The Agent Skills Directory

Indirect Prompt Injection (SAFE): The skill implements patterns for capturing application errors and metadata. While this creates a surface where untrusted data (like error messages or user IDs) is processed and sent to an external service, the risk is mitigated by explicit safety instructions.
Ingestion points: Data enters the telemetry stream via Sentry.captureException and helper methods described in SKILL.md.
Boundary markers: The skill contains multiple warnings (e.g., '❌ NEVER expose sensitive data in error context') and a checklist to prevent PII/sensitive data leakage.
Capability inventory: The skill uses the @sentry/node package to transmit data to the configured DSN.
Sanitization: Instructions require manual validation to ensure no sensitive content is interpolated into error logs.
Note: Severity is downgraded to SAFE as the surface is fundamental to the skill's primary purpose and includes sufficient mitigation guidance.
Data Exposure & Exfiltration (SAFE): The skill facilitates sending application data to an external Sentry endpoint.
Evidence: The skill provides initialization and capture patterns in SKILL.md using the @sentry/node library.
Mitigation: The skill correctly utilizes environment variables (e.g., process.env.SENTRY_DSN) and configuration placeholders instead of hardcoded secrets, and it includes robust warnings against the exposure of sensitive data.

error-tracking