The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local codebase and external documentation sources to generate implementation plans.
Ingestion points: Research and Documentation retrieval steps in SKILL.md process codebase files and external library/API documentation.
Boundary markers: The skill does not define explicit delimiters or warnings to the model to ignore instructions embedded within the ingested codebase or documentation content.
Capability inventory: The skill has the capability to write markdown files to the local directory (Save Plan step) and spawn subagents (Subagent Review step) using the processed information.
Sanitization: No sanitization or validation logic is defined for the external content before it is used to generate the plan or instructions for the subagent.

swarm-planner