Hands the small judgments in a session to TypeSafe's Jev, a System One model that returns typed answers instead of generating text: what kind of request each prompt is, which model tier a subagent needs, whether an edit breaks a rule already written in your CLAUDE.md or AGENTS.md, and which transcript blocks survive compaction. Kept blocks carry over verbatim — no LLM writes a summary.
fleet-band
wrg32786/aigent-os
Read-only fleet status band above the prompt (seat, context %, Room unread, pending job results, needs-you, held seats) plus /fires, /pending and /held commands that answer with no model turn.
fleet-delivery
wrg32786/aigent-os
Delivers a seat's Room inbox and pending job results into the session through the prompt API, one prompt per item, tracked in a per-seat log, and moves each delivered Room file to processed/. A room_send from the seat's model also announces the file to the recipient session through the engine's cross-session channel (a pointer, never the body); the recipient reads it with room_drain, and the...
cache-warmer
paulbkim-dev/claude-code-cache-warmer
Keep the main conversation's prompt cache warm by automatically forking its last request, tools denied and billed like any request, shortly before the cache expires; switch its lifetime between 5m and 1h, and show each refresh's cost and estimated saving.
workflows
edwinhu/workflows
One loop — work — for work worth doing properly: clarify, an approved plan that is the run's only authority, delegated implementation, independent verification against computed gates, and human review. Domain workflows for code (test-first), data, writing and workshop presentations.
sudus
eas4ai/sudus
The referee for agent-led software development: reads the repository, names the next action.
bitfrost
SavaSoftworks/BitFrost
Use Codex, ZCode, ACP apps (opencode, Oh My Pi, Gemini CLI), Crush, Grok Build and Antigravity as Claude Code subagents
Make sense of your Claude Code session's context, prompt cache, and usage limits through quick tips and short lessons. Insights come from what Claude Code reports; everything stays within your machine.
pair
Mint-Security/open-kitchen
Share your Claude Code session with someone, peer to peer with no backend: they watch it live, drawn the way Claude Code draws it, and can talk to your Claude under their own name. It runs over a private Syncthing the plugin downloads on first use; no account needed.
proof9
frankbria/codeframe
Record a bug so it stays fixed: proof9 checks it before every commit.
harness-mods
ucsandman/Agnostic-AI
Harness Mod layer: heartbeat/canary, native usage, rewrite explanations, measured subagent accounting, supervisor routing, secret redaction. Modes per guard in ~/.claude/mods/mods-config.json (classic | shadow_mod | mod).
bender
ucsandman/Agnostic-AI
REALITY BENDER: intercepts a harmless tool result and alters what Claude receives; prints REAL vs RECEIVED
blackbox
ucsandman/Agnostic-AI
BLACK BOX RECORDER: structured per-session timeline of prompts, model steps, tool calls, subagents and denials; /blackbox prints it
guardian
ucsandman/Agnostic-AI
GUARDIAN: executable policy that protects lab/demos/protected/** from Write/Edit/Bash, whatever CLAUDE.md says
matrix
ucsandman/Agnostic-AI
MATRIX: freezes a harmless tool call before it runs, shows what Claude intended, asks you, then continues or denies. /matrix on|off
tunnel-cost
ucsandman/Agnostic-AI
MIDDLEWARE TUNNEL stage 2: Cost analyzer: estimates the call's token cost from the args size
tunnel-observer
ucsandman/Agnostic-AI
MIDDLEWARE TUNNEL stage 1: Observer: sees the call first, stamps t0, prints what enters