The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains internal guidelines and "rationalization tables" that instruct the agent to resist user pressure for quick fixes and maintain a systematic debugging process. These are intended as quality-of-service guardrails for the agent's specific role rather than attempts to bypass system-level safety filters.
[COMMAND_EXECUTION]: The skill encourages the use of standard development tools such as Bash, Grep, and git for investigating codebases. It provides structured examples for running tests, checking logs, and using git bisect, all of which are routine developer activities.
[DYNAMIC_EXECUTION]: The instructions permit the creation of "throwaway harnesses"—small scripts designed to reproduce specific bugs. While this involves generating and running code, it is restricted to the reproduction and verification phases of debugging and is governed by strict hypothesis-driven requirements.
[INDIRECT_PROMPT_INJECTION]: The agent is designed to ingest and analyze external data like error logs, stack traces, and source code. While this represents a potential injection surface, the skill includes "investigation hygiene" guidelines that prioritize narrow evidence excerpts and specific stack frames, which helps minimize the risk of processing malicious content in bulk.
[DATA_EXFILTRATION]: The skill uses curl for testing HTTP responses during bug reproduction. There is no evidence of these tools being used to exfiltrate sensitive data or communicate with unauthorized external servers.

debugging