The Agent Skills Directory

[DATA_EXFILTRATION]: The skill recommends logging sensitive 'Business Context' and 'Environment Characteristics' to enable effective debugging.
Evidence: rules/context.md lists user.email, lifetime_value_cents, commit_hash, and instance_id as standard fields for inclusion in wide events.
Risk: Logging personally identifiable information (PII) and detailed infrastructure metadata increases the risk of information disclosure if log storage or transmission is compromised.
[PROMPT_INJECTION]: The logging pattern involves capturing untrusted data from incoming requests, which creates a vulnerability to indirect prompt injection.
Ingestion points: rules/structure.md and rules/wide-events.md capture c.req.json() and user_agent header data.
Boundary markers: There are no instructions for using delimiters or boundary markers when logging user-provided strings.
Capability inventory: The skill is instructional and does not include active scripts with dangerous capabilities, but it outlines the creation of data that could be dangerous to downstream consumers.
Sanitization: The provided code examples lack sanitization or validation of user-controlled input before inclusion in the log payload.

logging-best-practices