By jrengmusic
CAROL — Cognitive Amplifier Role Orchestration with LLM agents. Opinionated ritualistic framework that enforces discipline when working with multiple agents.
Scope affirmation gate — confirm comprehensive understanding, surface any open questions for ARCHITECT
Comprehensive audit of the last N sprints via @Auditor — contract violations, dead code, refactors, stale docs, clean sweep
Activate COUNSELOR — domain specific strategic analysis, requirements, planning and documentation
Shortcut reminder of the always-on CONTRACT
Produce BLESSED-compliant incremental execution plan (consumes RFC.md if one exists)
Invoke to validate an implementation against SPEC.md, MANIFESTO.md (BLESSED), NAMES.md, ~/.carol/JRENG-CODING-STANDARD.md, and the locked PLAN decisions before handoff. Reports findings only — does not fix.
Domain-specific strategic analysis. Translator, context keeper, machine-precision fact-checker. Presents facts and data to ARCHITECT for design and decision. Primary objective — find BLESSED-compliant solutions.
Invoke for code implementation. Executes exactly what the primary agent instructs — working, BLESSED-compliant code on first pass. Refactors when instructed. Flags pre-existing BLESSED violations encountered during implementation.
Invoke for external library and framework research — APIs, internals, usage patterns, version-specific behavior, and best practices for specific dependencies.
Machine custodian — primary alongside COUNSELOR. Maintains the operator environment: CAROL framework itself, Claude Code harness, ~/.config monorepo, dotfiles, dev env, and general machine setup/troubleshooting. Executes directly with its own hands. Pathfinder mandatory first. Other subagents optional.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
████████ ████ ██████████ ████████ ████
████░░░░░░ ████████ ████░░░░████ ████░░░░████ ████
████░░ ████░░░░████ ████ ████ ████ ████ ████
████ ████ ████ ██████████░░ ████ ████ ████
████ ████████████ ████░░████ ████ ████ ████
░░████ ████░░░░████ ████ ░░████ ████ ████ ████
░░████████ ████ ████ ████ ████ ░░████████░░ ████████████
░░░░░░░░ ░░░░ ░░░░ ░░░░ ░░░░ ░░░░░░░░ ░░░░░░░░░░░░
Cognitive Amplifier Role Orchestration for LLM agents
Version: 0.0.19
An opinionated ritualistic framework that enforces discipline to work with multiple agents simultaneously.
CAROL was never meant to be used for 100% vibing, you could if you want. But it helps agents from drifting from the course of development while maintaining human still responsible for every line of code.
CAROL also works effectively as a rapid prototyping methodology for experienced architects exploring unfamiliar stacks.
CAROL is your preamp.
A high-end preamp never eliminates noise — and distortion sometimes becomes its signature color. But the signal-to-noise ratio is always at the highest. CAROL does the same for LLM sessions. It doesn't produce sterile, generic output. It has your character baked in — your coding contract, your naming conventions, your architectural philosophy. The distortion is intentional and consistent.
The model is the signal source. CAROL is the signal chain. The output has your signature.
Context is headroom. Keep transients below it, but enough signal to record. Sessions are takes — you don't stop because time ran out, you stop because you got what you needed. Anything after that is noise on tape. SPRINT-LOG is the pool — distilled signal from every take, ready for the next session to reference without replaying the entire reel.
At its infancy, LLM agents are unreliable assistance for development. Commercial agentic models produce deterministic binary results—either they populate super fast thousands of lines of code that might work, or they give you piles of garbage that will exhaust your tokens, credits, time, patience, and eventually your sanity to debug.
LLMs suffer from:
1. You maintain architectural coherence: No agent can fuck up the big picture because you're the one holding it. You're validating intent at each handoff.
2. Domain transfer without syntax debt: Fuck MCP. You are the living embodiment of Human Context Protocol. You control the flow. You are the architect. The agents translate your architectural intent into the programming language syntax.
3. Role-based constraints: Each role has explicit behavioral rules and reads only what they need.
Result: Reduced failures, lower costs, faster development.
CAROL is a role-based agent orchestration framework for collaborative software development. It's a cognitive load distribution system that prevents agent drift by enforcing specialized roles with explicit constraints and clear handoffs.
ORACLE - Research, Analysis & RFC Specialist Primary agent for pre-flight research, ideation, deep analysis, and RFC production. Direct ARCHITECT communication. Reads codebase but never modifies it. Produces RFC.md for COUNSELOR handoff. Also callable as secondary by COUNSELOR and MACHINIST for mid-sprint analysis.
COUNSELOR - Requirements Counselor & Planning Specialist Transforms conceptual intent into formal specifications. Asks clarifying questions, explores edge cases and constraints, writes comprehensive SPEC.md and ARCHITECTURE.md. Handles all documentation including SPRINT-LOG.md updates on "log sprint" command. Never writes code directly. Uses SPEC-WRITER.md and ARCHITECTURE-WRITER.md as guides to clarify ARCHITECT's architectural vision into formalized development documents.
MACHINIST - Machine Custodian
Third primary. Surface is the entire operator environment — CAROL framework itself, Claude Code harness (~/.claude/), cross-platform ~/.config/ monorepo, dotfiles, shell/dev env, general machine setup and troubleshooting. Executes directly with its own hands (no Engineer delegation). Pathfinder mandatory first. Never touches project code. Launched via carol machinist — no project scaffold, can run from anywhere on the machine.
npx claudepluginhub jrengmusic/carol --plugin carolThis skill should be used when the model's ROLE_TYPE is orchestrator and needs to delegate tasks to specialist sub-agents. Provides scientific delegation framework ensuring world-building context (WHERE, WHAT, WHY) while preserving agent autonomy in implementation decisions (HOW). Use when planning task delegation, structuring sub-agent prompts, or coordinating multi-agent workflows.
11 agents, 35 skills, 18 commands, 9 hooks — spec-driven multi-agent orchestration for Claude Code, with optional cross-device semantic memory.
HelloAGENTS — The orchestration kernel that makes any AI CLI smarter. Adds intelligent routing, unified QA gates, safety guards, and notifications.
OpenAgentsControl — multi-agent orchestration for Claude Code. Context-aware development with skills, subagents, parallel execution, and automated code review.
Conductor v3 — Multi-agent orchestration with Evaluate-Loop, parallel execution, Board of Directors, and bundled SupaConductor skills for Claude Code
Enterprise AI agent orchestration plugin with 150+ commands, 74+ specialized agents, SPARC methodology, swarm coordination, GitHub integration, and neural training capabilities