The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash for filesystem management and component creation. This is its intended functionality and is restricted by specific logic that prevents raw directory creation for plugins, requiring instead the use of audited scaffolding tools.
[PROMPT_INJECTION]: The skill processes untrusted external data from discovery plans to drive its build loop, which presents an indirect injection surface.
Ingestion points: Files in exploration/discovery-plans/ and the exploration/exploration-dashboard.md file.
Boundary markers: None explicitly present in the plan parsing logic.
Capability inventory: Filesystem Read/Write, shell execution via Bash, and invocation of external quality review skills.
Sanitization: The skill mitigates risks by requiring specific internal skills for plugin assembly and enforcing structural audits via the audit-plugin tool.

subagent-driven-prototyping