The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze logs, error messages, and stack traces, which are untrusted external data sources that could contain malicious instructions.
Ingestion points: Diagnostic data such as server logs, console output, and error stack traces are processed by the agent in SKILL.md.
Boundary markers: The skill proactively addresses this risk with a 'Treating Error Output as Untrusted Data' section, explicitly instructing the agent to treat these as data and not instructions.
Capability inventory: The skill uses shell-based capabilities via npm and git for testing and debugging.
Sanitization: Instructions explicitly direct the agent to surface any embedded instructions to the user rather than executing them.
[COMMAND_EXECUTION]: The skill utilizes standard development shell commands for its primary function.
Evidence: Employs git bisect for regression tracking and various npm test flags (e.g., --grep, --verbose) for verification.
Context: These commands are standard for software development and are used within their intended functional context without evidence of misuse.

debugging-and-error-recovery