Multi-LLM Task Router Suite. runagent.py routes bounded tasks to six backends: Claude CLI, GitHub Copilot, Gemini CLI, Antigravity (Agy), Codex/OpenAI-compatible, and direct local Gemma 4 12B via llama-server (2s, no proxy). 11 expert agent personas covering adversarial patterns (red-team, debate synthesis, output va.
★not rated 6
changed 2d agoA
tokens not measured
originalMIT
Cost-first delegation control for Claude Code: L0-L4 task budgets, role-to-model routing, a Fable critical-worker gate, bounded worker agents, and a PreToolUse hook that denies model inheritance on Agent, Workflow, and SendMessage dispatches.
Architecture governance for AI coding agents — derive the real architecture from any repo, enforce declared rules with a net-new ratchet, and stop agents drifting off pattern before code ships.
★not rated 5 1mo agoA
tokens not measured
originalMIT
Delegate coding tasks to Google's Antigravity (agy) CLI / Antigravity IDE from Claude Code (Google Agy) — full rescue, resume, background execution, code review, and stop-time review gate.
★not rated 5 20d agoA
tokens not measured
originalMIT
Agent-to-agent messaging for AI coding agents — sessions, direct/broadcast/channel messages, shared state, real-time feed, and a file-coordination hook for parallel edits.
★not rated 5 4mo agoA
tokens not measured
originalMIT
Deterministic, resumable multi-agent workflows that orchestrate Codex, Claude, Cursor, and Kimi agents — five MCP tools plus the workflow-authoring skill.
★not rated 5
changed 3d agoA
tokens not measured
originalMIT
Contract-first skill package for Claude Code. Public goal-contract-writer produces canonical Goal Contracts with paired fresh-subagent verifier and progress tracker runtime skills.
★not rated 5 3mo agoA
tokens not measured
originalMIT
Turn Claude Code into a structured development partner — session lifecycle, named code reviewers, research pipelines, and 35+ workflow skills that enforce quality from plan through execution.
Your AI agent maintains its own context files — a Claude Code plugin with an AGENTS-first model that covers Codex, Copilot, Cursor, Gemini, and 3 more tools. Signal Gate filtering, Context Guard hooks, health scoring, and MEMORY.md promotion.
★not rated 5 2mo agoA
tokens not measured
originalMIT
Quality-first harness for autonomous AI coding agents: deterministic gates + a default-FAIL evaluator in the close path of the loop, over a Beads task spine. Done is earned, not asserted. Unifies GSD + Beads; stack-agnostic.
★not rated 5 3mo agoA
tokens not measured
originalMIT
Nothing ships unproven. Heimdall is an autonomous multi-agent orchestrator for Claude Code that turns one prompt into a finished project — and keeps moving the bar on what counts as done: every plan wires an external, falsifiable oracle so the implementation can never grade its own homework, and the merge stays…
★not rated 5
changed 4d agoA
tokens not measured
originalMIT
Pollinations AI for Claude Code CLI. Generate images, text, audio, video, and transcribe speech. BYOP OAuth login — users fund their own usage with pollen credits.
★not rated 5 5mo agoA
tokens not measured
originalMIT
Make any repo AI-native. Analyzes your codebase and generates a complete Claude Code configuration — agents, skills, rules, and hooks tailored to your stack.
★not rated 5 6d agoA
tokens not measured
originalMIT
A disciplined control loop for multi-window Claude Code work: a controller window dispatches scoped task packages across a tmux-resident fleet, collects immutable target results, obtains independent Test validation, reviews the evidence, and decides the next step. Explicit on-disk state roots keep the run auditable…
★not rated 5 yesterdayA
tokens not measured
originalMIT
Turns your agent into a trainable employee. Learns rules from your corrections (with your approval), and skills earn autonomy through demonstrated reliability — proposed → verified → trusted → autonomous. No daemon. No cloud.
★not rated 5 2mo agoA
tokens not measured
originalMIT
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: