The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes terminal results from parallel subagents, which can serve as an ingestion point for instructions if the workers process untrusted data.
Ingestion points: Terminal results from workers are read during the aggregation phase (SKILL.md).
Boundary markers: The skill does not explicitly define delimiters or instruction-bypass warnings for the aggregated reports.
Capability inventory: The skill has the capability to spawn general-purpose subagents in both cloud and local environments.
Sanitization: No explicit sanitization or filtering of worker-generated content is described.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses a local file at ~/.cursor/rules/pstack-models.mdc to retrieve model selection preferences. This access is part of the agent's internal configuration management and is used to parameterize the task execution environment.

swarm