Same answer. No fluff. No gaps. No overlap.

Make your AI coding agent answer with MECE structure.
Same substance. 59% fewer output tokens on live eval,
up to 53% on input compression. Context intact. Register neutral.

See it · Install · Levels · What you get · vs. Caveman · Honest numbers

---

MECEIA is a skill/plugin for Cursor, Claude Code, Codex, OpenCode, and other agents. Install once. Agent drops filler and answers with MECE structure in a neutral technical register—keeping code, commands, and errors byte-for-byte exact.

Unlike caveman, MECEIA compresses structure and social surface, not grammar. Sentences stay intact; causal connectors stay when order matters.

Before / After

Normal agent	MECEIA
Sure! I'd be happy to help. The reason your React component is re-rendering is likely because you're creating a new object reference on each render cycle. When you pass an inline object as a prop, React's shallow comparison sees it as a different object every time. I'd recommend using useMemo.	Re-render caused by a new object reference each render cycle. Cause: inline object prop creates a new ref every render. Effect: shallow compare fails; component re-renders. Fix: wrap the object in useMemo.
Good catch! The auth middleware is probably not validating token expiry correctly. Let me suggest a fix.	Auth middleware rejects valid tokens at exact expiry boundary. Cause: expiry check uses < instead of <=. Fix: change comparison operator in token validation.

Same fix. Fewer words. Context preserved.

Benchmarks

Live Cursor agent CLI eval — 10 prompts, CURSOR_MODEL=composer-2.5-fast. Output tokens only. Full artifact: evals/results/2026-07-30T20-28-09-888Z.md.

agent login
$env:CURSOR_MODEL = "composer-2.5-fast"   # PowerShell; see evals/README.md
node evals/run.mjs --runner cursor --limit 10

Task	Baseline	MECEIA	Saved
Explain React re-render	837	316	62%
Fix auth token expiry	484	220	55%
Explain PostgreSQL pooling	852	430	50%
Explain git rebase vs merge	577	273	53%
Refactor callback to async/await	586	252	57%
Microservices vs monolith	789	263	67%
Review PR for security	654	116	82%
Docker multi-stage build	800	381	52%
Debug PostgreSQL race condition	1253	502	60%
Implement React error boundary	690	361	48%
Average	752	311	59%

Also measured: 90% MECE structural pass rate; 90% zero machine-register violations; context retention 3.8 avg on keyword rubric (synonym-aware scorer in evals/score-lib.mjs).

Honest scope limits: output tokens only; skill input overhead not subtracted. See docs/HONEST-NUMBERS.md.

vs Caveman

	Caveman	MECEIA
Compresses	Grammar / mouth	Structure + social surface
Voice	Meme caveman	Neutral technical / CLI
Risk	Context & connectors drop	Over-structuring (mitigated by anti-theater rules)
Self-reference	Allowed ("me caveman think")	Banned (passive / "this model")

Install

One command. Finds every agent on your machine. Installs for each.

# macOS · Linux · WSL · Git Bash
curl -fsSL https://raw.githubusercontent.com/uncle-rennus/meceia/main/install.sh | bash

# Windows · PowerShell 5.1+
irm https://raw.githubusercontent.com/uncle-rennus/meceia/main/install.ps1 | iex

~30 seconds. Needs Node ≥18. Skips agents you don't have. Safe to re-run.

Turn it on: type /mece or say "mece mode". Turn it off: say "stop mece" or "normal mode". With --with-init on Cursor, MECEIA is always-on from message one.

Install for one agent, or from a clone

Every agent has its own path (skill file, rule file, or npx skills add). The full per-agent matrix, all flags, dry-run, and uninstall live in INSTALL.md. A few common ones:

# Cursor — always-on rules in current repo
node bin/install.js --only cursor --with-init

# OpenCode — skills + AGENTS.md ruleset
node bin/install.js --only opencode

# Claude Code / Codex — via skills registry
npx skills add uncle-rennus/meceia -a cursor -y
npx skills add uncle-rennus/meceia -a claude -y

Install broke? Open your agent in this repo and say: "Read CLAUDE.md and INSTALL.md, install MECEIA for me." Agent read repo, agent fix own brain.

Levels

Level	Behavior
tight (default)	Verdict + minimal MECE bullets; full grammar
brief	Slightly more complete; still no fluff
full	Exhaustive MECE coverage when the task needs it

Switch: /mece tight|brief|full

What you get

Command	What it does
/mece [tight|brief|full]	MECE + machine register on every reply
/mece-commit	Conventional Commit messages, why over what
/mece-review	Line-anchored PR comments, no praise fluff
/mece-compress <path>	Compress memory files (AGENTS.md, rules) to save input tokens
/mece-compress: + paste	One-shot inline compress; chat output only, no file write

/mece saves output tokens on replies. /mece-compress saves input tokens on files or pasted prose loaded into context.

How it works

1. Install drops skill files into your agent.
2. Skill tells agent: MECE structure, neutral register, no first-person, no closing solicitations.
3. With --with-init, always-on rule files land in .cursor/rules/mece.mdc (and AGENTS.md for Claude).
4. Code, paths, errors: never touched.

Privacy

MECEIA makes no network calls after install. No telemetry, no analytics, no accounts, no backend. The skill is a prompt; the installer writes local files only.

License

MIT — see LICENSE.