Best Claude Code agents and subagents

Competitive ruliology for Ruflo swarms — arenas, tournaments, and adaptive co-evolution of program strategies (ADR-147/148). Strategies-as-programs compete under payoff games; tournaments produce Wolfram-style competitive arrays; hill-climb and mutual co-evolution discover winners empirically.

Gives Claude a register_assumption tool and shows the turn's assumptions, decisions and considered-but-not-done items above the prompt

Test wrapper mod for hermitd, an always-on Claude Code agent plugin with chat channels, routines, heartbeats, and a learning loop.

Graph Intelligence Engine for RuFlo: single-entry personalized PageRank, streaming delta updates and witness-signed reasoning artifacts (ADR-123), built on sublinear-time-solver. Library: `npm test`, `npm run build`.

A Claude Code plugin that forks the conversation after any turn that did real work, asks whether the goal was met, where corners were cut and what to approve, then draws a verdict band above the prompt with one-click next-step buttons.

Publishes Claude Code session context from a REMOTE host to localvoxtral on your Mac, over an SSH RemoteForward tunnel. Command hooks running a bundled POSIX-sh shim that POSTs via curl — needs only sh and curl on the remote host, and fails open silently when either is missing.

universal-modder is an MIT-licensed Claude Code plugin that can identify game engines, read real code, build mods, generate assets with fal, test in-game, and produce showcase videos.

A Reddit post describing a Claude Code mod that uses a runtime judge model to automatically select between Haiku, Sonnet and Opus and estimate reasoning effort per prompt, plus warnings for cold prompt caches and overloaded chats, and a smart handoff feature. The mod claims to read recent messages before switching to avoid mid-task model flips. It is described as free and open source (MIT) with a website for installation, but no repository link or verifiable source is included.

A Claude Code plugin that rewrites the model on every agent.spawn to a pinned target (default fable) unless the spawn is a fork; /fable-pin on, off and status toggle it per session, with state kept in the plugin's own store.

Knowledge graph construction — entity extraction, relation mapping, and pathfinder graph traversal. As a mod (ADR-445): keeps secrets out of the graph, confirms graph deletes, /kg-mod, and a status file the console shows

Picks the model for each subagent that names none, with OpenAI's Decisions API or Jev, from its short label only. One line per spawn above the prompt; /route shows each pick and what each subagent cost.

Project Sentinel is an open-source Claude Code mod for long autonomous runs that separately tracks work progress, context usage with automatic handoffs, and prompt-cache health/expiry. It adds autopilot fresh-context handoffs, cache protection, permissions/guardrails, resource limits, activity, profiles, and an optional companion, and can be installed via the plugin marketplace.