The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to analyze error messages, logs, and stack traces provided by the user. This data represents untrusted input that could potentially contain adversarial instructions (Indirect Prompt Injection) designed to manipulate the agent's logic during the debugging process.\n
Ingestion points: The skill explicitly requests "Complete error message content", "logs", and "reproduction steps" from the user in SKILL.md.\n
Boundary markers: While a structured question template is provided, the skill does not instruct the agent to use explicit delimiters or markers to isolate this untrusted content from the rest of the prompt context.\n
Capability inventory: The skill encourages the agent to use powerful system tools like the Python debugger (pdb), Node.js inspector, and Bash command tracing (set -x), all of which involve local code execution (SKILL.md, references/debugging-tools.md).\n
Sanitization: No logic is provided for sanitizing or escaping the user-provided logs or error messages before they are processed by the agent.\n- [COMMAND_EXECUTION]: The skill documents and provides examples for using various command-line tools for debugging and troubleshooting (e.g., bash -x, python -m pdb, and examples involving rm -rf). These tools are used legitimately for troubleshooting but represent a capability that should be exercised with caution, especially when reproduction steps involve running user-provided code snippets.

bug-detective