The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes a dedicated 'Redact' section that mandates the removal of secrets and credentials from all commands, outputs, and captured artifacts. It specifically advises using environment variables to keep sensitive data out of logs, demonstrating a clear security-first approach to data handling.
[PROMPT_INJECTION]: The skill processes external data like logs, HAR files, and error messages (ingestion points in SKILL.md), creating an indirect injection surface. While specific boundary markers for prompt interpolation are not defined, the capability inventory (curl, git, playwright, and shell execution) is constrained by the debugging scope. Risk is managed through a mandatory sanitization phase requiring the redaction of all secrets with .
[COMMAND_EXECUTION]: The skill leverages standard development and debugging utilities like curl, git, and headless browser frameworks (Playwright, Puppeteer) to create deterministic feedback loops. These tools are used appropriately for reproduction and verification within the skill's intended scope.

diagnosing-bugs