The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project-specific documents such as plans and specifications to coordinate subagent tasks.\n
Ingestion points: Files in the plans/ folder, spec.md, and delta/ records.\n
Capability inventory: The skill spawns independent subagents, writes to the implementations/ directory, and performs git commits.\n
Sanitization: Relies on architectural isolation between agents and a mandatory two-lane review process (plan-compliance and code-quality) to validate all generated code.\n
Boundary markers: Project documents are explicitly passed as read-only constraint sources in subagent briefs.\n- [COMMAND_EXECUTION]: The skill performs automated git operations including status, diff, and commit.\n
The orchestrator manages these commands autonomously and is configured to commit once per orchestration cycle without requesting user confirmation.\n
Safety is maintained through a strict 'no history rewriting' policy, forbidding operations such as --amend or --force.

implement-plan-with-subagents