[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-70483":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":15,"stars7d":16,"stars30d":17,"stars90d":14,"forks30d":14,"starsTrendScore":16,"compositeScore":18,"rankGlobal":9,"rankLanguage":9,"license":19,"archived":20,"fork":20,"defaultBranch":21,"hasWiki":20,"hasPages":20,"topics":22,"createdAt":9,"pushedAt":9,"updatedAt":23,"readmeContent":24,"aiSummary":25,"trendingCount":14,"starSnapshotCount":14,"syncStatus":26,"lastSyncTime":27,"discoverSource":28},70483,"autoharness","kayba-ai\u002Fautoharness","kayba-ai","🔁 Improve your agent harness autonomously overnight",null,"Python",290,12,54,0,1,3,122,56.84,"MIT License",false,"main",[],"2026-06-12 04:00:55","\u003Cp align=\"center\">\n  \u003Cimg src=\"docs\u002Fimages\u002Fharness_on_harness2.png\" alt=\"autoharness banner\" width=\"900\">\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fdiscord.com\u002Finvite\u002FmqCqH7sTyK\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fdiscord\u002F1429935408145236131?label=Discord&logo=discord&logoColor=white&color=5865F2\" alt=\"Discord\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fx.com\u002Fkaybaai\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Ffollow\u002Fkaybaai?style=social\" alt=\"Twitter Follow\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fkayba.ai\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fkayba.ai-0B1220?style=flat&labelColor=0B1220&color=0B1220\" alt=\"kayba.ai\">\n  \u003C\u002Fa>\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-MIT-16a34a\" alt=\"MIT license\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-3.11%2B-3776AB?logo=python&logoColor=white\" alt=\"Python 3.11+\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FCodex-supported-2563EB\" alt=\"Codex supported\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FClaude%20Code-supported-2563EB\" alt=\"Claude Code supported\">\n\u003C\u002Fp>\n\n# autoharness\n\nLet autoharness run overnight and come back to an optimized agent harness, so your production agents never make mistakes again.\n\nautoharness improves agent harnesses by proposing or applying prompt, config, middleware, and source changes, running evals, and keeping or discarding candidates based on benchmark results.\n\nIt is a control plane for an existing harness repo. You point it at a target root and a benchmark command; autoharness manages proposals, iterations, campaigns, and champion state under `.autoharness\u002F`.\n\n## Install\n\nFastest setup with Codex or Claude:\n\n1. `pipx install \"git+https:\u002F\u002Fgithub.com\u002Fkayba-ai\u002Fautoharness.git\"`\n2. `cd` into your harness repo\n3. open Codex or Claude Code in that repo\n4. tell the assistant: `Run autoharness guide --assistant codex --print-next-prompt, then use the generated onboarding packet to finish setup.`\n\nFor Claude Code, swap `--assistant codex` for `--assistant claude`.\n\nElse:\n\n```bash\npipx install \"git+https:\u002F\u002Fgithub.com\u002Fkayba-ai\u002Fautoharness.git\"\nautoharness --help\n```\n\nIf you do not use `pipx`:\n\n```bash\npython3 -m pip install --user \"git+https:\u002F\u002Fgithub.com\u002Fkayba-ai\u002Fautoharness.git\"\n```\n\n## How It Works\n\n- `guide` inspects a repo, asks a few focused setup questions in a TTY, stays scriptable with flags in non-interactive use, writes a starter `autoharness.yaml` plus benchmark config, and runs a readiness check.\n- `doctor` reruns config, generator, and benchmark validation when you want an explicit readiness gate.\n- `setup` and `init` remain available when you want to manage bootstrap explicitly.\n- `run-benchmark` executes one benchmark directly.\n- `generate-proposal` previews one candidate change without running it.\n- `run-iteration` or `optimize` executes one candidate or a resumable search loop.\n- `promote` or `promote-from-compare` moves a winner into champion state.\n\n## Mental Model\n\n- `target root`: the harness repo or deployment tree to edit\n- `benchmark config`: the command or adapter config that scores candidates\n- `workspace`: the long-lived optimization effort\n- `track`: one comparable lane inside a workspace\n- `campaign`: a resumable search run over candidate proposals\n- `.autoharness\u002F`: persisted settings, proposals, records, iterations, and champions\n\n## Batteries Included\n\n- Adapters: `generic_command`, `pytest`, `harbor`, `tau2_bench`, `hal`, `car_bench`\n- Proposal generators: `manual`, `failure_summary`, `local_template`, `local_command`, `openai_responses`, `codex_cli`, `claude_code`\n- Extension model: Python plugins can add generators, preflight checks, and search strategies from `.autoharness\u002Fplugins\u002F` or `AUTOHARNESS_PLUGIN_PATHS`\n\n\n## Quick Start\n\nLet autoharness generate a starter project config:\n\n```bash\nautoharness guide\n```\n\nIn a TTY, `guide` asks a few setup questions. In scripts or CI, use flags like `--non-interactive`, `--benchmark-command`, `--generator`, and `--autonomy`.\n\nIf you want Codex or Claude to help you refine the setup, generate an assistant brief too:\n\n```bash\nautoharness guide --assistant codex --print-next-prompt\n# or\nautoharness guide --assistant claude --print-next-prompt\n```\n\nThis writes `autoharness.codex.md` or `autoharness.claude.md` plus a structured `autoharness.onboarding.json` handoff next to `autoharness.yaml`, then prints a ready-to-paste assistant prompt. Assistant wrapper prompts live under [`contrib\u002Fagents\u002F`](contrib\u002Fagents\u002FREADME.md).\n\n`guide` ends with a doctor pass. Run `autoharness doctor` again later if you want an explicit re-check or a repeated benchmark probe.\n\nOn a fresh install, `guide` prefers a local assistant backend when `codex` or `claude` is installed, otherwise uses `openai_responses` when OpenAI credentials are configured, and falls back to `failure_summary` only when no model-backed generator is available.\n\nThen run the benchmark directly:\n\n```bash\nautoharness run-benchmark\n```\n\nIf `autoharness.yaml` is present, autoharness will auto-bootstrap missing settings and workspace state on this common path. `setup` and `init` are still available when you want explicit control.\n\nGenerate a proposal against a target harness root:\n\n```bash\nautoharness generate-proposal\n```\n\nIf you switch the project config to `openai_responses`, export an API key first:\n\n```bash\nexport OPENAI_API_KEY=...\n```\n\nRun the outer loop:\n\n```bash\nautoharness optimize\nautoharness report\n```\n\n## Early Results\n\nExample from one `tau2` airline benchmark study. Relative deltas are measured against the baseline harness on the same workload. Results depend on the benchmark, harness, and evaluation setup, and some intervention combinations can regress.\n\n![tau2 intervention deltas](docs\u002Fimages\u002Ftau2_airline_intervention_deltas.svg)\n\n## Docs\n\n- [Quickstart](docs\u002Fquickstart.md)\n- [Usage](docs\u002Fusage.md)\n\n## For Power Users\n\n- Background campaign workers plus queue and worker-state inspection\n- Root-level memory, transfer suggestions, and portfolio scheduling\n- Retention policies, pruning, and portable report and bundle exports\n- Event logs, inspection commands, and operational reporting surfaces\n- Python plugin hooks for generators, preflight checks, and search strategies\n\n> [!TIP]\n> Want deeper analysis or a custom optimization workflow? [Kayba](https:\u002F\u002Fkayba.ai) offers managed harness optimization and agent-improvement support tailored to your stack.\n\n\u003Cp align=\"center\">\n  \u003Cstrong>Star this repo if you find it useful!\u003C\u002Fstrong>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cstrong>Built with ❤️ by \u003Ca href=\"https:\u002F\u002Fkayba.ai\">Kayba\u003C\u002Fa> and the open-source community.\u003C\u002Fstrong>\n\u003C\u002Fp>\n","autoharness 是一个用于优化代理框架的自动化工具，通过提出或应用提示、配置、中间件和源代码变更来改进代理性能。它支持Python 3.11及以上版本，并且兼容Codex和Claude Code等AI助手。核心功能包括运行评估、基于基准测试结果保留或丢弃候选方案，以及管理提案、迭代和竞选活动。适用于需要持续优化生产环境中代理表现的场景，如提高软件代理的稳定性和准确性。使用该工具可以简化开发流程，减少人工干预，让开发者能够专注于更高层次的设计决策。",2,"2026-06-11 03:32:30","CREATED_QUERY"]