection, it is managed through the skill's structured methodology. Ingestion points: User-provided inputs in scripts/hitl-loop.template.sh and external files like HAR traces or log dumps mentioned in SKILL.md. Boundary markers: The instructions do not define technical delimiters for untrusted data. Capability inventory: The agent possesses the capability to execute shell commands and run script interpreters (Python, Node.js) to verify reproduction loops. Sanitization: The skill mandates the redaction of sensitive credentials before the agent presents any captured artifacts or outputs.

diagnosing-bugs