The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves processing potentially untrusted data from external sources such as network traces, event logs, and user-provided error messages during the debugging process. This data could be manipulated to contain instructions targeting the agent.
Ingestion points: Captured network traces, log files, and user input via the capture function in the HITL bash script.
Boundary markers: The instructions do not specify the use of clear delimiters or 'ignore' instructions when processing these artifacts, although redaction is mandated.
Capability inventory: The skill allows for the execution of CLI tools (curl, git) and the generation/execution of scripts (bash, Node.js).
Sanitization: The instructions explicitly require the agent to redact secrets and auth headers from any captured artifacts.
[DYNAMIC_EXECUTION]: The skill directs the agent to create and run various scripts to establishing reproduction loops.
Evidence: Instructions suggest creating throwaway harnesses, headless browser scripts (Playwright/Puppeteer), and custom bash scripts for human-in-the-loop interactions.
[COMMAND_EXECUTION]: The skill requires the execution of multiple command-line tools for testing and reproduction.
Evidence: Direct instructions to use curl, git bisect, and general CLI invocations to verify bug reproduction and fixes.

zj-diagnosing-bugs