agent-group-evolving-molecular-system-AGEM

Formal reasoning via FOL theorem proving, model finding, and categorical logic

Use modern, token-efficient command-line tools (fd, ripgrep, ast-grep, bat, sd, jq, yq, tokei, eza, dust, procs, delta, hyperfine, just) instead of legacy coreutils (find, grep, cat, sed, ls, du, ps) for any shell, file, or codebase task — on any platform (Linux/macOS/Windows) and through any executor (a terminal, Claude Code's bash tool, Desktop Commander, or an MCP shell bridge). Trigger whenever you are about to search files, grep for text, read or inspect files, find-and-replace, refactor code, traverse a repository, process JSON/YAML, or inspect disk usage or processes — even if the user did not name a specific tool. These tools are faster, .gitignore-aware, and emit clean structured output that protects the context window. For installation across operating systems and platform-specific quirks, see references/. If you run commands specifically through the LocalREPL MCP bridge, also consult the local-repl-cli skill for environment-specific gotchas.

Operating guide for the AGEM reasoning engine: native tools, honest metric interpretation, formal-logic verification via mcp-logic, and the meta-tool MCP interface.

Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract, or otherwise improve a frontend interface. Covers websites, landing pages, dashboards, product UI, app shells, components, forms, settings, onboarding, and empty states. Handles UX review, visual hierarchy, information architecture, cognitive load, accessibility, performance, responsive behavior, theming, anti-patterns, typography, fonts, spacing, layout, alignment, color, motion, micro-interactions, UX copy, error states, edge cases, i18n, and reusable design systems or tokens. Also use for bland designs that need to become bolder or more delightful, loud designs that should become quieter, live browser iteration on UI elements, or ambitious visual effects that should feel technically extraordinary. Not for backend-only or non-UI tasks.

Master orchestrator for the five-question decision tree that selects the right agentic pattern from Sequential, ReAct, Planning, Reflection, or Multi-Agent.

Multi-agent pattern for tasks requiring specialization or scale beyond single-agent capacity. Covers topology, routing, and coordination design.

Planning pattern for structurally articulable tasks. Planner defines stages and dependencies upfront; ReAct handles local uncertainty within each step.

ReAct (Reasoning + Acting) pattern for adaptive, exploratory tasks with unknown solution paths. Iteratively alternate between reasoning and tool actions.

Reflection pattern for quality-critical outputs using generate-critique-refine cycles. Requires clear evaluation criteria and acceptable latency.

Sequential workflow pattern for deterministic, fixed-path agentic tasks. Use when solution steps are known upfront and the process follows a stable, repeatable sequence.

Tool use pattern for agents requiring external world interaction. Foundation layer for most real-world agentic tasks with design principles for effective tool contracts.

Hierarchical supervisor pattern: Multi-level supervisor chains where supervisors delegate to supervisors who delegate to workers.

Parallel execution pattern: Multiple agents work simultaneously on independent sub-tasks. Best for embarrassingly parallel workloads.

Single-agent pattern: One LLM handles all tasks without sub-agent delegation. Use when simplicity beats sophistication.

Supervisor delegation pattern: One supervisor LLM routes tasks to specialized sub-agents. Best for multi-capability routing.

Supervisor-orchestrator pattern: Orchestrator controls task planning and delegation to specialized agents with full decomposition logic.

Complete AGEM reference: nine conflation distinctions, three conceptual bridges, the full Price equation + SOC regime table, Move B bridge formalization, and live cycle 11 evidence.

Hard-won patterns for running AGEM on philosophical/ethical topics without falling into the AI Comfort Trap or consuming grief as evidence. Based on the Duke conversation.

EFHF three-tier ethics framework with Omega1-4 axioms, Paraclete Protocol, and sheaf-based value-drift detection. Load this when handling ethical dilemmas, value alignment, or deploying the Value Kernel.