
确定性项目智能、持久的工作流编排以及可重复使用的 MCP 工程工具,适用于 Codex。
概览
2718lab DevKit 是一个基于 Codex 的工程工具包:一个本地、仅限 stdio 的 MCP 运行时,用于有界项目索引、Atlas 证据、中继生命周期协调和确定性快速通道规划,以及一个紧凑的 Skill 参考手册包。该仓库包含版本化的 v1.2.1 包。已签入的清单和允许列表定义了可执行的运行时表面;下面的手动映射、安装、构建和验证部分描述了支持的工作流。 公共的 Python 编译器和 CLI 保留了故意关闭的 team-efficiency/fast-lane-plan-v1 预览。对于新工作,准备 fast-lane-request-v2 并使用 prepare_model_neutral_fast_lane_request;MCP fastlane_compile 通过本地 RuntimeRoot 解析工作区/快照选择器,并返回经过验证的团队效率/快速通道计划 v3。其赋值描述需求,模型选择则交由协调员使用当前的 Codex 工具元数据完成。
概述(AI 翻译)
简体中文
2718lab DevKit — Codex + MCP v1.2.1
2718lab DevKit 是一个基于 Codex 的工程工具包:一个本地、仅限 stdio 的 MCP 有界项目索引的运行时、图谱证据、继电生命周期 协调、确定性快速车道规划,以及紧凑的技能组合包 参考手册。该仓库包含经过版本化的 v1.2.1 软件包。 签入的清单和允许列表定义可执行运行时表面; 以下手册地图、安装、构建和验证部分描述了 支持工作流程。
公共的 Python 编译器和 CLI 保留了故意失败关闭的
team-efficiency/fast-lane-plan-v1 预览。新作准备
fast-lane-request-v2 与 prepare_model_neutral_fast_lane_request;MCP
fastlane_compile 然后通过其解析工作区/快照选择器
本地 RuntimeRoot 并返回已验证的 team-efficiency/fast-lane-plan-v3。
其任务描述需求,模型选择则交由
协调员使用当前的Codex工具元数据。所有描述符依然保留
plan_only:它从未发货、不主张租约、创建工作树,或
授权执行。请求-v1/plan-v2仍可保持精确遗留重放。
本页由 AI 翻译,技术细节请以原文为准。
README
以下为原文 README 节选,完整内容请查看原文。
2718lab DevKit — Codex + MCP v1.2.1
2718lab DevKit is a Codex-first engineering toolkit: a local, stdio-only MCP runtime for bounded project indexing, Atlas evidence, Relay lifecycle coordination, and deterministic Fast Lane planning, plus a compact Skill bundle of reference manuals. This repository carries the versioned v1.2.1 package. The checked-in manifest and allowlist define the executable runtime surface; the manual map, install, build, and verification sections below describe the supported workflow.
The public Python compiler and CLI retain the deliberately fail-closed
team-efficiency/fast-lane-plan-v1 preview. For new work, prepare
fast-lane-request-v2 with prepare_model_neutral_fast_lane_request; MCP
fastlane_compile then resolves the workspace/snapshot selectors through its
local RuntimeRoot and returns a verified team-efficiency/fast-lane-plan-v3.
Its assignments describe requirements and leave model selection to the
coordinator using current Codex tool metadata. Every descriptor remains
plan_only: it never dispatches, claims a lease, creates a worktree, or
authorizes execution. Request-v1/plan-v2 stays available for exact legacy replay.
The prior v1.1.3 storage-governance integration references compatible Host source only
on Ayleovelle’s user-fork
codex/host-1.1.3-storage-governance-upstream
branch, pinned to immutable commit
c3dde23bec21c45d10740f2eec09d9a1b87cd329.
That fork is buildable Host source, not an OpenAI upstream merge or a component
shipped by this package. Stock Codex Hosts have no attested protected broker
and continue to fail closed.
[!IMPORTANT] Workflow reminder: route from bounded evidence. Parallel A1/A2/A3 work is allowed only with disjoint, exclusively owned write scopes and independent G: task roots. Claim and bind before execution; prewarm is read-only and
action="retain"is not a new spawn. The coordinator integrates only after PR-style independent review, then accepts before archiving. Fast Lane does not use an account-usage coordinator or account-usage inputs. Do not commit runtime state or credentials.
Scheduler topology V1:
2718lab-devkit/scheduler-topology-v1binds the plan, lease, and G-drive worktree as auditable opaque identities. A/B/C means main-conversation review/integration, scheduler, and writer; each scheduler has at most a 1:3 writer relationship. Design/prewarm are read-only and stay subject to actual host slots, host capability, and safety gates. Cross-scope work requires a declared-child split that strictly reduces conflict, or is UNSPLITTABLE.
1.2.1 hotfix
安装
This server does not publish a one-line install command.
Open the repository installation guide 新手?先看 mcpskill.org 教程 →