The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting user-defined tasks and codebase artifacts to coordinate subagent activities, creating a surface for potential indirect prompt injection. The system includes roles capable of file modification and command execution.
Ingestion points: Identified in SKILL.md (Step 1) and role-explorer.md, where the skill frames tasks based on user objectives and gathers codebase context.
Boundary markers: The orchestration logic in SKILL.md (Steps 1, 4, and 6) explicitly defines safety boundaries, approvals, and forbidden areas for each subagent roster.
Capability inventory: The tester role possesses command-execution capabilities, while implementer and integrator roles are authorized for file writes under disjoint-write and designated-writer policies.
Sanitization: Mitigation is achieved through task decomposition, role-based scoping, and a requirement in role-tester.md to seek approval for risky commands.
[EXTERNAL_DOWNLOADS]: The skill's documentation directs users to install the skill from https://skills.flc.io or the author's GitHub repository. These endpoints are the primary distribution channels provided by the author and are consistent with the skill's metadata.

subagent-orchestrator