Will this pack work on stock MCode without local models?

Mostly no. It was built and tested for a local GLM model through vLLM with zero token cost, while cloud TypeSafe stays opt-in and model serving remains separately deployable.

Do I need to reapply patches after updating MCode?

No manual reapply. Launching through mcode-enhanced checks the active release and automatically reapplies or rebuilds the pack after an update while preserving your config.

Can I use snap Chromium for the browser lane?

No. Snap confinement blocks CDP binding, so the verified lane is Playwright arm64 Chromium headless with --no-sandbox attached via BU_CDP_URL, confirmed by smoke-test.

Where do the ledger and performance numbers come from?

They come from the runtime's own sqlite ledger plus timed probes with a reproducible collector, including the long session that failed before the fix and prefill measurements.

How do I install the pack?

Run MCODE_SOURCE_TREE=/path/to/minimax-code ./install-one-shot.sh; it applies the compatible patches, installs the launch-time compaction enforcer, sets the 64K local-vLLM ceiling, and adds ~/.local/bin/mcode-enhanced, which auto-reapplies the pack after an MCode update.

What does the compaction fix actually do?

It addresses INVALID_CHECKPOINT auto-compaction failures on local vLLM by enforcing a never-reset limit: a 64K trigger at about 48K with a worst-case pair of about 104K against the 257K KV cache, checked on every launch, with a keeper.sh watchdog alongside it.

What is the pre-built container?

An arm64 GHCR image, ghcr.io/drowzeys/keys-mcode-continuous-context-browser-decision-enhancement-pack, that overlays the official @minimax-ai/code 0.4.12 with the patched bundles, arm64 Playwright Chromium, and the pack tooling.

CommunityJev Demos & Experiments

Keys Mcode Continuous Context Browser Decision Enhancement Pack

Enhancement pack for the MiniMax Code CLI with TUI patches, continuous context renewal, a local-vLLM compaction fix, and a measured arm64 browser lane.

See highlights

Open github.com

Stars: 8
Language: Python
Published by: Community
Author: drowzeys
Last verified: Oct 2026

drowzeys/keys-mcode-continuous-context-browser-decision-enhancement-pack

User repository on GitHub

View on GitHub

MiniMax Code CLI enhancement pack: continuous 25% context renewal, context progress meter, browser tooling, Syntra and djev-spark local decision integrations.

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A pack of validated enhancements for the MiniMax Code (MCode) CLI, built and battle-tested on a 4× NVIDIA DGX Spark (GB10, arm64) running a local GLM-5.3-EXL3 3.0 bpw (abliterated) model through vLLM — no cloud API, $0 token cost.

Quoted from the project's README. Read the full README.

Tagscommunity awesomejev catalog-import

Works forJev for Classification Jev for Tool calling

Used inJev for Software & DevTools

Languages: Python

Highlights

Adds transcript scrollbar, opt-in context gauge, and renewal after 25% use with tests.
Repairs INVALID_CHECKPOINT failures with a never-reset limit enforced on launch.
Provides a verified CDP lane using Playwright arm64 Chromium and BU_CDP_URL.
Applies via one-shot installer that preserves config and rebuilds after updates.

Watch out

Context limit is cached per session, so restart MCode to activate changes; snap Chromium cannot drive CDP and vision tools need a live VLM. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-kit

Claude Code kit that guards PreToolUse calls with a free pre-filter plus one Jev question for ambiguous cases, and adds sub-agent sizing, rechecks, search, daemon speedups, and monitoring.

Jev Demos & Experiments#community#awesomejev#catalog-import

Community

jonathanavis96/jev-kitPython

jev-ultrafast-mcp

MCP server where one browser_goal call hands a URL, a goal, and code-checked assertions to a server-side loop driven by a decision model over live page elements.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

jiawei686/jev-ultrafast-mcpPython

aside-jev

Browser workflow that keeps one Aside REPL session open, compacts page observations into allowed candidates, and has Jev pick the next action with confidence checks.

Jev Demos & Experiments#community#awesomejev#catalog-import

Community

himomohi/aside-jevPython

jev-harness

Decision layer for coding agents that triages test failures deterministically before spending frontier tokens. Offline heuristics handle missing packages, flaky transients, and loop breaking, with optional Jev micro-decisions and CLI, MCP, hook, and SDK use.

Jev Demos & Experiments#community#awesomejev#catalog-import

Community

ismaelsoilet/jev-harnessPython

open-spark-jev

Independent open-source local decision model inspired by Jev, with a spark-s1 family plus smaller control models for agent gates, served locally with probabilities and policy actions.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

abhishek085/open-spark-jevPython

agi-jev-containment

Local agent observability stack that scores chains of tool events with Jev, Sentinel rules, and SafetyDrift, then sets escalate-only incident levels with graph forensics.

Jev Demos & Experiments#community#awesomejev#catalog-import

Community

carlosedm10/agi-jev-containmentPython

From the Jev guides

Original write-ups on what this entry is for.

Per-Tick Decisions in Games

Sep 28, 2026

A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.

Jev Demos & Experiments#gaming#agents#control

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai