The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes the creation of job processors that ingest external, untrusted data via the job.data payload. This creates an indirect prompt injection surface where instructions embedded in job data could potentially influence the agent's behavior during task execution.
Ingestion points: Job data is ingested by workers in SKILL.md and references/setup-connections.md.
Boundary markers: The skill does not explicitly provide delimiters for prompt interpolation but does emphasize data handling precautions.
Capability inventory: The skill has capabilities for network communication with Redis/PostgreSQL databases and executing code via job processors.
Sanitization: The documentation recommends encrypting sensitive fields and explicitly states that sandboxed processors are not a security boundary for untrusted code.
[DYNAMIC_EXECUTION]: The skill provides instructions for implementing sandboxed processors and using worker threads. These are standard features in BullMQ for isolation and performance but involve the execution of logic based on specified file paths at runtime.
Evidence: Described in the 'Sandboxed Processors' section of references/setup-connections.md.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to install standard library dependencies from public registries and references official documentation from the BullMQ project.
Evidence: Installation commands in references/setup-connections.md and documentation links to docs.bullmq.io and GitHub in references/source-map.md.

bullmq