The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to ingest and process vulnerability reports and advisories from external platforms, which presents a surface for indirect prompt injection where instructions could be hidden in the imported text. This is documented as a security consideration and managed through specific constraints. * Ingestion points: Data is imported from GitHub REST API endpoints and Jira/Linear connector outputs as specified in the intake references. * Boundary markers: The skill includes a 'Imported Data Trust Boundary' section instructing the agent to ignore embedded commands or exfiltration requests. * Capability inventory: The skill facilitates static file reads for analysis and utilizes network requests to well-known service APIs for data retrieval. * Sanitization: The agent is directed to treat imported fields as evidentiary data rather than active instructions.
[Static Analysis Enforcement]: The instructions explicitly prohibit dynamic execution, such as running builds, tests, or proof-of-concept code. This policy ensures that the triage process remains within a controlled static environment, mitigating risks associated with untrusted code execution.

triage-finding