[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-74310":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":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":17,"stars30d":18,"stars90d":15,"forks30d":15,"starsTrendScore":19,"compositeScore":20,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":23,"hasPages":21,"topics":24,"createdAt":9,"pushedAt":9,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":15,"starSnapshotCount":15,"syncStatus":28,"lastSyncTime":29,"discoverSource":30},74310,"meta-harness-tbench2-artifact","stanford-iris-lab\u002Fmeta-harness-tbench2-artifact","stanford-iris-lab","Meta-Harness: 76.4% on Terminal-Bench 2.0 (Claude Opus 4.6)",null,"Python",1087,160,12,1,0,6,16,63,18,77.92,false,"main",true,[],"2026-06-12 04:01:14","# Meta-Harness\n\nAgent scaffold for [Terminal-Bench 2.0](https:\u002F\u002Ftbench.ai), built on top of [Terminus-KIRA](https:\u002F\u002Fgithub.com\u002Fkrafton-ai\u002FKIRA) by KRAFTON AI and [Harbor](https:\u002F\u002Fgithub.com\u002Flaude-institute\u002Fharbor)'s Terminus-2 framework.\n\n## Results\n\n76.4% on Terminal-Bench 2.0 (89 tasks × 5 trials, Claude Opus 4.6).\n\n| Split  | N  | Score |\n|--------|---:|------:|\n| Easy   |  4 | 100.0 |\n| Medium | 55 |  81.1 |\n| Hard   | 30 |  64.7 |\n| **All**| 89 |**76.4**|\n\n## Usage\n\n```bash\npip install harbor\n\nexport ANTHROPIC_API_KEY=\u003Cyour-key>\n\nharbor run \\\n  --agent-import-path agent:AgentHarness \\\n  -d terminal-bench@2.0 \\\n  -m anthropic\u002Fclaude-opus-4-6 \\\n  -e runloop \\\n  -n 20 \\\n  --n-attempts 5\n```\n\n## Method\n\nMeta-Harness extends the [Terminus-KIRA](https:\u002F\u002Fgithub.com\u002Fkrafton-ai\u002FKIRA) agent with environment bootstrapping: before the agent loop starts, it gathers a snapshot of the sandbox environment (working directory, file listing, available languages\u002Ftools, package managers, memory) and injects it into the initial prompt. This saves 2-5 early exploration turns that the agent normally spends on `ls`, `which python3`, etc.\n\nThe agent was discovered through automated harness evolution. More details coming soon.\n\n## Acknowledgements\n\nWe thank KRAFTON AI for compute support.\n","Meta-Harness 是一个针对 Terminal-Bench 2.0 的代理框架，基于 KRAFTON AI 的 Terminus-KIRA 和 Harbor 的 Terminus-2 框架构建。该项目实现了76.4%的得分（89个任务×5次试验，使用Claude Opus 4.6模型）。其核心功能包括环境初始化快照收集，能够自动获取工作目录、文件列表等信息，并将其注入初始提示中，从而节省了早期探索所需的2到5轮交互。Meta-Harness 适用于需要高效执行命令行操作或自动化脚本测试的场景，特别是在评估AI助手在终端环境中解决问题能力时尤为有用。",2,"2026-06-11 03:49:54","high_star"]