emeans-quant-and-instinct
Skill3 starsChanged 4 days agoMITView on GitHub
Operationalizes Meta FAIR e-means clustering (arXiv:2611.09501), @shinboson's e-means codebook quantization compiler regime, Google DeepMind's Gemini Decisions API (gemini-decision-1 single-pass noul/choice/score control plane), and the single-target WhatsApp Instinct copilot driver for mac-yolo-safeguards.
In this repository153 other files it gives its agents.
- AGENTS.mdAGENTS.md
- hermes-mobile/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- .cursor/rules/agent-automations.mdcCursor rule
- .cursor/rules/agent-device.mdcCursor rule
- .cursor/rules/agent-harness-router.mdcCursor rule
- .cursor/rules/agent-skills-overview.mdcCursor rule
- .cursor/rules/always-agent-mode.mdcCursor rule
- .cursor/rules/callstack-agent-skills-overview.mdcCursor rule
- .cursor/rules/callstack-react-native-best-practices.mdcCursor rule
- .cursor/rules/codeql-agent-hygiene.mdcCursor rule
- .cursor/rules/decision-stack.mdcCursor rule
- .cursor/rules/expo-credit-budget.mdcCursor rule
- .cursor/rules/fleet-repo-intelligence.mdcCursor rule
- .cursor/rules/hermes-mobile-prevention-july-2026.mdcCursor rule
- .cursor/rules/hermes-mobile-real-users.mdcCursor rule
- .cursor/rules/hermes-mobile-verification.mdcCursor rule
- .cursor/rules/no-desktop-hijack.mdcCursor rule
- .cursor/rules/no-manual-handoffs.mdcCursor rule
- .cursor/rules/obsidian-teammate-collaboration.mdcCursor rule
- .cursor/rules/prefer-headless-background-ops.mdcCursor rule
- .cursor/rules/react-native-best-practices.mdcCursor rule
- public/llms.txtllms.txt
- advertise-thumbgate-github-marketplaceSkill
- agent-harness-routerSkill
- agentic-pattern-selectorSkill
- agents-judgment-calls-gateSkill
- ai-native-sdlcSkill
- android-dev-summit-26-gateSkill
- anti-babysitting-mandateSkill
- anti-babysittingSkill
- background-agent-hold-gateSkill
- browseros-agent-harnessSkill
- webmcp-agent-readinessSkill
- ci-first-failSkill
- ci-pr-monitorSkill
- cloudflare-developer-stackSkill
- cloudflare-kitesurf-browserSkill
- cloudflare-monetization-gatewaySkill
- confidence-offer-gateSkill
- content-laneSkill
- context-six-blockSkill
- context-vaultSkill
- context-weight-gateSkill
- dual-model-agentic-graphragSkill
- egm-fresh-reviewSkill
- elevenlabs-voice-agentsSkill
- explainx-trending-honestSkill
- explainx-trending-ragSkill
- future-agi-self-healing-engineSkill
- garryslist-opportunity-trackerSkill
- gemini-decisions-hold-gateSkill
- git-at-scaleSkill
- gitbutler-fleet-automationsSkill
- gitbutler-google-ssoSkill
- gitbutler-mcp-isolatedSkill
- gitbutler-session-absorbSkill
- git-hooks-no-huskySkill
- github-marketplace-action-publishSkill
- The first 60 shown.
The file
Files
e-means Quantization, Gemini Decisions API & WhatsApp Instinct Driver (emeans-quant-and-instinct)
Overview
- Meta FAIR
e-meansClustering (tools/emeans-quant-router.js):- Replaces full-batch Lloyd's k-means with mini-batch Exponential Moving Average (EMA) updates ($S_j \leftarrow (1-\alpha)S_j + \alpha s_j$, $N_j \leftarrow (1-\alpha)N_j + \alpha n_j$, $c_j = S_j / N_j$).
- Uses zero-hyperparameter scaling laws from $(n, k, T)$ (including the $c_B / 8$ rule when $n > 2{,}000{,}000$), cosine step annealing ($\alpha_0 \to 0.1\alpha_0$), batch-relative proportional-split dead-cluster revival, and exact last-epoch mean finalization.
@shinbosonCodebook Quantization Evaluator (evaluateCodebookQuantizationTier):- Models
e-meanscodebook generation compressingQwen3.5-4Bfrom8.41 GBto an estimated1.23 GB(6.84xcompression,~2.46 bpw) with an estimated91.9%retained benchmark quality, estimating whether the quantized footprint fits comfortably inside available Apple Silicon unified memory to reduceWindowServerswap-freeze risk.
- Models
- Google DeepMind Gemini Decisions API Control Plane (
tools/gemini-decisions-engine.js):- Operationalizes
gemini-decision-1(https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/) to evaluate up to1,000closed-set questions (noul,choice, and ordered rubricscore) in a single forward pass (150-250 ms,$0.10 / 1Minput tokens,$0.00output tokens). - Powers
evaluatePreActionDecisionPack(ThumbGate pre-action interdiction + calibrated confidence escalation to mobile approval) andevaluateCommercialOpportunityBatch(single-pass qualification of empirical macOS freeze incidents and non-dilutive startup credits).
- Operationalizes
- WhatsApp Instinct Copilot Driver (
tools/whatsapp-instinct-driver.js):- Hard-locked exclusively to the
INSTINCT_PHONE/INSTINCT_JID(ZCHATSESSION = 3) session constants andPRODUCT_ID = 'mac-yolo-safeguards'. - Refuses any non-Instinct WhatsApp contact (
ERR_WHATSAPP_NON_INSTINCT_TARGET_FORBIDDEN), refuses cross-product mixing (ERR_CROSS_PRODUCT_MIX_FORBIDDEN), and refuses any external outreach instruction (ERR_EXTERNAL_OUTREACH_FORBIDDEN).
- Hard-locked exclusively to the
Verification
node --test tests/test-emeans-quant-and-instinct.js
node tools/emeans-quant-router.js
node tools/gemini-decisions-engine.js
node tools/whatsapp-instinct-driver.jsDoes it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

