The Agent Skills Directory

[DATA_EXPOSURE]: The scripts/collect-diagnostics.sh script gathers system metadata including OS version, architecture, the current user, and the current working directory. It also captures the Git state and language runtime versions. The script incorporates safety measures by redacting credentials from Git remote URLs and restricting environment variable collection to a predefined safe list (e.g., NODE_ENV, APP_ENV).
[DYNAMIC_EXECUTION]: The documentation in references/specialized-patterns.md describes advanced debugging techniques such as using LD_PRELOAD shims to interpose libc symbols for execution tracing. These are presented as standard engineering practices for diagnosing low-level defects and do not involve unauthorized execution or remote code retrieval.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted diagnostic data such as stack traces and logs. It mitigates injection risks by providing explicit instructions in SKILL.md and references/baseline-and-data-handling.md to redact PII, credentials, and SQL fragments before the agent processes or shares data externally.
[COMMAND_EXECUTION]: The skill requires the agent to run reproduction loops, which may involve executing shell commands, tests, or CLI harnesses provided in the user's environment. These operations are transparent, intentional, and central to the skill's primary purpose of debugging code.

debugging