The Agent Skills Directory

[SAFE] (SAFE): The skill contains legitimate code snippets for the Sentry SDK. There is no evidence of prompt injection, data exfiltration, or obfuscation.
[Indirect Prompt Injection] (LOW): The skill processes untrusted external data (error messages, user metadata, breadcrumbs) which could be used as an injection vector if the agent later interprets this data as instructions.
Ingestion points: Methods like Sentry.captureException(error), Sentry.setUser(user), and Sentry.addBreadcrumb(data) in SKILL.md define where external data enters the context.
Boundary markers: None provided in the snippets to distinguish data from instructions.
Capability inventory: The skill allows powerful tools including Bash, Write, and Edit across all referenced files.
Sanitization: No sanitization or validation logic is demonstrated for the incoming error or context data.

sentry-error-capturing