The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute various system commands and developer tools for bug diagnosis, including curl for network debugging, CLI invocations for system interaction, and browser automation frameworks like Playwright or Puppeteer (SKILL.md).
[INDIRECT_PROMPT_INJECTION]: The workflow involves processing data from external sources that may be untrusted, creating a surface for indirect prompt injection. Ingestion points: Data is ingested from application logs, network request traces, and HTTP responses during the feedback loop construction phase (SKILL.md), as well as via human input captured by the scripts/hitl-loop.template.sh utility. Boundary markers: The instructions do not mandate the use of delimiters or specific boundary markers when the agent processes these external data sources. Capability inventory: The agent is authorized to run tests, execute CLI commands, and perform browser automation, providing a significant capability tier for any potential injection to target (SKILL.md). Sanitization: There are no explicit instructions for the agent to sanitize or validate the content of logs, traces, or HTTP responses before analysis.

diagnosing-bugs