Skip to content

已核验参考文献 ​

研究登记表共 174 项,其中 146 项已核验身份与版本且可用于技术事实引用的来源列于下方;其余待核验或社区发现线索保留在仓库。资料维护至 2026-09-19,产品行为以各章版本为准。来源身份核验不等于全文每个结论均被独立证明。

  1. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / README.md。版本:git cd5ef8148158c3a752a658978873241fdf8e2bbc;访问:2026-09-19T09:44:36.928264+00:00。

  2. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / docs/architecture.md。版本:git cd5ef8148158c3a752a658978873241fdf8e2bbc;访问:2026-09-19T09:44:36.953863+00:00。

  3. 机构/作者未登记 (n.d.). Cordis Primer。版本:git cd5ef8148158c3a752a658978873241fdf8e2bbc;访问:2026-08-28。

  4. Zhang et al. (n.d.). Self-Harness: Harnesses That Improve Themselves。版本:arXiv:2606.09498v3;访问:2026-09-19T09:53:07.552Z。

  5. Luo et al. (n.d.). Self-Evolving Agent Harnesses via Gated Semantic Quality-Diversity。版本:arXiv:2607.13683v1;访问:2026-09-19T09:55:33.641Z。

  6. Du et al. (n.d.). Living-Harness Is an Interactive-Agent Evolver。版本:arXiv:2607.26598v2;访问:2026-09-19T09:53:07.560Z。

  7. Tailin Zhou (n.d.). Hierarchical Self-Improvement: A Framework for Task-Specific Evolvable Agent Harnesses。版本:arXiv:2608.08466v1;访问:2026-09-19T09:53:07.542Z。

  8. 机构/作者未登记 (n.d.). Unrolling the Codex agent loop | OpenAI。版本:web snapshot 2026-09-19;访问:2026-09-19T11:23:43.915Z。

  9. 机构/作者未登记 (n.d.). Unlocking the Codex harness: how we built the App Server | OpenAI。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:42.679Z。

  10. 机构/作者未登记 (n.d.). Introducing the Codex app。版本:web snapshot 2026-08-28;访问:2026-08-28。

  11. 机构/作者未登记 (n.d.). Harness engineering: leveraging Codex in an agent-first world | OpenAI。版本:web snapshot 2026-09-19;访问:2026-09-19T11:23:43.962Z。

  12. 机构/作者未登记 (n.d.). openai/codex。版本:repository metadata snapshot 2026-09-19;访问:2026-09-19T09:43:24.100927+00:00。

  13. 机构/作者未登记 (n.d.). 持续改进我们的智能体框架 · Cursor。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:27.160Z。

  14. 机构/作者未登记 (n.d.). Dynamic context discovery · Cursor。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:27.172Z。

  15. 机构/作者未登记 (n.d.). Runtime Architecture - OpenHands Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:27.253Z。

  16. Earendil Works / Mario Zechner (2026). pi/packages/coding-agent/README.md at main · earendil-works/pi。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:37.437Z。

  17. 机构/作者未登记 (n.d.). ReAct: Synergizing Reasoning and Acting in Language Models。版本:arXiv:2210.03629v3;访问:2026-09-19T11:45:11.199Z。

  18. 机构/作者未登记 (n.d.). Reflexion: Language Agents with Verbal Reinforcement Learning。版本:arXiv:2303.11366v4;访问:2026-09-19T11:45:11.193Z。

  19. 机构/作者未登记 (n.d.). Voyager: An Open-Ended Embodied Agent with Large Language Models。版本:arXiv:2305.16291v2;访问:2026-09-19T11:45:11.205Z。

  20. Richard Fikes; Nils Nilsson (1971). STRIPS: A New Approach to the Application of Theorem Proving to Problem Solving。版本:Artificial Intelligence 2 (1971), pages 189-208; author-hosted original PDF;访问:2026-09-19T12:04:42.938Z。

  21. Reid G. Smith (1980). The Contract Net Protocol: High-Level Communication and Control in a Distributed Problem Solver | IEEE Journals & Magazine | IEEE Xplore。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:37.414Z。

  22. 机构/作者未登记 (2020). BDI Agent Architectures: A Survey | IJCAI。版本:web snapshot 2026-09-19;访问:2026-09-19T11:12:16.586Z。

  23. 机构/作者未登记 (2024). SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering。版本:arXiv:2405.15793v3;访问:2026-09-19T11:49:17.153Z。

  24. 机构/作者未登记 (2023). babyagi_archive/babyagi.py at main · yoheinakajima/babyagi_archive。版本:web snapshot 2026-09-19;访问:2026-09-19T11:24:43.773Z。

  25. 机构/作者未登记 (2023). Autonomous Agents & Agent Simulations。版本:web snapshot 2026-09-19;访问:2026-09-19T11:17:44.573Z。

  26. 机构/作者未登记 (2025). Building LangGraph: Designing an Agent Runtime from first principles。版本:web snapshot 2026-09-19;访问:2026-09-19T11:24:43.754Z。

  27. 机构/作者未登记 (n.d.). Repository map | aider。版本:web snapshot 2026-09-19;访问:2026-09-19T11:17:44.588Z。

  28. 机构/作者未登记 (n.d.). GPT code editing benchmarks | aider。版本:web snapshot 2026-09-19;访问:2026-09-19T11:17:44.580Z。

  29. 机构/作者未登记 (n.d.). Building Effective AI Agents \ Anthropic。版本:web snapshot 2026-09-19;访问:2026-09-19T11:23:43.972Z。

  30. 机构/作者未登记 (n.d.). How the agent loop works - Claude Code Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:20.843Z。

  31. 机构/作者未登记 (n.d.). How Claude Code works - Claude Code Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T11:21:13.461Z。

  32. 机构/作者未登记 (n.d.). Idempotency and retries - AWS Durable Execution SDK Developer Guide。版本:web snapshot 2026-09-19;访问:2026-09-19T11:22:42.486Z。

  33. 机构/作者未登记 (2024). Lost in the Middle: How Language Models Use Long Contexts - ACL Anthology。版本:web snapshot 2026-09-19;访问:2026-09-19T11:24:43.782Z。

  34. 机构/作者未登记 (n.d.). How Claude remembers your project - Claude Code Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T11:21:13.439Z。

  35. 机构/作者未登记 (n.d.). Explore the context window - Claude Code Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T11:21:13.453Z。

  36. 机构/作者未登记 (n.d.). Architecture - Model Context Protocol。版本:MCP specification 2025-06-18; web snapshot 2026-09-19;访问:2026-09-19T11:22:39.243Z。

  37. 机构/作者未登记 (n.d.). Schema Reference - Model Context Protocol。版本:MCP specification 2025-11-25; web snapshot 2026-09-19;访问:2026-09-19T11:22:42.560Z。

  38. 机构/作者未登记 (n.d.). Authorization - Model Context Protocol。版本:MCP specification 2025-11-25; web snapshot 2026-09-19;访问:2026-09-19T11:22:42.478Z。

  39. 机构/作者未登记 (n.d.). Scale to many tools with tool search - Claude Code Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T11:21:13.530Z。

  40. 机构/作者未登记 (n.d.). deepseek-harness/packages/core/tools/README.md at master · deepseek-ai/deepseek-harness。版本:web snapshot 2026-09-19;访问:2026-09-19T11:27:11.346Z。

  41. 机构/作者未登记 (n.d.). Making Claude Code more secure and autonomous with sandboxing \ Anthropic。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:42.689Z。

  42. 机构/作者未登记 (n.d.). codex/codex-rs/execpolicy/README.md at main · openai/codex。版本:web snapshot 2026-09-19;访问:2026-09-19T11:27:11.358Z。

  43. Carlos E. Jimenez et al. (2023). SWE-bench: Can Language Models Resolve Real-World GitHub Issues?。版本:arXiv:2310.06770v3;访问:2026-09-19T11:45:11.210Z。

  44. OpenAI (2024). Introducing SWE-bench Verified | OpenAI。版本:web snapshot 2026-09-19;访问:2026-09-19T11:23:43.941Z。

  45. OpenAI (2026). Why SWE-bench Verified no longer measures frontier coding capabilities | OpenAI。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:42.659Z。

  46. Anthropic (2026). Demystifying evals for AI agents \ Anthropic。版本:web snapshot 2026-09-19;访问:2026-09-19T11:24:43.765Z。

  47. Koki Wataoka, Tsubasa Takahashi, Ryokan Ri (2024). Self-Preference Bias in LLM-as-a-Judge。版本:arXiv:2410.21819v2;访问:2026-09-19T11:49:17.172Z。

  48. Lin Shi et al. (2024). Judging the Judges: A Systematic Study of Position Bias in LLM-as-a-Judge。版本:arXiv:2406.07791v9;访问:2026-09-19T11:49:17.181Z。

  49. Jonathan Gabor, Jayson Lynch, Jonathan Rosenfeld (2025). EvilGenie: A Reward Hacking Benchmark。版本:arXiv:2511.21654v2;访问:2026-09-19T11:49:17.165Z。

  50. Bingchen Zhao et al. (2026). SpecBench: Measuring Reward Hacking in Long-Horizon Coding Agents。版本:arXiv:2605.21384v1;访问:2026-08-28。

  51. Anthropic (2025). How we built our multi-agent research system \ Anthropic。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:47.317Z。

  52. OpenAI (2026). openai-agents-python/docs/handoffs.md at main · openai/openai-agents-python。版本:web snapshot 2026-09-19;访问:2026-09-19T11:27:11.396Z。

  53. Anthropic (2026). Extend Claude Code - Claude Code Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:20.870Z。

  54. Anthropic (2026). Automate actions with hooks - Claude Code Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:20.863Z。

  55. Cursor (2026). Cursor Agent Security。版本:web snapshot 2026-08-28;访问:2026-08-28。

  56. Cursor (2026). Cursor Background Agents。版本:web snapshot 2026-08-28;访问:2026-08-28。

  57. Cursor (2026). 钩子 | Cursor Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T12:18:51.997Z。

  58. DeepSeek AI (2026). deepseek-ai/deepseek-harness / docs/tool-catalog.md。版本:git cd5ef8148158c3a752a658978873241fdf8e2bbc;访问:2026-09-19T09:44:37.154753+00:00。

  59. OpenAI (2026). openai-agents-python/docs/agents.md at main · openai/openai-agents-python。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:37.447Z。

  60. Model Context Protocol (2025). Key Changes - Model Context Protocol。版本:MCP specification 2025-11-25; web snapshot 2026-09-19;访问:2026-09-19T11:49:42.634Z。

  61. Microsoft Research (2026). AutoGen - Microsoft Research: Publications。版本:web snapshot 2026-09-19;访问:2026-09-19T11:49:47.330Z。

  62. 机构/作者未登记 (n.d.). OpenHands: An Open Platform for AI Software Developers as Generalist Agents。版本:arXiv:2407.16741v3;访问:2026-09-19T11:13:46.085Z。

  63. 机构/作者未登记 (n.d.). Toolformer: Language Models Can Teach Themselves to Use Tools。版本:arXiv:2302.04761v1;访问:2026-09-19T11:12:46.050Z。

  64. 机构/作者未登记 (n.d.). AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation。版本:arXiv:2308.08155v2;访问:2026-09-19T11:12:46.058Z。

  65. 机构/作者未登记 (n.d.). Executable Code Actions Elicit Better LLM Agents。版本:arXiv:2402.01030v4;访问:2026-09-19T11:12:16.595Z。

  66. 机构/作者未登记 (n.d.). AI Harness Engineering: A Runtime Substrate for Foundation-Model Software Agents。版本:arXiv:2605.13357v1;访问:2026-09-19T11:17:44.562Z。

  67. 机构/作者未登记 (n.d.). MemGPT: Towards LLMs as Operating Systems。版本:arXiv:2310.08560v2;访问:2026-09-19T11:12:46.065Z。

  68. 机构/作者未登记 (n.d.). MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework。版本:arXiv:2308.00352v7;访问:2026-09-19T11:12:46.070Z。

  69. 机构/作者未登记 (n.d.). Why Do Multi-Agent LLM Systems Fail?。版本:arXiv:2503.13657v3;访问:2026-09-19T11:13:46.093Z。

  70. 机构/作者未登记 (n.d.). Can LLM Agents Really Debate? A Controlled Study of Multi-Agent Debate in Logical Reasoning。版本:arXiv:2511.07784v1;访问:2026-09-19T11:13:46.108Z。

  71. 机构/作者未登记 (n.d.). MultiAgentBench: Evaluating the Collaboration and Competition of LLM agents。版本:arXiv:2503.01935v1;访问:2026-09-19T11:13:46.101Z。

  72. OpenAI (n.d.). Changelog | OpenAI API。版本:web snapshot 2026-09-19;访问:2026-09-19T09:38:31.208Z。

  73. OpenAI (n.d.). Agents API | OpenAI API。版本:web snapshot 2026-09-19;访问:2026-09-19T09:34:14.265Z。

  74. Anthropic (n.d.). Claude Platform release notes - Claude Platform Docs。版本:web snapshot 2026-09-19;访问:2026-09-19T09:34:14.296Z。

  75. Anthropic (n.d.). Scaling Managed Agents: Decoupling the brain from the hands \ Anthropic。版本:web snapshot 2026-09-19;访问:2026-09-19T09:55:33.629Z。

  76. Cursor (n.d.). 在您自行管理的机器上运行云端智能体 · Cursor。版本:web snapshot 2026-09-19;访问:2026-09-19T09:34:14.310Z。

  77. Cursor (n.d.). Cursor 最新动态 — 最新更新与发布说明。版本:web snapshot 2026-09-19;访问:2026-09-19T09:38:31.226Z。

  78. T. Bengre; C. Curme / LangChain (n.d.). Organizing Context in a Multi-Agent Harness。版本:web snapshot 2026-09-19;访问:2026-09-19T09:34:14.279Z。

  79. Microsoft (n.d.). Agent Harness | Microsoft Learn。版本:web snapshot 2026-09-19;访问:2026-09-19T09:44:20.634Z。

  80. Wu et al. (n.d.). HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?。版本:arXiv:2609.01437v1;访问:2026-09-19T09:45:11.363Z。

  81. Yan et al. (n.d.). Harness-of-Harness: Multi-Day Autonomous Software Development with Continual Improvement。版本:arXiv:2609.01481v1;访问:2026-09-19T09:45:11.377Z。

  82. Zhang et al. (n.d.). JIT-Agent: Scaling Harness Intelligence via Just-in-Time Harness Evolution。版本:arXiv:2608.25593v2;访问:2026-09-19T09:45:11.405Z。

  83. Jiang et al. (n.d.). HarnessEvolve: Learning from Reference Trajectories for Reliable Agent Self-Evolution。版本:arXiv:2609.00829v1;访问:2026-09-19T09:52:13.305Z。

  84. Fan et al. (n.d.). An Empirical Study of Harness Design for Coding Agents。版本:arXiv:2609.20804v1;访问:2026-09-19T09:45:11.392Z。

  85. Li et al. (n.d.). A Blind Trust, the Bloody Thrust: When Attacker-Controlled Hook UpdatesSteer AI Agent Harnesses towards Malicious Behaviors。版本:arXiv:2609.03884v2;访问:2026-09-19T09:52:13.352Z。

  86. Chen et al. (n.d.). MemSecBench: Tracking Agent Memory Poisoning from Persistence to Consequence and Repair。版本:arXiv:2607.27080v1;访问:2026-09-19T09:50:41.795Z。

  87. Ben Brandt / ACP (n.d.). ACP v2 is available in Draft - Agent Client Protocol。版本:web snapshot 2026-09-19;访问:2026-09-19T09:52:13.322Z。

  88. Palash Shah / LangChain (n.d.). LangSmith Engine: How We Built an Agent for Improving Agents。版本:web snapshot 2026-09-19;访问:2026-09-19T09:52:13.337Z。

  89. Jin et al. (n.d.). Harness Engineering in LLM Tool Use via Agent-Native Reusable Tool Primitives。版本:arXiv:2609.01736v1;访问:2026-09-19T09:44:20.654Z。

  90. Duan et al. (n.d.). The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement。版本:arXiv:2609.11873v2;访问:2026-09-19T10:00:34.225Z。

  91. Chen et al. (n.d.). Show-Harness: Just a VLM Agent Can Play Robots。版本:arXiv:2609.10522v1;访问:2026-09-19T09:54:21.136Z。

  92. Yue et al. (n.d.). Ecdysis: Efficient and Effective Training of Runtime Harnesses for LLM Agents。版本:arXiv:2609.11677v1;访问:2026-09-19T10:00:34.247Z。

  93. Lin et al. (n.d.). Stellar Colosseum: A Many-Agent Harness for Long-Horizon Research in Mathematics and Theoretical Computer Science。版本:arXiv:2609.15983v2;访问:2026-09-19T10:00:34.237Z。

  94. Liu et al. (n.d.). SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness。版本:arXiv:2609.20519v1;访问:2026-09-19T10:00:34.203Z。

  95. Luo et al. (n.d.). HarnessBank: Semantic Gene-Bank Search with Gated Verification for Agent-Harness Self-Evolution。版本:arXiv:2607.13683v2;访问:2026-09-19T09:55:33.606Z。

  96. NLE authors / Facebook Research (n.d.). facebookresearch/nle: The NetHack Learning Environment。版本:web snapshot 2026-09-19;访问:2026-09-19T09:55:33.617Z。

  97. Taylor Mullen; Christian Gunderman / Google (n.d.). The Anatomy of Harness Engineering: How to Evaluate, Iterate, and Guard AI Coding Agents - Google Developers Blog。版本:web snapshot 2026-09-19;访问:2026-09-19T09:59:03.556Z。

  98. openJiuwen Team (n.d.). openJiuwen: Beyond Static Harnesses for Long-Horizon Coding Agents。版本:arXiv:2608.27969v1;访问:2026-09-19T09:35:42.335Z。

  99. Google Cloud (n.d.). Query syntax  |  BigQuery  |  Google Cloud Documentation。版本:web snapshot 2026-09-19;访问:2026-09-19T10:14:48.326Z。

  100. 机构/作者未登记 (n.d.). openai/codex / codex-rs/app-server/README.md。版本:git 426fa8cdab4247e5623e9617d531f6917482b947;访问:2026-09-19T09:44:39.463010+00:00。

  101. 机构/作者未登记 (n.d.). openai/codex / codex-rs/app-server-protocol/src/protocol/common.rs。版本:git be2951ea34f0d295ed0becf97079f92fa5f6950e;访问:2026-09-19T09:54:19.199639+00:00。

  102. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / README.md。版本:git ddefc45fbc7f8e46dd73185e68295696d1297887;访问:2026-09-19T09:44:36.923997+00:00。

  103. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / packages/extensions/cordis-host-runner/src/sandbox.ts。版本:git ddefc45fbc7f8e46dd73185e68295696d1297887;访问:2026-09-19T09:49:04.757600+00:00。

  104. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / packages/code-runtime/code-runtime-worker-thread/README.md。版本:git cd5ef8148158c3a752a658978873241fdf8e2bbc;访问:2026-09-19T09:54:18.498962+00:00。

  105. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / packages/ptc-runtime/ptc-runtime-node/README.md。版本:git ddefc45fbc7f8e46dd73185e68295696d1297887;访问:2026-09-19T09:51:32.840667+00:00。

  106. 机构/作者未登记 (n.d.). OpenHands/OpenHands / openhands/runtime/README.md。版本:git 7fbb48c40679afd674970966b96185657d92a487;访问:2026-09-19T09:54:19.573502+00:00。

  107. 机构/作者未登记 (n.d.). OpenHands/software-agent-sdk / openhands-sdk/openhands/sdk/conversation/conversation.py。版本:git d128a786ee2ee570eb23ff5862ec148b43cfad0b;访问:2026-09-19T09:54:19.421087+00:00。

  108. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / packages/boot/plugin-manager/README.md。版本:git ddefc45fbc7f8e46dd73185e68295696d1297887;访问:2026-09-19T09:49:04.882477+00:00。

  109. 机构/作者未登记 (n.d.). fix(sdk): persist events before publishing them, return the assigned seq。版本:git 94fca578b720df758b9bbf8a2639511b303c78e6;访问:2026-09-19T09:49:07.883806+00:00。

  110. 机构/作者未登记 (n.d.). feat(agent-server): add /sockets/session/{id} with a non-Event envelope。版本:git 2ab274897ac5e2c66b0ba17e9a6d39367b769876;访问:2026-09-19T09:49:07.828869+00:00。

  111. 机构/作者未登记 (n.d.). feat(agent-server): add docker runtime mode for per-conversation containers。版本:git 3ff6924d8564b3d47a22a6c7e71377a701ae014f;访问:2026-09-19T09:54:20.078080+00:00。

  112. 机构/作者未登记 (n.d.). Architecture - Model Context Protocol。版本:MCP specification 2026-07-28; web snapshot 2026-09-19;访问:2026-09-19T11:27:11.371Z。

  113. 机构/作者未登记 (n.d.). Codex local schema probe。版本:local Codex 0.142.5 probe 2026-09-19;访问:2026-09-19。

  114. 机构/作者未登记 (n.d.). Codex local initialize probe。版本:local Codex 0.142.5 probe 2026-09-19;访问:2026-09-19。

  115. 机构/作者未登记 (n.d.). OpenHands/OpenHands / openhands/runtime/action_execution_server.py。版本:git 7fbb48c40679afd674970966b96185657d92a487;访问:2026-09-19T09:51:34.569498+00:00。

  116. 机构/作者未登记 (n.d.). OpenHands/OpenHands / package.json。版本:git 9737f713616a1e452f822c2967f0e2c8bf2dc308;访问:2026-09-19T09:54:19.076593+00:00。

  117. 机构/作者未登记 (n.d.). OpenHands/OpenHands / README.md。版本:git b50c60c6728e2ce123ccb6e125bee3eb88ac87d1;访问:2026-09-19T09:44:39.765917+00:00。

  118. 机构/作者未登记 (n.d.). OpenHands/OpenHands 1.0.0。版本:release 1.0.0; published_at=2025-12-16T16:03:32Z;访问:2026-09-19T09:51:33.954077+00:00。

  119. 机构/作者未登记 (n.d.). OpenHands/OpenHands v1.20.0。版本:release v1.20.0; published_at=2026-09-17T07:15:18Z;访问:2026-09-19T09:43:26.012204+00:00。

  120. 机构/作者未登记 (n.d.). OpenHands/software-agent-sdk / openhands-agent-server/openhands/agent_server/docker_runtime/provisioning.py。版本:git d128a786ee2ee570eb23ff5862ec148b43cfad0b;访问:2026-09-19T09:54:19.190723+00:00。

  121. 机构/作者未登记 (n.d.). OpenHands/software-agent-sdk v1.45.0。版本:release v1.45.0; published_at=2026-09-07T03:07:34Z;访问:2026-09-19T09:43:27.809445+00:00。

  122. 机构/作者未登记 (n.d.). OpenHands/software-agent-sdk v1.48.0。版本:release v1.48.0; published_at=2026-09-15T20:58:25Z;访问:2026-09-19T09:43:27.809445+00:00。

  123. 机构/作者未登记 (n.d.). OpenHands/software-agent-sdk v1.49.1。版本:release v1.49.1; published_at=2026-09-17T04:35:48Z;访问:2026-09-19T09:43:27.809445+00:00。

  124. 机构/作者未登记 (n.d.). OpenHands/software-agent-sdk v1.49.2。版本:release v1.49.2; published_at=2026-09-17T20:46:52Z;访问:2026-09-19T09:43:26.388854+00:00。

  125. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / docs/architecture.md。版本:git cd5ef8148158c3a752a658978873241fdf8e2bbc;访问:2026-09-19T09:44:36.953863+00:00。

  126. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / docs/tool-catalog.md。版本:git cd5ef8148158c3a752a658978873241fdf8e2bbc;访问:2026-09-19T09:44:37.154753+00:00。

  127. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / docs/architecture.md。版本:git ddefc45fbc7f8e46dd73185e68295696d1297887;访问:2026-09-19T09:44:36.981066+00:00。

  128. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness / docs/tool-catalog.md。版本:git ddefc45fbc7f8e46dd73185e68295696d1297887;访问:2026-09-19T09:44:37.714636+00:00。

  129. 机构/作者未登记 (n.d.). feat(code-runtime): execute Node programs through confined processes。版本:git 75ed8da3e0c9103b3b2174b2981b7129e1fba21d;访问:2026-09-19T09:57:13.361793+00:00。

  130. 机构/作者未登记 (n.d.). refactor(ptc): align runtime packages and services with PTC naming。版本:git 7c9bb5914cedec80e46197a8c894037fcfd12faf;访问:2026-09-19T09:54:19.613914+00:00。

  131. 机构/作者未登记 (n.d.). feat: add current-profile plugin manager service and Web controls。版本:git 98b92b683c39fc60771daa774492105c2d3e8076;访问:2026-09-19T09:51:33.515620+00:00。

  132. 机构/作者未登记 (n.d.). feat(agent): await initialization through agent/created。版本:git 9b7a8ccc9fabc2e87386acf7f8b0741baf978022;访问:2026-09-19T09:54:18.826378+00:00。

  133. 机构/作者未登记 (n.d.). refactor(hmr): own profile reload lifecycle through YAML。版本:git abd765a6001ff9d9c9772b8b407e0b7f18fe25ab;访问:2026-09-19T09:51:33.188488+00:00。

  134. 机构/作者未登记 (n.d.). feat(session)!: add released format migration。版本:git d1521ea7838f19a78a9cca7b4a93622d301149bb;访问:2026-09-19T09:51:35.337469+00:00。

  135. 机构/作者未登记 (n.d.). Revert #932 transactional Cordis reload changes。版本:git e07f41d5fd8ca172287fda0f923b4d1f69c592f3;访问:2026-09-19T09:51:33.122985+00:00。

  136. 机构/作者未登记 (n.d.). feat(creator): use Plugin Manager for persistent plugins。版本:git ed32f57f88ef6bba983e30a0b434fe5d77e5773b;访问:2026-09-19T09:51:33.349909+00:00。

  137. 机构/作者未登记 (n.d.). feat(session)!: embed assistant streams in format v2。版本:git f99b06eaed81d6fe4fc64d44687450e18ef68a67;访问:2026-09-19T09:51:34.765805+00:00。

  138. 机构/作者未登记 (n.d.). deepseek-ai/deepseek-harness dsh-v0.1.6-alpha.2。版本:release dsh-v0.1.6-alpha.2; published_at=2026-09-17T13:30:16Z;访问:2026-09-19T09:43:26.873237+00:00。

  139. 机构/作者未登记 (n.d.). openai/codex / codex-rs/app-server-protocol/src/protocol/common.rs。版本:git 426fa8cdab4247e5623e9617d531f6917482b947;访问:2026-09-19T09:44:39.266059+00:00。

  140. 机构/作者未登记 (n.d.). openai/codex python-v0.154.0。版本:release python-v0.154.0; published_at=2026-09-10T19:51:43Z;访问:2026-09-19T09:43:37.004725+00:00。

  141. 机构/作者未登记 (n.d.). openai/codex rust-v0.152.0。版本:release rust-v0.152.0; published_at=2026-09-01T01:58:32Z;访问:2026-09-19T09:43:37.004725+00:00。

  142. 机构/作者未登记 (n.d.). openai/codex rust-v0.153.0。版本:release rust-v0.153.0; published_at=2026-09-03T01:37:38Z;访问:2026-09-19T09:43:37.004725+00:00。

  143. 机构/作者未登记 (n.d.). openai/codex rust-v0.154.0。版本:release rust-v0.154.0; published_at=2026-09-09T22:35:38Z;访问:2026-09-19T09:43:37.004725+00:00。

  144. 机构/作者未登记 (n.d.). openai/codex rust-v0.155.1。版本:release rust-v0.155.1; published_at=2026-09-18T20:03:04Z;访问:2026-09-19T09:43:26.336187+00:00。

  145. 机构/作者未登记 (n.d.). Cognitive Architectures for Language Agents。版本:arXiv:2309.02427v3;访问:2026-09-19T11:12:16.577Z。

  146. 机构/作者未登记 (n.d.). SpecBench: Measuring Reward Hacking in Long-Horizon Coding Agents。版本:arXiv:2605.21384v2;访问:2026-09-19T11:49:20.851Z。

公开阅读版本