The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's core methodology involves generating and running various shell commands, scripts (like the provided hitl-loop.template.sh), and testing tools (curl, Playwright, Puppeteer) to diagnose software issues. This involves high-privilege operations that could be misused if the agent is not carefully guided.
[DYNAMIC_EXECUTION]: The skill guides the agent to generate throwaway scripts, harnesses, and browser automation code at runtime to facilitate bug reproduction and testing.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, including terminal output, error logs, and user responses, to drive its logic. This data could potentially contain malicious instructions intended to manipulate the agent's behavior.
Ingestion points: User input captured by scripts/hitl-loop.template.sh and output from reproduction scripts in Phase 1.
Boundary markers: The bash script uses a '--- Captured ---' header to delimit data, which helps the agent distinguish between tool execution and data collection.
Capability inventory: The agent is encouraged to use a wide array of powerful command-line tools and scripting environments.
Sanitization: The skill includes mandatory instructions to redact secrets and auth headers from any captured output, significantly reducing the risk of accidental data exposure.

diagnosing-bugs