用闭环控制框架实现可追溯的自动软件开发，保障安全与确定性。
Closed-Loop Autonomous Software Development via Jira-Integrated Backlog Orchestration: A Case Study in Deterministic Control and Safety-Constrained Automation
- 将软件生命周期管理建模为带七阶段流水线的闭环控制系统。
- 152次运行全部成功，95%置信区间达[97.6%, 100%]，零误报。
- 适合需高可靠性、强审计能力的自动化开发场景。
本文提出一种以控制架构为核心的软件生命周期闭环管理系统，而非单纯代码生成工具。系统管理约1,602行任务数据，覆盖七个任务类别，处理13份结构化文档，执行由七条自动化流水线构成的确定性七阶段流程。自动化栈包含约12,661行Python代码（23个脚本）及6,907行版本化提示规范，配备检查点时间预算、101个异常处理机制和12个集中式锁机制，通过四个核心函数与八个可复用模式实现。Jira状态契约提供外部可见的冲突锁定，降级模式支持Jira不可用时本地持续运行。人工智能辅助受限于结构化上下文包、资源上限、输出重验证与人工审查门控。初始152次运行评估显示100%终端状态成功，95% Clopper-Pearson置信区间为[97.6%, 100%]；系统已积累超过795个运行产物。三轮对抗式代码审查发现51个问题，均在研究范围内关闭（48个完全修复，3个延期加固），注入样本中无误报。在10个自主安全工单中，6个通过流水线自动调度与验证完成，2个需人工修复，2个由策略决策关闭。结果表明，在嵌入明确控制、恢复与审计机制的前提下，受约束的可追溯自动化是可行的。
原文摘要 · Abstract (English)
This paper presents a closed-loop system for software lifecycle management framed as a control architecture rather than a code-generation tool. The system manages a backlog of approximately 1,602 rows across seven task families, ingests 13 structured source documents, and executes a deterministic seven-stage pipeline implemented as seven scheduled automation lanes. The automation stack comprises approximately 12,661 lines of Python across 23 scripts plus 6,907 lines of versioned prompt specifications, with checkpoint-based time budgets, 101 exception handlers, and 12 centralized lock mechanisms implemented through four core functions and eight reusable patterns. A Jira Status Contract provides externally observable collision locking, and a degraded-mode protocol supports continued local operation when Jira is unavailable. Artificial-intelligence assistance is bounded by structured context packages, configured resource caps, output re-validation, and human review gates. A formal evaluation of the initial 152-run window yielded 100% terminal-state success with a 95% Clopper-Pearson interval of [97.6%, 100%]; the system has since accumulated more than 795 run artifacts in continuous operation. Three rounds of adversarial code review identified 51 findings, all closed within the study scope (48 fully remediated, 3 closed with deferred hardening), with zero false negatives within the injected set. In an autonomous security ticket family of 10 items, six were completed through pipeline-autonomous dispatch and verification, two required manual remediation, and two were closed by policy decision. The results indicate that bounded, traceable lifecycle automation is practical when autonomy is embedded within explicit control, recovery, and audit mechanisms.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。