[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-93506":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":16,"stars7d":16,"stars30d":16,"stars90d":16,"forks30d":16,"starsTrendScore":16,"compositeScore":17,"rankGlobal":10,"rankLanguage":10,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":19,"hasPages":19,"topics":21,"createdAt":10,"pushedAt":10,"updatedAt":22,"readmeContent":23,"aiSummary":24,"trendingCount":16,"starSnapshotCount":16,"syncStatus":25,"lastSyncTime":26,"discoverSource":27},93506,"harness-engineering","lopopolo\u002Fharness-engineering","lopopolo","🐎 Ryan Lopopolo’s anthology, field guide, and agent context bundle for harness engineering","https:\u002F\u002Fopenai.com\u002Findex\u002Fharness-engineering\u002F",null,"Python",2298,234,5,1,0,59.11,"Creative Commons Attribution 4.0 International",false,"trunk",[],"2026-08-25 04:01:20","# Harness Engineering\n\n> “Most people do not know that they can just point their agents at my writing,\n> tweets, podcasts, and talks and improve the output of their agents by 100x.”\n>\n> — [Ryan Lopopolo]\n\n[Ryan Lopopolo]: https:\u002F\u002Fx.com\u002F_lopopolo\u002Fstatus\u002F2050698864542482709\n\nHarness engineering, the practice of improving agent output by shaping the\nenvironment around it, holds a chosen model and coding agent constant as a black\nbox. It improves the two external levers—context and tools—and curates the\nenvironment around them. The worker should be able to recover intent, operate\nthe real system, respect authority, prove the outcome, and leave the next run\nbetter equipped.\n\nA central purpose of that environment is to carry an organization's\nnonfunctional requirements: the quality attributes and constraints governing\nreliability, security, compatibility, maintainability, performance, operability,\nrisk posture, and polish. The harness also carries local decisions about how to\nprioritize, trade off, and satisfy those requirements. Ryan adopted a\n[systems-level framing] from 2026’s [\\[un\\]prompted conference] that describes this\nas getting the whole universe of nonfunctional requirements into code. [Make the\nRepository Teach the Agent] develops how the requirements and decisions become\nretrievable context, examples, tools, and executable constraints.\n\n[systems-level framing]: https:\u002F\u002Fx.com\u002F_lopopolo\u002Fstatus\u002F2028982729145237775\n[\\[un\\]prompted conference]: https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=U2O14Jd3MBU\n[Make the Repository Teach the Agent]:\n  docs\u002Fdomain-modeling\u002F#make-nonfunctional-requirements-recoverable\n\nBecause [work is an iterative game], a harness can make organizational judgment\ncumulative. Lessons from accepted work, corrections, failures, and user\nresponses become context, boundaries, tools, examples, and checks that shape\nlater trajectories. Over time, that feedback loop can [make coherence\ncumulative] across agent-maintained artifacts.\n\n[work is an iterative game]: https:\u002F\u002Fx.com\u002F_lopopolo\u002Fstatus\u002F2052858891835465813\n[make coherence cumulative]: docs\u002Fdurable-systems\u002F#make-coherence-cumulative\n\n[Code is how an agent uses a computer]. That internal action language can\nproduce reliable domain outcomes for people who never review the implementation\nwhen [last-mile deployment] supplies the organization’s context, capabilities,\nauthority, and proof.\n\n[Code is how an agent uses a computer]:\n  https:\u002F\u002Fx.com\u002F_lopopolo\u002Fstatus\u002F2043495733375230026\n[last-mile deployment]: docs\u002Flast-mile-deployment\u002F\n\nGeneral model weights contain only the visible tip of an organization’s\nprocess-data iceberg. Below the waterline sit the current operational state,\nlocal ontology, quality bar, procedures, exception history, and authority\nrelationships that an agent needs to do a particular job. Organizations cannot\npresume that this private, changing process data will be present in general\nmodel weights, nor that agents will reliably intuit which process data matters\nto the organization. Harness engineering is the last-mile work of making it\navailable to a capable worker as context and tools.\n\nPoint a coding agent at this repository alongside the system it should improve.\n[`AGENTS.md`] routes the task to the relevant arguments, cases, and proof. For\ndirect reading, start with the [thesis index]. For an application, choose from\nthe [playbooks].\n\n[`AGENTS.md`]: AGENTS.md\n[thesis index]: docs\u002F\n[playbooks]: playbooks\u002F\n\n## Sources and related work\n\n- [“Harness engineering: leveraging Codex in an agent-first world”] ([fetch\n  helper] for agents blocked by the canonical page)\n- [Source library]\n- [Influences and alternate framings]\n\n[“Harness engineering: leveraging Codex in an agent-first world”]:\n  https:\u002F\u002Fopenai.com\u002Findex\u002Fharness-engineering\u002F\n[fetch helper]: sources\u002Fscripts\u002Ffetch_openai.py\n[Source library]: sources\u002F\n[Influences and alternate framings]: docs\u002Flineage\u002F\n\nRepository-authored material is licensed under [CC BY 4.0]. See [`COPYING.md`]\nfor attribution and rights in source material.\n\n[CC BY 4.0]: LICENSE\n[`COPYING.md`]: COPYING.md\n","Harness Engineering 是一套面向 AI 代理（agent）的工程化方法论与实践框架，旨在通过系统性构建代理运行环境（即“harness”）来提升其输出质量与可靠性。核心在于固定模型与编码代理作为黑盒，聚焦优化外部两大杠杆：上下文（含组织非功能性需求、领域知识、历史反馈）和工具（可执行约束、检查机制、领域接口），使代理能准确理解意图、操作真实系统、尊重权限边界并持续累积组织判断。适用于需要将企业私有流程、质量要求、权责规则等隐性知识注入 AI 代理的场景，如企业级自动化开发、合规性敏感的智能运维与领域专用 Agent 部署。",2,"2026-07-20 02:30:03","CREATED_QUERY"]