The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill architecture is designed to ingest and process voluminous external data through subagents, creating a pathway for malicious instructions to reach the primary coordinator.
Ingestion points: The skill is triggered by tasks involving "bulk material" such as codebase sweeps, log triage, document reviews, and web research (found in SKILL.md).
Boundary markers: The skill defines a "worker brief" and "distilled-report contract" to shape output, but it explicitly advises the coordinator "Don't re-read what a worker read," which increases the risk of the coordinator executing instructions hidden within those summaries.
Capability inventory: The coordinator agent retains full authority for synthesis, planning, and decision-making based on the data provided by the subagents.
Sanitization: There are no instructions for the coordinator to sanitize, filter, or validate the content of the worker reports for injection patterns before processing them.

fable-orchestrator