The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various commands including curl, git bisect, and CLI tools to verify bug signals and automate bisection. It also refers to a shell script scripts/hitl-loop.template.sh for user-interactive reproduction loops.
[DYNAMIC_EXECUTION]: The skill encourages the creation and execution of temporary reproduction scripts (e.g., bash, python, Playwright/Puppeteer scripts) and throwaway harnesses to exercise specific code paths in isolation.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant surface area for indirect injection as it is designed to ingest and process untrusted external data such as network traces and logs.
Ingestion points: The skill reads GLOSSARY.md, ADRs, and captured artifacts like HAR files, log dumps, and core dumps (referenced in SKILL.md).
Boundary markers: None explicitly mentioned for separating data from instructions when replaying traces or logs.
Capability inventory: The agent is empowered to perform file writing, network requests via curl, browser automation, and shell command execution across various scripts.
Sanitization: The skill explicitly mandates redacting secrets from output, but does not provide specific guidance on sanitizing instructions or malicious code embedded within the processed artifacts.

diagnosing-bugs