hud 0.6.16.dev4__tar.gz → 0.6.17__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (320) hide show
  1. {hud-0.6.16.dev4 → hud-0.6.17}/.gitignore +0 -2
  2. {hud-0.6.16.dev4 → hud-0.6.17}/PKG-INFO +17 -15
  3. {hud-0.6.16.dev4 → hud-0.6.17}/README.md +14 -12
  4. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/agent.py +6 -0
  5. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/sdk/agent.py +18 -5
  6. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/agent.py +25 -2
  7. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_base.py +1 -0
  8. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_claude_agent.py +44 -1
  9. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_claude_sdk_agent.py +27 -6
  10. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_openai_agent.py +38 -1
  11. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_tool_agent.py +106 -2
  12. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tool_agent.py +21 -5
  13. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/types.py +6 -0
  14. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/mcp.py +27 -2
  15. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/__init__.py +4 -4
  16. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/deploy.py +17 -0
  17. hud-0.6.17/hud/cli/init.py +129 -0
  18. hud-0.6.17/hud/cli/presets.py +115 -0
  19. hud-0.6.17/hud/cli/project.py +198 -0
  20. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/sync.py +25 -2
  21. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/task.py +34 -20
  22. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_deploy.py +93 -0
  23. hud-0.6.17/hud/cli/tests/test_init.py +276 -0
  24. hud-0.6.17/hud/cli/tests/test_project_command.py +133 -0
  25. hud-0.6.17/hud/cli/tests/test_sync_projects.py +176 -0
  26. hud-0.6.17/hud/cli/tests/test_task.py +43 -0
  27. hud-0.6.17/hud/cli/tests/test_trace.py +47 -0
  28. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/trace.py +0 -13
  29. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/api.py +1 -2
  30. hud-0.6.17/hud/cli/utils/project.py +270 -0
  31. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/source.py +6 -0
  32. hud-0.6.17/hud/cli/utils/tests/test_project.py +276 -0
  33. {hud-0.6.16.dev4 → hud-0.6.17}/hud/clients/client.py +22 -12
  34. {hud-0.6.16.dev4 → hud-0.6.17}/hud/clients/tests/test_connect.py +17 -5
  35. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/server.py +54 -12
  36. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_server.py +81 -2
  37. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_sessions.py +23 -2
  38. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_tunnel.py +18 -1
  39. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_workspace.py +27 -0
  40. hud-0.6.17/hud/environment/utils.py +134 -0
  41. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/workspace.py +3 -1
  42. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/job.py +2 -0
  43. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/run.py +10 -2
  44. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/runtime/hosted.py +42 -11
  45. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/sync.py +4 -0
  46. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_hosted.py +96 -1
  47. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_job.py +2 -0
  48. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_rollout.py +114 -13
  49. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_sync.py +35 -0
  50. {hud-0.6.16.dev4 → hud-0.6.17}/hud/settings.py +6 -0
  51. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/context.py +21 -10
  52. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/span.py +4 -9
  53. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/gateway.py +2 -4
  54. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/tests/test_gateway.py +32 -0
  55. {hud-0.6.16.dev4 → hud-0.6.17}/hud/version.py +1 -1
  56. {hud-0.6.16.dev4 → hud-0.6.17}/pyproject.toml +4 -2
  57. hud-0.6.16.dev4/hud/cli/init.py +0 -187
  58. hud-0.6.16.dev4/hud/cli/login.py +0 -238
  59. hud-0.6.16.dev4/hud/cli/presets.py +0 -217
  60. hud-0.6.16.dev4/hud/cli/templates.py +0 -141
  61. hud-0.6.16.dev4/hud/cli/tests/test_init.py +0 -134
  62. hud-0.6.16.dev4/hud/cli/tests/test_trace.py +0 -26
  63. hud-0.6.16.dev4/hud/environment/utils.py +0 -84
  64. {hud-0.6.16.dev4 → hud-0.6.17}/LICENSE +0 -0
  65. {hud-0.6.16.dev4 → hud-0.6.17}/hud/__init__.py +0 -0
  66. {hud-0.6.16.dev4 → hud-0.6.17}/hud/__main__.py +0 -0
  67. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/__init__.py +0 -0
  68. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/base.py +0 -0
  69. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/browser_use/__init__.py +0 -0
  70. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/browser_use/agent.py +0 -0
  71. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/__init__.py +0 -0
  72. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/sdk/__init__.py +0 -0
  73. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/sdk/computer_mcp.py +0 -0
  74. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/tools/__init__.py +0 -0
  75. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/tools/base.py +0 -0
  76. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/tools/coding.py +0 -0
  77. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/tools/computer.py +0 -0
  78. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/tools/hosted.py +0 -0
  79. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/tools/mcp_proxy.py +0 -0
  80. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/tools/settings.py +0 -0
  81. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/tools/tests/__init__.py +0 -0
  82. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/claude/tools/tests/test_computer.py +0 -0
  83. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/__init__.py +0 -0
  84. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/agent.py +0 -0
  85. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/settings.py +0 -0
  86. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/tools/__init__.py +0 -0
  87. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/tools/base.py +0 -0
  88. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/tools/coding.py +0 -0
  89. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/tools/computer.py +0 -0
  90. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/tools/filesystem.py +0 -0
  91. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/tools/hosted.py +0 -0
  92. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/tools/mcp_proxy.py +0 -0
  93. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/tools/tests/__init__.py +0 -0
  94. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/gemini/tools/tests/test_computer.py +0 -0
  95. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/misc/__init__.py +0 -0
  96. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/misc/response_automation.py +0 -0
  97. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/__init__.py +0 -0
  98. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/__init__.py +0 -0
  99. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/apply_patch.py +0 -0
  100. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/base.py +0 -0
  101. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/coding.py +0 -0
  102. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/computer.py +0 -0
  103. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/hosted.py +0 -0
  104. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/mcp_proxy.py +0 -0
  105. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/strict_schema.py +0 -0
  106. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/tests/__init__.py +0 -0
  107. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/tests/test_computer.py +0 -0
  108. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai/tools/tests/test_strict_schema.py +0 -0
  109. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai_compatible/__init__.py +0 -0
  110. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai_compatible/agent.py +0 -0
  111. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai_compatible/tools/__init__.py +0 -0
  112. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai_compatible/tools/base.py +0 -0
  113. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai_compatible/tools/filesystem.py +0 -0
  114. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/openai_compatible/tools/mcp_proxy.py +0 -0
  115. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/robot/__init__.py +0 -0
  116. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/robot/adapter.py +0 -0
  117. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/robot/agent.py +0 -0
  118. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/robot/batching.py +0 -0
  119. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/robot/dataset.py +0 -0
  120. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/robot/model.py +0 -0
  121. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/__init__.py +0 -0
  122. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_apply_patch.py +0 -0
  123. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_gemini_agent.py +0 -0
  124. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_openai_compatible_agent.py +0 -0
  125. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_provider_native_tools.py +0 -0
  126. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_robot_dataset_writer.py +0 -0
  127. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tests/test_trace.py +0 -0
  128. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tools/__init__.py +0 -0
  129. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tools/base.py +0 -0
  130. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tools/hosted.py +0 -0
  131. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tools/mcp.py +0 -0
  132. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tools/rfb.py +0 -0
  133. {hud-0.6.16.dev4 → hud-0.6.17}/hud/agents/tools/ssh.py +0 -0
  134. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/__init__.py +0 -0
  135. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/base.py +0 -0
  136. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/cdp.py +0 -0
  137. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/rfb.py +0 -0
  138. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/robot.py +0 -0
  139. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/ssh.py +0 -0
  140. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/tests/__init__.py +0 -0
  141. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/tests/test_rfb.py +0 -0
  142. {hud-0.6.16.dev4 → hud-0.6.17}/hud/capabilities/tests/test_ssh.py +0 -0
  143. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/__main__.py +0 -0
  144. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/cancel.py +0 -0
  145. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/client.py +0 -0
  146. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/eval.py +0 -0
  147. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/jobs.py +0 -0
  148. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/models.py +0 -0
  149. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/qa.py +0 -0
  150. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/qa_analysis.py +0 -0
  151. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/serve.py +0 -0
  152. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/__init__.py +0 -0
  153. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_cli_init.py +0 -0
  154. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_cli_main.py +0 -0
  155. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_cli_more_wrappers.py +0 -0
  156. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_eval_bedrock.py +0 -0
  157. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_eval_config.py +0 -0
  158. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_jobs.py +0 -0
  159. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_main_module.py +0 -0
  160. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_qa.py +0 -0
  161. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_qa_analysis.py +0 -0
  162. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_sync_export.py +0 -0
  163. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/tests/test_usage.py +0 -0
  164. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/__init__.py +0 -0
  165. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/build_display.py +0 -0
  166. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/build_logs.py +0 -0
  167. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/config.py +0 -0
  168. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/context.py +0 -0
  169. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/display.py +0 -0
  170. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/jobs.py +0 -0
  171. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/registry.py +0 -0
  172. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/tasks.py +0 -0
  173. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/tests/__init__.py +0 -0
  174. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/tests/test_build_display.py +0 -0
  175. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/tests/test_config.py +0 -0
  176. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/tests/test_context.py +0 -0
  177. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/tests/test_registry.py +0 -0
  178. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/tests/test_source.py +0 -0
  179. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/tests/test_tasks.py +0 -0
  180. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/tests/test_version_check.py +0 -0
  181. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/usage.py +0 -0
  182. {hud-0.6.16.dev4 → hud-0.6.17}/hud/cli/utils/version_check.py +0 -0
  183. {hud-0.6.16.dev4 → hud-0.6.17}/hud/clients/__init__.py +0 -0
  184. {hud-0.6.16.dev4 → hud-0.6.17}/hud/clients/tests/__init__.py +0 -0
  185. {hud-0.6.16.dev4 → hud-0.6.17}/hud/conftest.py +0 -0
  186. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/__init__.py +0 -0
  187. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/arguments.py +0 -0
  188. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/egress.py +0 -0
  189. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/env.py +0 -0
  190. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/file_tracker.py +0 -0
  191. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/namespace.py +0 -0
  192. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/robot/__init__.py +0 -0
  193. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/robot/bridge.py +0 -0
  194. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/robot/endpoint.py +0 -0
  195. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/robot/gym.py +0 -0
  196. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/__init__.py +0 -0
  197. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/conftest.py +0 -0
  198. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_arguments.py +0 -0
  199. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_capability_backing.py +0 -0
  200. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_file_tracker.py +0 -0
  201. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_file_tracking.py +0 -0
  202. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_loader.py +0 -0
  203. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_manifest.py +0 -0
  204. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_platform.py +0 -0
  205. {hud-0.6.16.dev4 → hud-0.6.17}/hud/environment/tests/test_robot_regressions.py +0 -0
  206. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/__init__.py +0 -0
  207. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/chat.py +0 -0
  208. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/docker-seccomp.json +0 -0
  209. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/file_tracking.py +0 -0
  210. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/runtime/__init__.py +0 -0
  211. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/runtime/compose.py +0 -0
  212. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/runtime/core.py +0 -0
  213. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/runtime/daytona.py +0 -0
  214. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/runtime/docker.py +0 -0
  215. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/runtime/hud.py +0 -0
  216. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/runtime/local.py +0 -0
  217. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/runtime/modal.py +0 -0
  218. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/task.py +0 -0
  219. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/taskset.py +0 -0
  220. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/__init__.py +0 -0
  221. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_chat.py +0 -0
  222. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_docker_provider.py +0 -0
  223. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_file_tracking_observer.py +0 -0
  224. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_local_runtime.py +0 -0
  225. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_shared.py +0 -0
  226. {hud-0.6.16.dev4 → hud-0.6.17}/hud/eval/tests/test_task.py +0 -0
  227. {hud-0.6.16.dev4 → hud-0.6.17}/hud/graders/__init__.py +0 -0
  228. {hud-0.6.16.dev4 → hud-0.6.17}/hud/graders/base.py +0 -0
  229. {hud-0.6.16.dev4 → hud-0.6.17}/hud/graders/bash.py +0 -0
  230. {hud-0.6.16.dev4 → hud-0.6.17}/hud/graders/combine.py +0 -0
  231. {hud-0.6.16.dev4 → hud-0.6.17}/hud/graders/judge.py +0 -0
  232. {hud-0.6.16.dev4 → hud-0.6.17}/hud/graders/results.py +0 -0
  233. {hud-0.6.16.dev4 → hud-0.6.17}/hud/graders/text.py +0 -0
  234. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/__init__.py +0 -0
  235. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/Dockerfile +0 -0
  236. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/__init__.py +0 -0
  237. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/adapt.py +0 -0
  238. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/env.py +0 -0
  239. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/export.py +0 -0
  240. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/install.sh +0 -0
  241. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/__init__.py +0 -0
  242. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/conftest.py +0 -0
  243. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/agent-lifecycle/environment/Dockerfile +0 -0
  244. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/agent-lifecycle/instruction.md +0 -0
  245. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/agent-lifecycle/solution/solve.sh +0 -0
  246. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/agent-lifecycle/task.toml +0 -0
  247. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/agent-lifecycle/tests/test.sh +0 -0
  248. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/hello-mcp/environment/Dockerfile +0 -0
  249. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/hello-mcp/environment/docker-compose.yaml +0 -0
  250. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/hello-mcp/environment/mcp-server/Dockerfile +0 -0
  251. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/hello-mcp/environment/mcp-server/__init__.py +0 -0
  252. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/hello-mcp/environment/mcp-server/server.py +0 -0
  253. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/hello-mcp/instruction.md +0 -0
  254. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/hello-mcp/solution/solve.sh +0 -0
  255. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/hello-mcp/task.toml +0 -0
  256. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/hello-mcp/tests/test.sh +0 -0
  257. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/phase-boundary/environment/Dockerfile +0 -0
  258. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/phase-boundary/environment/entrypoint.sh +0 -0
  259. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/phase-boundary/instruction.md +0 -0
  260. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/phase-boundary/solution/solve.sh +0 -0
  261. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/phase-boundary/task.toml +0 -0
  262. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/phase-boundary/tests/test.sh +0 -0
  263. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/sidecar-reachability/environment/Dockerfile +0 -0
  264. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/sidecar-reachability/environment/docker-compose.yaml +0 -0
  265. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/sidecar-reachability/instruction.md +0 -0
  266. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/sidecar-reachability/solution/solve.sh +0 -0
  267. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/sidecar-reachability/task.toml +0 -0
  268. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/sidecar-reachability/tests/Dockerfile +0 -0
  269. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/sidecar-reachability/tests/test.sh +0 -0
  270. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/verifier-lifecycle/environment/Dockerfile +0 -0
  271. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/verifier-lifecycle/environment/entrypoint.sh +0 -0
  272. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/verifier-lifecycle/instruction.md +0 -0
  273. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/verifier-lifecycle/solution/solve.sh +0 -0
  274. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/verifier-lifecycle/task.toml +0 -0
  275. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/tasks/verifier-lifecycle/tests/test.sh +0 -0
  276. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/test_contract.py +0 -0
  277. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/test_harbor.py +0 -0
  278. {hud-0.6.16.dev4 → hud-0.6.17}/hud/integrations/harbor/tests/test_integration.py +0 -0
  279. {hud-0.6.16.dev4 → hud-0.6.17}/hud/patches/__init__.py +0 -0
  280. {hud-0.6.16.dev4 → hud-0.6.17}/hud/patches/mcp_patches.py +0 -0
  281. {hud-0.6.16.dev4 → hud-0.6.17}/hud/patches/tests/__init__.py +0 -0
  282. {hud-0.6.16.dev4 → hud-0.6.17}/hud/patches/tests/test_warnings.py +0 -0
  283. {hud-0.6.16.dev4 → hud-0.6.17}/hud/patches/warnings.py +0 -0
  284. {hud-0.6.16.dev4 → hud-0.6.17}/hud/py.typed +0 -0
  285. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/__init__.py +0 -0
  286. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/exporter.py +0 -0
  287. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/instrument.py +0 -0
  288. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/robot/__init__.py +0 -0
  289. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/robot/recorder.py +0 -0
  290. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/robot/video.py +0 -0
  291. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/tests/__init__.py +0 -0
  292. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/tests/test_exporter.py +0 -0
  293. {hud-0.6.16.dev4 → hud-0.6.17}/hud/telemetry/tests/test_instrument.py +0 -0
  294. {hud-0.6.16.dev4 → hud-0.6.17}/hud/train/__init__.py +0 -0
  295. {hud-0.6.16.dev4 → hud-0.6.17}/hud/train/base.py +0 -0
  296. {hud-0.6.16.dev4 → hud-0.6.17}/hud/train/client.py +0 -0
  297. {hud-0.6.16.dev4 → hud-0.6.17}/hud/train/tests/__init__.py +0 -0
  298. {hud-0.6.16.dev4 → hud-0.6.17}/hud/train/tests/test_client.py +0 -0
  299. {hud-0.6.16.dev4 → hud-0.6.17}/hud/train/types.py +0 -0
  300. {hud-0.6.16.dev4 → hud-0.6.17}/hud/types.py +0 -0
  301. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/__init__.py +0 -0
  302. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/docker.py +0 -0
  303. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/exceptions.py +0 -0
  304. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/hints.py +0 -0
  305. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/hud_console.py +0 -0
  306. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/modules.py +0 -0
  307. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/naming.py +0 -0
  308. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/platform.py +0 -0
  309. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/process.py +0 -0
  310. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/requests.py +0 -0
  311. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/serialization.py +0 -0
  312. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/tests/__init__.py +0 -0
  313. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/tests/test_exceptions.py +0 -0
  314. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/tests/test_hints.py +0 -0
  315. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/tests/test_hud_console.py +0 -0
  316. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/tests/test_platform.py +0 -0
  317. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/tests/test_process.py +0 -0
  318. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/tests/test_requests.py +0 -0
  319. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/tests/test_serialization.py +0 -0
  320. {hud-0.6.16.dev4 → hud-0.6.17}/hud/utils/time.py +0 -0
@@ -62,8 +62,6 @@ hud/rl/checkpoints_test/
62
62
 
63
63
  docs/internal
64
64
 
65
- environments/
66
-
67
65
  experiments/
68
66
  .memories/
69
67
 
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: hud
3
- Version: 0.6.16.dev4
3
+ Version: 0.6.17
4
4
  Summary: SDK for the HUD platform.
5
5
  Project-URL: Homepage, https://github.com/hud-evals/hud-python
6
6
  Project-URL: Bug Tracker, https://github.com/hud-evals/hud-python/issues
@@ -34,7 +34,7 @@ Classifier: Programming Language :: Python :: 3
34
34
  Classifier: Programming Language :: Python :: 3.11
35
35
  Classifier: Programming Language :: Python :: 3.12
36
36
  Requires-Python: <3.13,>=3.11
37
- Requires-Dist: anthropic>=0.78.0
37
+ Requires-Dist: anthropic>=0.98.0
38
38
  Requires-Dist: asyncssh>=2.23.0
39
39
  Requires-Dist: asyncvnc>=1.3.0
40
40
  Requires-Dist: fastmcp==3.2.0
@@ -56,7 +56,7 @@ Requires-Dist: websockets>=15.0.1
56
56
  Provides-Extra: agent
57
57
  Provides-Extra: agents
58
58
  Provides-Extra: bedrock
59
- Requires-Dist: anthropic[bedrock]>=0.78.0; extra == 'bedrock'
59
+ Requires-Dist: anthropic[bedrock]>=0.98.0; extra == 'bedrock'
60
60
  Provides-Extra: browseruse
61
61
  Requires-Dist: browser-use>=0.11.13; extra == 'browseruse'
62
62
  Provides-Extra: daytona
@@ -123,10 +123,12 @@ hud set HUD_API_KEY=your-key-here
123
123
  # or: export HUD_API_KEY=your-key-here
124
124
  ```
125
125
 
126
- Then scaffold your first environment:
126
+ Then start from the coding environment, or choose another example environment:
127
127
 
128
128
  ```bash
129
129
  hud init my-env
130
+ hud init my-desktop-env --template cua
131
+ hud init my-custom-env --template blank
130
132
  ```
131
133
 
132
134
  ![Agent running on SheetBench](docs/src/images/trace_sheet.gif)
@@ -166,16 +168,6 @@ hud sync tasks my-taskset
166
168
  hud eval my-taskset --remote
167
169
  ```
168
170
 
169
- For local iteration, the same protocol works against a container on your laptop:
170
-
171
- ```bash
172
- docker build -f Dockerfile.hud -t my-env .
173
- docker run -d --name run1 -p 8765:8765 my-env
174
- hud task start fix_bug --url tcp://127.0.0.1:8765
175
- hud task grade fix_bug --url tcp://127.0.0.1:8765 --answer "..."
176
- docker rm -f run1
177
- ```
178
-
179
171
  → [Run & deploy](https://docs.hud.ai/v6/reference/runtime)
180
172
 
181
173
  ## Environments & templates
@@ -242,7 +234,17 @@ A **capability** is a connection the environment exposes; a **harness** attaches
242
234
 
243
235
  From the [platform UI](https://hud.ai) you can run batches, compare models on the same taskset, and inspect every trace.
244
236
 
245
- → [Run & deploy](https://docs.hud.ai/v6/reference/runtime)
237
+ A **project** holds the environments and tasksets a team creates and decides who can see them. Deploying without one uses your team's default project, so nothing here is required to get started. To put an environment somewhere else, pin the directory once and every later `hud deploy` and `hud sync tasks` follows it:
238
+
239
+ ```bash
240
+ hud project list # projects you can see
241
+ hud project use browser-evals # writes projectId to .hud/config.json
242
+ hud project # where does a deploy here land?
243
+ ```
244
+
245
+ `hud deploy` and `hud sync tasks` also take `--project <name-or-id>` as a one-command override; it does not change the directory configuration. Use `hud set HUD_DEFAULT_PROJECT=<name-or-id>` for a machine-wide fallback. Precedence is the flag, then the directory's `.hud/config.json`, then `HUD_DEFAULT_PROJECT`, then your team default. An environment or taskset that already exists stays where it is; naming a different project fails rather than moving it.
246
+
247
+ → [Projects](https://docs.hud.ai/v6/reference/projects) · [Run & deploy](https://docs.hud.ai/v6/reference/runtime)
246
248
 
247
249
  ## Train on rewards
248
250
 
@@ -37,10 +37,12 @@ hud set HUD_API_KEY=your-key-here
37
37
  # or: export HUD_API_KEY=your-key-here
38
38
  ```
39
39
 
40
- Then scaffold your first environment:
40
+ Then start from the coding environment, or choose another example environment:
41
41
 
42
42
  ```bash
43
43
  hud init my-env
44
+ hud init my-desktop-env --template cua
45
+ hud init my-custom-env --template blank
44
46
  ```
45
47
 
46
48
  ![Agent running on SheetBench](docs/src/images/trace_sheet.gif)
@@ -80,16 +82,6 @@ hud sync tasks my-taskset
80
82
  hud eval my-taskset --remote
81
83
  ```
82
84
 
83
- For local iteration, the same protocol works against a container on your laptop:
84
-
85
- ```bash
86
- docker build -f Dockerfile.hud -t my-env .
87
- docker run -d --name run1 -p 8765:8765 my-env
88
- hud task start fix_bug --url tcp://127.0.0.1:8765
89
- hud task grade fix_bug --url tcp://127.0.0.1:8765 --answer "..."
90
- docker rm -f run1
91
- ```
92
-
93
85
  → [Run & deploy](https://docs.hud.ai/v6/reference/runtime)
94
86
 
95
87
  ## Environments & templates
@@ -156,7 +148,17 @@ A **capability** is a connection the environment exposes; a **harness** attaches
156
148
 
157
149
  From the [platform UI](https://hud.ai) you can run batches, compare models on the same taskset, and inspect every trace.
158
150
 
159
- → [Run & deploy](https://docs.hud.ai/v6/reference/runtime)
151
+ A **project** holds the environments and tasksets a team creates and decides who can see them. Deploying without one uses your team's default project, so nothing here is required to get started. To put an environment somewhere else, pin the directory once and every later `hud deploy` and `hud sync tasks` follows it:
152
+
153
+ ```bash
154
+ hud project list # projects you can see
155
+ hud project use browser-evals # writes projectId to .hud/config.json
156
+ hud project # where does a deploy here land?
157
+ ```
158
+
159
+ `hud deploy` and `hud sync tasks` also take `--project <name-or-id>` as a one-command override; it does not change the directory configuration. Use `hud set HUD_DEFAULT_PROJECT=<name-or-id>` for a machine-wide fallback. Precedence is the flag, then the directory's `.hud/config.json`, then `HUD_DEFAULT_PROJECT`, then your team default. An environment or taskset that already exists stays where it is; naming a different project fails rather than moving it.
160
+
161
+ → [Projects](https://docs.hud.ai/v6/reference/projects) · [Run & deploy](https://docs.hud.ai/v6/reference/runtime)
160
162
 
161
163
  ## Train on rewards
162
164
 
@@ -403,6 +403,12 @@ class ClaudeAgent(ToolAgent[BetaMessageParam, ClaudeConfig]):
403
403
  if thinking_parts:
404
404
  result.reasoning = "\n".join(thinking_parts)
405
405
  result.finish_reason = response.stop_reason
406
+ if response.stop_reason == "refusal" and response.stop_details is not None:
407
+ explanation = response.stop_details.explanation
408
+ if explanation:
409
+ result.refusal = explanation
410
+ elif response.stop_details.category is not None:
411
+ result.refusal = f"This request was refused ({response.stop_details.category})."
406
412
  return result
407
413
 
408
414
  @staticmethod
@@ -10,6 +10,7 @@ Inspired by harbor-framework/harbor's ClaudeCode agent.
10
10
 
11
11
  from __future__ import annotations
12
12
 
13
+ import base64
13
14
  import json
14
15
  import logging
15
16
  import shlex
@@ -20,6 +21,7 @@ from typing import TYPE_CHECKING, Any, cast
20
21
  from hud.agents.base import Agent
21
22
  from hud.agents.types import AgentStep, ClaudeSDKConfig, Usage
22
23
  from hud.settings import settings
24
+ from hud.telemetry.context import get_current_trace_id, get_trace_headers
23
25
  from hud.types import Step
24
26
 
25
27
  if TYPE_CHECKING:
@@ -83,6 +85,7 @@ class ClaudeSDKAgent(Agent):
83
85
 
84
86
  async def __call__(self, run: Run) -> None:
85
87
  mcp_servers: dict[str, dict[str, Any]] = {}
88
+ trace_id = get_current_trace_id()
86
89
  manifest = run.client.manifest
87
90
  bindings = manifest.bindings if manifest is not None else []
88
91
  families = {c.protocol.split("/", 1)[0] for c in bindings}
@@ -100,8 +103,11 @@ class ClaudeSDKAgent(Agent):
100
103
  token = cap.params.get("auth_token")
101
104
  transport = "http" if cap.params["transport"] == "streamable-http" else "sse"
102
105
  server_config: dict[str, Any] = {"type": transport, "url": cap.url}
106
+ headers = {"Trace-Id": trace_id} if trace_id is not None else {}
103
107
  if token:
104
- server_config["headers"] = {"Authorization": f"Bearer {token}"}
108
+ headers["Authorization"] = f"Bearer {token}"
109
+ if headers:
110
+ server_config["headers"] = headers
105
111
  if cap.name in mcp_servers:
106
112
  raise RuntimeError(f"duplicate MCP server name {cap.name!r}")
107
113
  mcp_servers[cap.name] = server_config
@@ -188,6 +194,11 @@ class ClaudeSDKAgent(Agent):
188
194
  if settings.api_key:
189
195
  env["ANTHROPIC_BASE_URL"] = settings.hud_gateway_url
190
196
  env["ANTHROPIC_API_KEY"] = settings.api_key
197
+ trace_headers = get_trace_headers()
198
+ if trace_headers:
199
+ env["ANTHROPIC_CUSTOM_HEADERS"] = "\n".join(
200
+ f"{name}: {value}" for name, value in trace_headers.items()
201
+ )
191
202
  elif settings.anthropic_api_key:
192
203
  env["ANTHROPIC_API_KEY"] = settings.anthropic_api_key
193
204
 
@@ -259,16 +270,18 @@ class ClaudeSDKAgent(Agent):
259
270
  # Solution: use `cmd /c claude [args]` (no inline prompt) and feed the
260
271
  # prompt via stdin from .hud_prompt.txt. claude --print reads stdin as
261
272
  # the initial message when no -- argument is provided.
262
- set_parts = [f"set {k}={v}" for k, v in env_vars.items()]
263
273
  cmd_args = ["cmd", "/c", "claude"] + base_args[1:] # noqa: RUF005
264
274
  py_args_repr = "[" + ",".join(f"'{a}'" for a in cmd_args) + "]"
275
+ encoded_env = base64.b64encode(json.dumps(env_vars).encode()).decode()
265
276
  python_launcher = (
266
277
  'python -c "'
267
- "import subprocess,sys;"
268
- f"r=subprocess.run({py_args_repr},stdin=open('.hud_prompt.txt','rb'));"
278
+ "import base64,json,os,subprocess,sys;"
279
+ "env=os.environ.copy();"
280
+ f"env.update(json.loads(base64.b64decode('{encoded_env}')));"
281
+ f"r=subprocess.run({py_args_repr},stdin=open('.hud_prompt.txt','rb'),env=env);"
269
282
  'sys.exit(r.returncode)"'
270
283
  )
271
- return " && ".join([*set_parts, python_launcher])
284
+ return python_launcher
272
285
 
273
286
  # POSIX path: shell-quote everything and embed prompt inline.
274
287
  cli_parts = [shlex.quote(a) for a in base_args]
@@ -24,7 +24,7 @@ from openai.types.responses.response_input_param import (
24
24
  )
25
25
  from openai.types.shared_params.reasoning import Reasoning # noqa: TC002
26
26
 
27
- from hud.agents.tool_agent import RunState, ToolAgent
27
+ from hud.agents.tool_agent import DegenerateTurnError, RunState, ToolAgent
28
28
  from hud.agents.types import AgentStep, Citation, OpenAIConfig, Usage
29
29
  from hud.settings import settings
30
30
  from hud.types import MCPToolCall, MCPToolResult
@@ -41,6 +41,16 @@ if TYPE_CHECKING:
41
41
  logger = logging.getLogger(__name__)
42
42
 
43
43
 
44
+ class EmptyShellCallError(DegenerateTurnError):
45
+ """The model emitted a shell_call with an empty ``commands`` array.
46
+
47
+ OpenAI stores such a turn but returns 500 for every request that continues
48
+ from it via ``previous_response_id``. The turn is rejected before its
49
+ response id is committed, so the run state still points at the last
50
+ healthy response and the next sample continues from there.
51
+ """
52
+
53
+
44
54
  @dataclass
45
55
  class OpenAIRunState(RunState[ResponseInputItemParam]):
46
56
  last_response_id: str | None = None
@@ -223,10 +233,23 @@ class OpenAIAgent(ToolAgent[ResponseInputItemParam, OpenAIConfig]):
223
233
  previous_response_id=(
224
234
  oai_state.last_response_id if oai_state.last_response_id is not None else Omit()
225
235
  ),
236
+ prompt_cache_key=self.config.prompt_cache_key,
226
237
  truncation=self.truncation if self.truncation is not None else Omit(),
227
238
  include=include_param,
228
239
  )
229
240
 
241
+ empty_shell_calls = [
242
+ item.call_id
243
+ for item in response.output
244
+ if item.type == "shell_call" and item.action.to_dict().get("commands") == []
245
+ ]
246
+ if empty_shell_calls:
247
+ raise EmptyShellCallError(
248
+ f"Model emitted shell_call(s) with an empty commands array "
249
+ f"({', '.join(empty_shell_calls)}); continuing from response {response.id} "
250
+ "would fail with OpenAI 500s, so it was not committed to the run state."
251
+ )
252
+
230
253
  oai_state.last_response_id = response.id
231
254
  oai_state.message_cursor = len(messages)
232
255
 
@@ -335,4 +358,4 @@ class OpenAIAgent(ToolAgent[ResponseInputItemParam, OpenAIConfig]):
335
358
  )
336
359
 
337
360
 
338
- __all__ = ["OpenAIAgent"]
361
+ __all__ = ["EmptyShellCallError", "OpenAIAgent"]
@@ -105,6 +105,7 @@ def test_create_agent_value_shortcut_leaves_client_out_of_config(
105
105
  assert isinstance(agent, OpenAIAgent)
106
106
  assert agent.config.model_client is None
107
107
  assert agent.openai_client is sentinel
108
+ assert agent.hosted_spec()["config"]["prompt_cache_key"] == agent.config.prompt_cache_key
108
109
 
109
110
 
110
111
  def test_create_agent_resolves_gateway_model_metadata(
@@ -54,11 +54,12 @@ class FakeAnthropic:
54
54
  self.beta = SimpleNamespace(messages=FakeMessages(*outcomes))
55
55
 
56
56
 
57
- def _final(*content: Any, stop_reason: str) -> Any:
57
+ def _final(*content: Any, stop_reason: str, stop_details: Any = None) -> Any:
58
58
  """A fake ``BetaMessage``: content blocks plus the always-present envelope."""
59
59
  return SimpleNamespace(
60
60
  content=list(content),
61
61
  stop_reason=stop_reason,
62
+ stop_details=stop_details,
62
63
  model="claude-test-v9",
63
64
  usage=SimpleNamespace(input_tokens=11, output_tokens=7, cache_read_input_tokens=3),
64
65
  )
@@ -128,6 +129,48 @@ async def test_get_response_done_on_text_only() -> None:
128
129
  assert result.done is True
129
130
  assert result.content == "done"
130
131
  assert result.tool_calls == []
132
+ assert result.refusal is None
133
+
134
+
135
+ async def test_get_response_surfaces_refusal_explanation() -> None:
136
+ explanation = (
137
+ "This request triggered restrictions on violative cyber content and was "
138
+ "blocked under Anthropic's Usage Policy."
139
+ )
140
+ final = _final(
141
+ stop_reason="refusal",
142
+ stop_details=SimpleNamespace(
143
+ type="refusal",
144
+ category="cyber",
145
+ explanation=explanation,
146
+ ),
147
+ )
148
+ agent = _agent(final)
149
+ result = await agent.get_response(_state(agent))
150
+
151
+ assert result.finish_reason == "refusal"
152
+ assert result.refusal == explanation
153
+ assert result.done is True
154
+ assert result.tool_calls == []
155
+
156
+
157
+ async def test_get_response_surfaces_refusal_category_when_explanation_missing() -> None:
158
+ final = _final(
159
+ stop_reason="refusal",
160
+ stop_details=SimpleNamespace(
161
+ type="refusal",
162
+ category="cyber",
163
+ explanation=None,
164
+ ),
165
+ )
166
+ agent = _agent(final)
167
+ result = await agent.get_response(_state(agent))
168
+
169
+ assert result.finish_reason == "refusal"
170
+ assert result.refusal is not None
171
+ assert "cyber" in result.refusal
172
+ assert result.done is True
173
+ assert result.tool_calls == []
131
174
 
132
175
 
133
176
  async def test_get_response_collects_thinking() -> None:
@@ -26,6 +26,8 @@ from hud.agents.claude.sdk.agent import ClaudeSDKAgent, build_remote_invocation
26
26
  from hud.agents.types import ClaudeSDKConfig
27
27
  from hud.capabilities import Capability, SSHClient
28
28
  from hud.capabilities.rfb import WebPScreenshotEncoding
29
+ from hud.settings import settings
30
+ from hud.telemetry.context import set_trace_context
29
31
 
30
32
  if TYPE_CHECKING:
31
33
  from pathlib import Path
@@ -52,6 +54,20 @@ def test_posix_shell_runs_inline_with_install_check() -> None:
52
54
  assert inv.command.endswith(" && claude --print -- hi")
53
55
 
54
56
 
57
+ def test_gateway_trace_headers_are_forwarded_to_claude_cli(
58
+ monkeypatch: pytest.MonkeyPatch,
59
+ ) -> None:
60
+ monkeypatch.setattr(settings, "api_key", "test-key")
61
+ agent = ClaudeSDKAgent()
62
+
63
+ with set_trace_context("child-trace", parent_trace_id="parent-trace"):
64
+ env = agent._build_env_vars()
65
+
66
+ assert env["ANTHROPIC_CUSTOM_HEADERS"] == (
67
+ "Trace-Id: child-trace\nX-HUD-Parent-Trace-Id: parent-trace"
68
+ )
69
+
70
+
55
71
  # ─── _exec end-to-end over a fake SSH workspace ────────────────────────
56
72
 
57
73
 
@@ -246,17 +262,22 @@ async def test_manifest_mcp_capability_is_written_for_remote_claude(
246
262
  execute = AsyncMock()
247
263
  monkeypatch.setattr(agent, "_exec", execute)
248
264
 
249
- await agent(
250
- cast(
251
- "Any",
252
- SimpleNamespace(client=Client(), prompt_text="call the tool"),
265
+ with set_trace_context("child-trace"):
266
+ await agent(
267
+ cast(
268
+ "Any",
269
+ SimpleNamespace(client=Client(), prompt_text="call the tool"),
270
+ )
253
271
  )
254
- )
255
272
 
256
273
  await_args = execute.await_args
257
274
  assert await_args is not None
258
275
  assert await_args.kwargs["mcp_servers"] == {
259
- "database": {"type": claude_type, "url": "http://database:8000/mcp"}
276
+ "database": {
277
+ "type": claude_type,
278
+ "url": "http://database:8000/mcp",
279
+ "headers": {"Trace-Id": "child-trace"},
280
+ }
260
281
  }
261
282
  execute.assert_awaited_once()
262
283
 
@@ -8,9 +8,10 @@ from types import SimpleNamespace
8
8
  from typing import Any, cast
9
9
 
10
10
  import mcp.types as mcp_types
11
+ import pytest
11
12
  from openai.types.responses import ResponseOutputText
12
13
 
13
- from hud.agents.openai.agent import OpenAIAgent, OpenAIRunState
14
+ from hud.agents.openai.agent import EmptyShellCallError, OpenAIAgent, OpenAIRunState
14
15
  from hud.agents.openai.tools.base import format_openai_result
15
16
  from hud.agents.openai.tools.computer import OPENAI_COMPUTER_SPEC, OpenAIComputerTool
16
17
  from hud.agents.types import OpenAIConfig
@@ -139,6 +140,19 @@ async def test_get_response_done_when_no_tool_calls() -> None:
139
140
  assert result.finish_reason is None
140
141
 
141
142
 
143
+ async def test_get_response_reuses_prompt_cache_key() -> None:
144
+ agent = _agent(_api_response("resp_cache", []))
145
+ state = OpenAIRunState(messages=[agent._format_message("user", "first")])
146
+
147
+ await agent.get_response(state)
148
+ state.messages.append(agent._format_message("user", "second"))
149
+ await agent.get_response(state)
150
+
151
+ calls = cast("Any", agent.openai_client.responses).calls
152
+ assert calls[0]["prompt_cache_key"] == agent.config.prompt_cache_key
153
+ assert calls[1]["prompt_cache_key"] == agent.config.prompt_cache_key
154
+
155
+
142
156
  async def test_get_response_surfaces_token_cap_truncation() -> None:
143
157
  # Responses has no finish_reason; incomplete_details.reason carries the cap.
144
158
  response = _api_response(
@@ -181,3 +195,26 @@ async def test_get_response_parses_shell_call() -> None:
181
195
 
182
196
  result = await agent.get_response(state)
183
197
  assert [tc.name for tc in result.tool_calls] == ["shell"]
198
+
199
+
200
+ async def test_get_response_rejects_empty_shell_call() -> None:
201
+ # A shell_call with an empty commands array poisons the stored response
202
+ # chain (OpenAI 500s every continuation via previous_response_id), so the
203
+ # turn must fail fast and its response id must never be committed.
204
+ response = _api_response(
205
+ "resp_poisoned",
206
+ [
207
+ SimpleNamespace(
208
+ type="shell_call",
209
+ action=SimpleNamespace(to_dict=lambda: {"commands": [], "timeout_ms": 10000}),
210
+ call_id="call_empty",
211
+ ),
212
+ ],
213
+ )
214
+ agent = _agent(response)
215
+ state = OpenAIRunState(messages=[agent._format_message("user", "run")])
216
+
217
+ with pytest.raises(EmptyShellCallError, match="empty commands array"):
218
+ await agent.get_response(state)
219
+
220
+ assert state.last_response_id is None
@@ -7,6 +7,7 @@ scripted ``get_response`` so the loop, dispatch, and message formatting run offl
7
7
  from __future__ import annotations
8
8
 
9
9
  import asyncio
10
+ from contextlib import AsyncExitStack
10
11
  from types import SimpleNamespace
11
12
  from typing import TYPE_CHECKING, Any, Literal, cast
12
13
  from unittest.mock import AsyncMock, Mock
@@ -20,14 +21,22 @@ from hud.agents.claude.agent import ClaudeAgent
20
21
  from hud.agents.claude.tools.coding import ClaudeBashTool, ClaudeTextEditorTool
21
22
  from hud.agents.openai.tools.coding import OpenAIShellTool
22
23
  from hud.agents.openai.tools.mcp_proxy import OpenAIMCPProxyTool
23
- from hud.agents.tool_agent import RunState, ToolAgent
24
+ from hud.agents.tool_agent import DegenerateTurnError, RunState, ToolAgent
24
25
  from hud.agents.tools.base import AgentToolSpec, result_text
25
26
  from hud.agents.tools.rfb import RFBTool
26
27
  from hud.agents.tools.ssh import SSHInfrastructureErrorResult
27
28
  from hud.agents.types import AgentConfig, AgentStep, ClaudeConfig, ClaudeSDKConfig, ToolStep
28
- from hud.capabilities import Capability, CapabilityClient, MCPClient, RFBClient, SSHClient
29
+ from hud.capabilities import (
30
+ Capability,
31
+ CapabilityClient,
32
+ MCPClient,
33
+ RFBClient,
34
+ SSHClient,
35
+ )
36
+ from hud.capabilities.mcp import get_mcp_trace_id
29
37
  from hud.capabilities.rfb import PngScreenshotEncoding, WebPScreenshotEncoding
30
38
  from hud.capabilities.ssh import SSHConnectionError
39
+ from hud.telemetry.context import set_trace_context
31
40
  from hud.types import MCPToolCall, MCPToolResult, Step, Trace
32
41
 
33
42
  if TYPE_CHECKING:
@@ -253,6 +262,57 @@ async def test_mcp_client_uses_the_declared_http_transport(
253
262
  assert isinstance(transports[0], expected_type)
254
263
 
255
264
 
265
+ async def test_mcp_client_propagates_active_trace_context() -> None:
266
+ calls: list[dict[str, Any]] = []
267
+
268
+ class Client:
269
+ async def call_tool_mcp(self, **kwargs: Any) -> mcp_types.CallToolResult:
270
+ calls.append(kwargs)
271
+ return mcp_types.CallToolResult(content=[])
272
+
273
+ capability = Capability.mcp(url="https://tools.example/mcp")
274
+ client = MCPClient(capability, cast("Any", Client()), AsyncExitStack())
275
+
276
+ with set_trace_context("child-trace"):
277
+ await client.call_tool("verify", {})
278
+
279
+ assert calls == [
280
+ {
281
+ "name": "verify",
282
+ "arguments": {},
283
+ "meta": {"hud/trace-id": "child-trace"},
284
+ }
285
+ ]
286
+
287
+
288
+ def test_mcp_server_reads_propagated_trace_context(monkeypatch: pytest.MonkeyPatch) -> None:
289
+ request_context = SimpleNamespace(
290
+ meta=SimpleNamespace(model_extra={"hud/trace-id": "child-trace"})
291
+ )
292
+ monkeypatch.setattr(
293
+ "fastmcp.server.dependencies.get_context",
294
+ lambda: SimpleNamespace(request_context=request_context),
295
+ )
296
+
297
+ assert get_mcp_trace_id() == "child-trace"
298
+
299
+
300
+ def test_mcp_server_reads_propagated_trace_header(monkeypatch: pytest.MonkeyPatch) -> None:
301
+ request_context = SimpleNamespace(meta=SimpleNamespace(model_extra={}))
302
+ monkeypatch.setattr(
303
+ "fastmcp.server.dependencies.get_context",
304
+ lambda: SimpleNamespace(request_context=request_context),
305
+ )
306
+ monkeypatch.setattr(
307
+ "fastmcp.server.dependencies.get_http_headers",
308
+ lambda: {
309
+ "trace-id": "child-trace",
310
+ },
311
+ )
312
+
313
+ assert get_mcp_trace_id() == "child-trace"
314
+
315
+
256
316
  async def test_multiple_mcp_capabilities_qualify_tool_names() -> None:
257
317
  class Client(MCPClient):
258
318
  def __init__(self, *names: str) -> None:
@@ -532,6 +592,50 @@ async def test_loop_finishes_on_done_response() -> None:
532
592
  assert step.started_at is not None
533
593
 
534
594
 
595
+ class _DegenerateDictAgent(DictAgent):
596
+ """DictAgent whose scripted turns may be ``DegenerateTurnError`` instances."""
597
+
598
+ def __init__(self, turns: list[AgentStep | DegenerateTurnError], **config: Any) -> None:
599
+ super().__init__(cast("list[AgentStep]", turns), **config)
600
+
601
+ async def get_response(
602
+ self, state: RunState[_Msg], *, system_prompt: Any = None, citations_enabled: bool = False
603
+ ) -> AgentStep:
604
+ turn = await super().get_response(
605
+ state, system_prompt=system_prompt, citations_enabled=citations_enabled
606
+ )
607
+ if isinstance(turn, DegenerateTurnError):
608
+ raise turn
609
+ return turn
610
+
611
+
612
+ async def test_loop_discards_degenerate_turn_and_resamples() -> None:
613
+ agent = _DegenerateDictAgent(
614
+ [
615
+ DegenerateTurnError("empty shell_call"),
616
+ AgentStep(content="recovered", done=True),
617
+ ]
618
+ )
619
+ run = cast("Run", _FakeRun())
620
+
621
+ await agent._loop(run, RunState(), max_steps=3)
622
+
623
+ # The degenerate turn is discarded (consuming a step) and never recorded.
624
+ assert run.trace.status == "completed"
625
+ assert run.trace.content == "recovered"
626
+ assert [step.source for step in run.trace.steps] == ["agent"]
627
+
628
+
629
+ async def test_loop_fails_when_degenerate_turn_exhausts_steps() -> None:
630
+ agent = _DegenerateDictAgent([DegenerateTurnError("empty shell_call")] * 2)
631
+ run = cast("Run", _FakeRun())
632
+
633
+ await agent._loop(run, RunState(), max_steps=2)
634
+
635
+ assert run.trace.status == "error"
636
+ assert run.trace.error == "empty shell_call"
637
+
638
+
535
639
  async def test_loop_dispatches_tool_calls_then_finishes() -> None:
536
640
  agent = DictAgent(
537
641
  [
@@ -53,6 +53,16 @@ MessageT = TypeVar("MessageT")
53
53
  ConfigT = TypeVar("ConfigT", bound="AgentConfig")
54
54
 
55
55
 
56
+ class DegenerateTurnError(Exception):
57
+ """A model sample was unusable and was not committed to the run state.
58
+
59
+ Raised by ``get_response`` implementations that reject a degenerate
60
+ response before recording it. The loop discards the turn and samples a
61
+ fresh one; each discarded turn still consumes a step from ``max_steps``,
62
+ and a degenerate final step fails the run.
63
+ """
64
+
65
+
56
66
  def _message_text(message: mcp_types.PromptMessage) -> str:
57
67
  """Best-effort plain text for a prompt message (text content only for now)."""
58
68
  content = message.content
@@ -229,11 +239,17 @@ class ToolAgent(Agent, Generic[MessageT, ConfigT]):
229
239
  for turn in range(1, max_steps + 1):
230
240
  logger.info("step %d/%d", turn, max_steps)
231
241
  started_at = now_iso()
232
- step = await self.get_response(
233
- state,
234
- system_prompt=system_prompt,
235
- citations_enabled=citations_enabled,
236
- )
242
+ try:
243
+ step = await self.get_response(
244
+ state,
245
+ system_prompt=system_prompt,
246
+ citations_enabled=citations_enabled,
247
+ )
248
+ except DegenerateTurnError as exc:
249
+ if turn == max_steps:
250
+ raise
251
+ logger.warning("Discarded degenerate turn: %s", exc)
252
+ continue
237
253
  step.started_at = step.started_at or started_at
238
254
  step.model = step.model or self.config.model
239
255
  run.record(step)
@@ -17,6 +17,7 @@ schema understands this family's payload.
17
17
  from __future__ import annotations
18
18
 
19
19
  from typing import Any, ClassVar, Literal, cast
20
+ from uuid import uuid4
20
21
 
21
22
  from mcp.types import ContentBlock, ImageContent, TextContent
22
23
  from pydantic import (
@@ -121,6 +122,11 @@ class OpenAIConfig(AgentConfig):
121
122
  text: Any = None # {"verbosity": "low"|"medium"|"high"}
122
123
  truncation: Literal["auto", "disabled"] | None = None
123
124
  parallel_tool_calls: bool | None = None
125
+ prompt_cache_key: str = Field(
126
+ default_factory=lambda: uuid4().hex,
127
+ min_length=1,
128
+ max_length=64,
129
+ )
124
130
 
125
131
 
126
132
  class OpenAIChatConfig(AgentConfig):