The Agent Skills Directory

[SAFE]: The skill provides documentation and code templates for error handling and recovery. It does not include executable code, network operations, or sensitive data access. The patterns described follow standard software engineering practices for resilience.
[INDIRECT_PROMPT_INJECTION]: The skill provides logic for classifying and logging error messages from external services, which is a potential surface for indirect injection if an external service returns a malicious payload in an error response.
Ingestion points: modules/classification.md (via classify_by_message function) and modules/recovery-strategies.md (via log_error_with_context function).
Boundary markers: No delimiters or explicit warnings against embedded instructions are used in the provided code templates.
Capability inventory: No executable capabilities (subprocess calls, exec/eval, file writes) were found in the skill's modules.
Sanitization: No sanitization or escaping of the error message content is demonstrated in the examples. The risk is categorized as safe because the skill itself is instructional and does not provide an exploitable execution path.

error-patterns