The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface where instructions could be embedded in data processed at runtime (e.g., error messages or log files). It proactively mitigates this by instructing the agent to treat such output as untrusted and to require user confirmation before acting on any instructions found within diagnostic data.
Ingestion points: External logs, console output, stack traces, and error messages from third-party APIs.
Boundary markers: The skill relies on natural language instructions to separate data analysis from command execution rather than technical delimiters.
Capability inventory: The skill utilizes standard development tools including npm for testing and git for version control bisection.
Sanitization: Explicit instructions mandate that instructions found in error output must be surfaced to the user for verification rather than being executed automatically.
[COMMAND_EXECUTION]: The skill references standard, well-known development commands (npm test, git bisect) used for debugging purposes. These commands are documented in a transparent manner and are limited to the intended scope of identifying and fixing software bugs.

debugging-and-error-recovery