Ai Dev Group

A universal, multi-role AI engineering team for autonomous planning, implementation, and rigorous code review.

Install
cmdop skills install agensi-ai-dev-group

AI Dev Group is a skill that introduces a role-based routing engine into an AI agent, replacing single generalist prompting with a coordinated set of specialized roles. The roles described include planners, implementers, critics, and researchers, each targeted at a distinct part of a software engineering workflow. Rather than asking one model to handle every concern at once, the skill dispatches tasks to the appropriate role depending on the nature of the work.

The skill maintains three discrete operational modes — P7, P9, and P10 — spanning a range from bounded, focused execution up to high-level strategic thinking. This allows an agent to shift between tightly scoped implementation tasks and broader architectural reasoning within the same session.

A stated design principle is “closure discipline”: work is not considered done when started, but must be verified and completed to a defined standard before the skill considers the task closed. This is paired with a fact-driven execution model intended to reduce hallucinated or incomplete outputs on complex tasks such as database migrations, security audits, and architectural shifts.

This skill targets developers who need autonomous handling of multi-step engineering work — not simple code completion. It is not a database connector, not a data pipeline tool, and carries no tools in its tool list, so it does not expose discrete callable functions to an agent runtime in the way an MCP server would.

Use cases

  • Coordinate autonomous planning and implementation of a multi-step database migration
  • Run a security audit pass over a codebase using the critic role
  • Decompose a large architectural refactor into role-specific subtasks
  • Delegate research tasks to a specialized researcher role before implementation begins
  • Apply closure discipline to ensure a complex task reaches a verified, production-ready state

When to use it

  • When a task is too complex for a single generalist prompt and requires distinct planning, implementation, and review phases
  • When working on migrations, security audits, or architectural changes that span multiple concerns
  • When an agent needs to enforce completion standards rather than just starting work

When not to use it

  • When looking for a Postgres or other database connector — this skill has no database integration
  • When the agent runtime requires callable tool functions; this skill lists no tools
  • When a simple, single-step code generation or autocomplete task is all that is needed
  • When a specific transport protocol (HTTP, stdio, etc.) is required; no transport is specified