The Agent Skills Directory

[SAFE]: The skill implements a structured workflow for feature development through task decomposition and delegation. The following analysis confirms its safety:
Workflow and Delegation: It serves as a plan-to-DAG executor, delegating specific building tasks to an internal implement-and-review skill. This separation of concerns is a standard architectural pattern.
Execution Environment: It uses standard Git commands for branch management (git branch, git merge, git rev-parse) and worktree isolation. These are benign operations within a development environment.
External Dependencies: The skill relies on local scripts (discover_runners.py, launch.py) located within the .agents/ directory, which are expected components of the agent framework.
Data Handling: No evidence of hardcoded credentials, sensitive file exfiltration, or unauthorized network operations was found. Communication with other models is handled through a discovery probe that verifies available 'seats' (model instances) before starting tasks.
User Control: The skill enforces a 'Hard Rule' where writes are gated; the agent must obtain user approval for the task breakdown before writing code, unless explicitly overridden by the --auto flag.
Security Integration: The orchestration logic includes a security_focus flag for the final feature-wide review if any sub-tasks were marked as security-sensitive during the decomposition phase.

implement-feature