The Agent Skills Directory

[DATA_EXPOSURE]: The skill instructs the agent to read worker model configuration from specific paths in the user's home directory (~/.cursor/rules/pstack-models.mdc or ~/.agents/pstack-models.md). While these are intended as configuration files for the agent harness, reading from hidden directories in the user profile is a sensitive operation.
[INDIRECT_PROMPT_INJECTION]: The skill implements a pattern where a parent agent ingests data produced by multiple autonomous subagents, creating a surface for indirect instructions to influence the parent agent's behavior.
Ingestion points: The parent agent reads terminal results and logs from N parallel workers during the aggregation phase (Phase C).
Boundary markers: The instructions do not specify the use of strict boundary delimiters for the worker output, though it does advise against using raw worker dumps.
Capability inventory: The skill uses powerful subagent orchestration tools (such as Task, Agent, or spawn_agent) that can execute code and access files.
Sanitization: The skill advises providing a consolidated report with one-line summaries rather than raw output, which provides a layer of manual filtering but does not programmatically sanitize potential prompt injection content embedded in worker results.

swarm