The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes scripts/hitl-loop.template.sh, which is intended to be executed by the agent to facilitate a structured manual reproduction loop. The script uses standard shell commands (read, printf) to interact with the user and does not perform any suspicious background operations.
[PROMPT_INJECTION]: The scripts/hitl-loop.template.sh script represents an ingestion surface for indirect prompt injection (Category 8). It captures raw user input via the read command and outputs it as key-value pairs for the agent to parse. While this allows for potentially malicious instructions to be passed from the user into the agent's context, this is a standard risk for interactive debugging tools and is mitigated by the agent's own safety guardrails.
[SAFE]: The instructions in SKILL.md follow software engineering best practices for debugging, such as minimizing reproduction cases and using targeted instrumentation. All suggested actions (CLI invocations, test runners, headless browsers) are consistent with the skill's stated purpose of diagnosing technical issues.

diagnose