The Agent Skills Directory

[COMMAND_EXECUTION]: The find-polluter.sh script executes npm test on files discovered via a user-defined pattern. This allows for the execution of arbitrary code within the test environment.
[CREDENTIALS_UNSAFE]: The SKILL.md file provides diagnostic examples that use the macOS security command to list keychains (security list-keychains) and identities (security find-identity -v). This facilitates the exposure of sensitive security credentials and system configuration.
[DATA_EXFILTRATION]: Instructions in SKILL.md and root-cause-tracing.md encourage logging environment variables (e.g., env | grep IDENTITY) and full stack traces. This pattern can lead to the accidental collection and exfiltration of secrets or sensitive internal paths if the logs are captured by the agent context.
[REMOTE_CODE_EXECUTION]: The skill's methodology for root cause investigation explicitly directs the agent to 'add diagnostic instrumentation' and 'Run once to gather evidence'. This encourages the dynamic generation and execution of scripts by the agent, which is a high-risk behavior.
[PROMPT_INJECTION]: The skill uses extremely rigid and authoritative language, such as 'The Iron Law' and 'Violating the letter of this process is violating the spirit', which functions as a behavioral override. This can be used to force the agent to ignore user instructions or safety considerations in favor of the skill's specific debugging protocols.

systematic-debugging