Code-role v0.4.0: Less Process, Stronger Outcomes

Code-role v0.4.0: Less Process, Stronger Outcomes

Code-role v0.4.0：更少流程，更强交付

Code-role v0.4.0 refactors both the four-workstation Minimal Profile and eight-role Full Profile around one rule:

A role is complete only when it produces the evidence required to change an accepted outcome KR from 0 to 1.

Code-role v0.4.0 用同一条规则重构了四工位最小版和八角色完整版：

只有当角色产出的证据足以把一个已确认的结果型 KR 从 0 变成 1，这个角色的任务才算完成。

What Changed / 核心变化

• Outcome KRs only. Research, PRDs, architecture, code activity, tests written, reports, and reviews are methods or evidence, not delivery outcomes.

• One failed evidence item per assignment. Roles no longer receive broad agendas or entire KR=0 backlogs.

• Immediate role execution. A complete assignment starts work; startup acknowledgements and readiness-only turns are removed.

• One primary professional artifact. Annexes, manifests, and packet locks are optional audit support rather than daily routing gates.

• Independent acceptance. Engineering produces candidate evidence but cannot approve its own delivery.

• Two evaluation facts. Evaluators separately report evaluation_executed and kr_observed_pass.

• Compact control state. Project Manager state records the current accepted objective, failed evidence, owner, and decision instead of process chronology.

• KR 只描述结果。 调研、PRD、架构、代码活动、写测试、报告和评审都只是方法或证据，不是交付结果。

• 一次只处理一个失败证据。 角色不再接收宽泛议程或整组 KR=0 待办。

• 任务书即启动。 完整任务书到达后直接工作，不再进行启动确认和 readiness-only 回合。

• 一份主专业产物。 附件、manifest 和 packet lock 降为可选审计支持，不再充当日常路由门槛。

• 独立验收。 工程可以生成候选证据，但不能批准自己的交付。

• 两个评估事实。 评估师分别报告 evaluation_executed 与 kr_observed_pass。

• 紧凑控制状态。 项目经理只记录当前目标、失败证据、责任人和决策，不记录流程流水账。

Why It Matters / 为什么重要

The previous workflow could be strict and still fail in practice: roles spent too much time satisfying workflow mechanics, while the actual software outcome remained unchanged. v0.4.0 preserves the hard acceptance boundary and removes work that does not change milestone evidence.

之前的流程即使严格，也可能在实践中失败：角色花费大量时间满足流程动作，真实软件结果却没有变化。v0.4.0 保留严格验收边界，同时删除不能改变里程碑证据的工作。

The design was calibrated against two real private projects:

• DeepBrain: strong partial benchmark and runtime evidence remained insufficient for a production claim.
• Leaper Agent: a professional evaluation plan was rejected because its task artifacts, holdout isolation, grader mechanism, and runtime conditions were not executable.

这套设计来自两个真实私有项目的校准：

• DeepBrain： 大量局部 benchmark 与 runtime 证据仍不足以支持生产结论。
• Leaper Agent： 一份专业评估方案因为任务数据、holdout 隔离、grader 机制和运行条件不可执行而被打回。

Start in 60 Seconds / 60 秒启动

git clone https://github.com/Deepleaper/Code-role.git
cd Code-role
python3 scripts/init_loop_workflow.py "/absolute/path/to/your-project" \
  --project-name "Your Project"
python3 scripts/init_loop_workflow.py "/absolute/path/to/your-project" --check

Then define one Objective, binary outcome KRs, and the exact failed evidence item the next role owns.

然后定义一个 Objective、可二值验收的结果型 KR，以及下一个角色负责解决的唯一失败证据。

Verified / 已验证

• Repository test suite: 83 passed.
• Minimal Profile initializer replay and --check: passed.
• Full Profile initializer replay: passed.
• Manifest JSON validation and HTML parsing: passed.

Read more:

• README
• Complete four-workstation walkthrough
• DeepBrain case study
• Leaper Agent case study
• Two-case launch story

Code-role v0.3.1: Real project case studies

Two projects, two failure modes, one milestone-control loop

This release adds sanitized bilingual case studies from two private AI engineering projects:

• DeepBrain: strong local evidence included 1,750 passing unit tests, S50 at 50/50, LongMemEval-S at 499/500, and 100/100 grounded source joins. Independent Evaluation still held milestone completion and Reviewer routing at 0 because fair comparison, raw benchmark reruns, repair proof, clean reproducibility, and production cost/SLO evidence were incomplete.
• Leaper Agent: the Project Manager rejected a professional-looking evaluation baseline before Engineering started because task artifacts, holdout isolation, grader execution, runtime conditions, and integrity evidence were not yet executable. The correction replaced declarations with concrete task, holdout, calibration, command, and integrity artifacts.

Neither case is presented as a completed product milestone. The case studies show the core Code-role value: preserve proven progress while preventing unsupported claims and invalid handoffs.

Included

• README case-study proof points.
• Bilingual DeepBrain and Leaper Agent case studies.
• A ready-to-publish two-case technical launch story.
• Updated launch copy and public disclosure boundaries.

Verification

• 69 repository tests passed.
• All relative Markdown links resolve.
• Private source code, repository links, local paths, customer data, and implementation details are excluded.

Code-role v0.3.0: Minimal and Full workflow profiles

Reliable AI coding around one accepted milestone

Code-role v0.3.0 introduces two current operating profiles:

• Minimal Profile: a four-workstation goal loop for normal software delivery.
• Full Profile: an eight-role, versioned packet workflow for complex or audit-heavy milestones.

What is new

• PM-owned Objectives and binary Key Results.
• One current KR=0 per assignment.
• Independent evaluation before milestone acceptance.
• Minimal Profile initialization, synchronization, and validation.
• An evidence-complete walkthrough from PM Assignment to independent KR=1.
• Bilingual Roadmap, Issue forms, social preview, and launch materials.

Quick start

python3 scripts/init_loop_workflow.py /absolute/path/to/project --project-name "Project Name"
python3 scripts/init_loop_workflow.py /absolute/path/to/project --check

Verification

• 68 repository tests passed.
• Social preview verified at 1280 x 640 and under 1 MB.

Code-role remains local-first and discussion-first. It controls delivery evidence; it does not automatically execute an entire software project.

v0.1.1

What changed

• Added milestone contract as the hard goal anchor for Orchestrator / Project Manager.
• Added evaluation SOP as the hard evaluation anchor for Test Evaluator and Reviewer.
• Added Test Evaluator packet templates for evaluation SOP consumption and SOP calibration.
• Updated Reviewer standards to audit milestone-contract and evaluation-sop adherence across the workflow.
• Updated target-project bootstrap to generate milestone-contract.md and evaluation-sop.md.
• Updated README, PRD, HTML workflow guide, role standards, source map, and tests.

Validation

• .venv/bin/python -m pytest -q: 60 passed
• git diff --check: pass

Notes

This release keeps Code-role as local workflow assistance for target projects. Target-project code-role/ folders remain outside product release artifacts unless a project explicitly adopts them.

Code-role v0.1.0

Code-role v0.1.0

First public release of Code-role, a discussion-first local role workflow template for Codex-assisted software delivery.

What is included

• Eight role definitions: Workflow Orchestrator, Researcher, Product / PRD, Architect, Code Context, Implementer, Test Evaluator, and Reviewer.
• Lightweight milestone handoff as the default workflow.
• Optional strict packet locking for advanced handoff control.
• Target-project bootstrap script.
• Role output standards and packet templates.
• Public-source network research policy with explicit boundaries for private/authenticated resources.
• MIT License, contribution guide, security policy, CI, README workflow diagram, and minimal target-project example.

Boundary

Code-role is not an automation runner. It is designed to make each professional step discussable, reviewable, and explicitly accepted before the next role acts.

Generated target-project code-role/ folders are local operator assistance by default and should not be committed or shipped unless a project explicitly adopts them.