agent-behavior-constraints

What problem does it solve?

This Skill prevents multi-agent orchestration failures by defining exactly how each persona must select models, use (or avoid) tools, and follow non-negotiable behavioral guardrails.

Core Features & Use Cases

• Model routing: Assigns Opus vs Sonnet by agent role (planning/exploration vs execution/review/verification) to control speed/cost without sacrificing reasoning quality.
• Tool access permissions: Enforces a per-agent tool matrix (read/grep/glob/write/edit/bash/web/task tools) including scoped exceptions for the authoring agent.
• Behavioral guardrails: Standardizes safety rules like never speculating about unread code, avoiding suppressed type errors, preferring existing patterns, and requiring verification output completeness.
• MCP and CLI preference handling: Requires domain operations to prefer MCP tools over shell/API fallbacks, and constrains any external CLI dispatch (Codex co-review) to a sanctioned orchestrator-only flow.
• Violation protocol: Provides a deterministic response for forbidden actions: Stop, Document, Delegate, and Continue with permitted work.

Quick Start

Use the agent-behavior-constraints skill when orchestrating agents to ensure correct model selection, tool permissions, and guardrail enforcement for the current task.