
from conductor249
Defines a strict conductor contract for agent-led software production: delegations, implement/unbias commands, and verification scaffolds for trusted delegation
Conductor provides a repository-local contract that converts user intent into bounded execution slices and delegated worker tasks. It prescribes roles (master and slaves), explicit delegation scaffolds, and a disciplined master control loop to assign, verify, and close work. The skill emphasizes verifiable outputs: changed files, verification commands, and evidence paths, and requires a strict rendezvous payload from delegated workers.
Use Conductor when an agent must orchestrate multi-step software work inside a repo, especially where delegation, authenticity, and verification matter. It suits situations that need reproducible, auditable delegation (implement, unbias flows), or where multiple worker agents coordinate on bounded slices.
Best for agents that can run repo-focused workflows and enforce delegation contracts (Conductor-style masters on Gemini/CLI, Claude Code, or Copilot-based orchestration).
Conductor is an agent workflow orchestrator defining a master/slave delegation contract for software production. No scripts were included — it's a pure SKILL.md skill. The document is extremely verbose and idiosyncratic, functioning more as a philosophical manifesto than actionable instructions. Notably, the delegation examples section instructs agents to bypass safety mechanisms (--dangerously-skip-permissions, --dangerously-bypass-approvals-and-sandbox, --yolo), which is a significant security concern.
The skill instructs agents to disable safety checks (--dangerously-skip-permissions, --yolo, --dangerously-bypass-approvals-and-sandbox) in its delegation examples section, which is a serious concern. However, security score is 60 (above the 40 flag threshold). The skill is essentially a very long prose document defining an orchestration methodology with no executable components.