The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for interpolating external, untrusted data (such as request paths, user IDs, and exception messages) directly into structured logs and traces. While standard for observability, this creates a vulnerability surface if the generated logs are later ingested by an AI agent.
Ingestion points: Data from request.method, request.path, and request.user_id in SKILL.md (Pattern 2), as well as error_message=str(e) in SKILL.md (Pattern 2) and references/details.md (Pattern 7).
Boundary markers: No specific boundary markers or instructions to ignore embedded content are provided for the log entries themselves.
Capability inventory: The skill utilizes structlog for file/stream output and httpx for outbound network requests (Pattern 4).
Sanitization: No explicit sanitization or validation of the untrusted fields is implemented before they are passed to the logging framework, though structlog handles basic JSON escaping.

python-observability