praxis-agent 0.1.0
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.
- package/LICENSE +21 -0
- package/README.md +534 -0
- package/dist/application/background-agent-manager.d.ts +66 -0
- package/dist/application/background-agent-manager.d.ts.map +1 -0
- package/dist/application/background-agent-manager.js +362 -0
- package/dist/application/background-agent-manager.js.map +1 -0
- package/dist/application/background-bash-manager.d.ts +54 -0
- package/dist/application/background-bash-manager.d.ts.map +1 -0
- package/dist/application/background-bash-manager.js +354 -0
- package/dist/application/background-bash-manager.js.map +1 -0
- package/dist/application/background-task-id.d.ts +5 -0
- package/dist/application/background-task-id.d.ts.map +1 -0
- package/dist/application/background-task-id.js +25 -0
- package/dist/application/background-task-id.js.map +1 -0
- package/dist/application/claude-project-purge.d.ts +50 -0
- package/dist/application/claude-project-purge.d.ts.map +1 -0
- package/dist/application/claude-project-purge.js +459 -0
- package/dist/application/claude-project-purge.js.map +1 -0
- package/dist/application/managed-worktree.d.ts +15 -0
- package/dist/application/managed-worktree.d.ts.map +1 -0
- package/dist/application/managed-worktree.js +127 -0
- package/dist/application/managed-worktree.js.map +1 -0
- package/dist/application/model-compactor.d.ts +8 -0
- package/dist/application/model-compactor.d.ts.map +1 -0
- package/dist/application/model-compactor.js +70 -0
- package/dist/application/model-compactor.js.map +1 -0
- package/dist/application/scheduled-prompt-manager.d.ts +69 -0
- package/dist/application/scheduled-prompt-manager.d.ts.map +1 -0
- package/dist/application/scheduled-prompt-manager.js +387 -0
- package/dist/application/scheduled-prompt-manager.js.map +1 -0
- package/dist/application/session-service.d.ts +142 -0
- package/dist/application/session-service.d.ts.map +1 -0
- package/dist/application/session-service.js +1690 -0
- package/dist/application/session-service.js.map +1 -0
- package/dist/application/session-worktree.d.ts +55 -0
- package/dist/application/session-worktree.d.ts.map +1 -0
- package/dist/application/session-worktree.js +311 -0
- package/dist/application/session-worktree.js.map +1 -0
- package/dist/application/subagent-service.d.ts +98 -0
- package/dist/application/subagent-service.d.ts.map +1 -0
- package/dist/application/subagent-service.js +1305 -0
- package/dist/application/subagent-service.js.map +1 -0
- package/dist/application/tool-use-summary.d.ts +8 -0
- package/dist/application/tool-use-summary.d.ts.map +1 -0
- package/dist/application/tool-use-summary.js +53 -0
- package/dist/application/tool-use-summary.js.map +1 -0
- package/dist/application/top-level-agent-manager.d.ts +58 -0
- package/dist/application/top-level-agent-manager.d.ts.map +1 -0
- package/dist/application/top-level-agent-manager.js +907 -0
- package/dist/application/top-level-agent-manager.js.map +1 -0
- package/dist/application/workflow-manager.d.ts +61 -0
- package/dist/application/workflow-manager.d.ts.map +1 -0
- package/dist/application/workflow-manager.js +582 -0
- package/dist/application/workflow-manager.js.map +1 -0
- package/dist/application/workflow-meta.d.ts +18 -0
- package/dist/application/workflow-meta.d.ts.map +1 -0
- package/dist/application/workflow-meta.js +146 -0
- package/dist/application/workflow-meta.js.map +1 -0
- package/dist/application/workflow-runtime.d.ts +29 -0
- package/dist/application/workflow-runtime.d.ts.map +1 -0
- package/dist/application/workflow-runtime.js +176 -0
- package/dist/application/workflow-runtime.js.map +1 -0
- package/dist/application/workflow-worktree.d.ts +9 -0
- package/dist/application/workflow-worktree.d.ts.map +1 -0
- package/dist/application/workflow-worktree.js +11 -0
- package/dist/application/workflow-worktree.js.map +1 -0
- package/dist/cli/agents-dashboard.d.ts +37 -0
- package/dist/cli/agents-dashboard.d.ts.map +1 -0
- package/dist/cli/agents-dashboard.js +408 -0
- package/dist/cli/agents-dashboard.js.map +1 -0
- package/dist/cli/controls.d.ts +15 -0
- package/dist/cli/controls.d.ts.map +1 -0
- package/dist/cli/controls.js +219 -0
- package/dist/cli/controls.js.map +1 -0
- package/dist/cli/debug.d.ts +13 -0
- package/dist/cli/debug.d.ts.map +1 -0
- package/dist/cli/debug.js +37 -0
- package/dist/cli/debug.js.map +1 -0
- package/dist/cli/interactive.d.ts +63 -0
- package/dist/cli/interactive.d.ts.map +1 -0
- package/dist/cli/interactive.js +600 -0
- package/dist/cli/interactive.js.map +1 -0
- package/dist/cli/protocol.d.ts +251 -0
- package/dist/cli/protocol.d.ts.map +1 -0
- package/dist/cli/protocol.js +2146 -0
- package/dist/cli/protocol.js.map +1 -0
- package/dist/cli.d.ts +116 -0
- package/dist/cli.d.ts.map +1 -0
- package/dist/cli.js +3901 -0
- package/dist/cli.js.map +1 -0
- package/dist/compatibility/claude/compaction.d.ts +19 -0
- package/dist/compatibility/claude/compaction.d.ts.map +1 -0
- package/dist/compatibility/claude/compaction.js +82 -0
- package/dist/compatibility/claude/compaction.js.map +1 -0
- package/dist/compatibility/claude/context.d.ts +24 -0
- package/dist/compatibility/claude/context.d.ts.map +1 -0
- package/dist/compatibility/claude/context.js +92 -0
- package/dist/compatibility/claude/context.js.map +1 -0
- package/dist/compatibility/claude/dynamic-context.d.ts +20 -0
- package/dist/compatibility/claude/dynamic-context.d.ts.map +1 -0
- package/dist/compatibility/claude/dynamic-context.js +168 -0
- package/dist/compatibility/claude/dynamic-context.js.map +1 -0
- package/dist/compatibility/claude/file-history.d.ts +14 -0
- package/dist/compatibility/claude/file-history.d.ts.map +1 -0
- package/dist/compatibility/claude/file-history.js +340 -0
- package/dist/compatibility/claude/file-history.js.map +1 -0
- package/dist/compatibility/claude/file-resources.d.ts +28 -0
- package/dist/compatibility/claude/file-resources.d.ts.map +1 -0
- package/dist/compatibility/claude/file-resources.js +168 -0
- package/dist/compatibility/claude/file-resources.js.map +1 -0
- package/dist/compatibility/claude/fork.d.ts +9 -0
- package/dist/compatibility/claude/fork.d.ts.map +1 -0
- package/dist/compatibility/claude/fork.js +225 -0
- package/dist/compatibility/claude/fork.js.map +1 -0
- package/dist/compatibility/claude/history.d.ts +4 -0
- package/dist/compatibility/claude/history.d.ts.map +1 -0
- package/dist/compatibility/claude/history.js +86 -0
- package/dist/compatibility/claude/history.js.map +1 -0
- package/dist/compatibility/claude/ownership.d.ts +87 -0
- package/dist/compatibility/claude/ownership.d.ts.map +1 -0
- package/dist/compatibility/claude/ownership.js +100 -0
- package/dist/compatibility/claude/ownership.js.map +1 -0
- package/dist/compatibility/claude/paths.d.ts +17 -0
- package/dist/compatibility/claude/paths.d.ts.map +1 -0
- package/dist/compatibility/claude/paths.js +54 -0
- package/dist/compatibility/claude/paths.js.map +1 -0
- package/dist/compatibility/claude/pr-links.d.ts +25 -0
- package/dist/compatibility/claude/pr-links.d.ts.map +1 -0
- package/dist/compatibility/claude/pr-links.js +80 -0
- package/dist/compatibility/claude/pr-links.js.map +1 -0
- package/dist/compatibility/claude/projection.d.ts +11 -0
- package/dist/compatibility/claude/projection.d.ts.map +1 -0
- package/dist/compatibility/claude/projection.js +325 -0
- package/dist/compatibility/claude/projection.js.map +1 -0
- package/dist/compatibility/claude/schema.d.ts +18 -0
- package/dist/compatibility/claude/schema.d.ts.map +1 -0
- package/dist/compatibility/claude/schema.js +983 -0
- package/dist/compatibility/claude/schema.js.map +1 -0
- package/dist/compatibility/claude/shared-resources.d.ts +51 -0
- package/dist/compatibility/claude/shared-resources.d.ts.map +1 -0
- package/dist/compatibility/claude/shared-resources.js +382 -0
- package/dist/compatibility/claude/shared-resources.js.map +1 -0
- package/dist/compatibility/claude/sidechain.d.ts +50 -0
- package/dist/compatibility/claude/sidechain.d.ts.map +1 -0
- package/dist/compatibility/claude/sidechain.js +90 -0
- package/dist/compatibility/claude/sidechain.js.map +1 -0
- package/dist/compatibility/claude/tool-links.d.ts +13 -0
- package/dist/compatibility/claude/tool-links.d.ts.map +1 -0
- package/dist/compatibility/claude/tool-links.js +61 -0
- package/dist/compatibility/claude/tool-links.js.map +1 -0
- package/dist/compatibility/claude/translation.d.ts +71 -0
- package/dist/compatibility/claude/translation.d.ts.map +1 -0
- package/dist/compatibility/claude/translation.js +443 -0
- package/dist/compatibility/claude/translation.js.map +1 -0
- package/dist/compatibility/claude/workflow-replay.d.ts +10 -0
- package/dist/compatibility/claude/workflow-replay.d.ts.map +1 -0
- package/dist/compatibility/claude/workflow-replay.js +66 -0
- package/dist/compatibility/claude/workflow-replay.js.map +1 -0
- package/dist/compatibility/claude/workflow.d.ts +33 -0
- package/dist/compatibility/claude/workflow.d.ts.map +1 -0
- package/dist/compatibility/claude/workflow.js +76 -0
- package/dist/compatibility/claude/workflow.js.map +1 -0
- package/dist/core/compaction.d.ts +16 -0
- package/dist/core/compaction.d.ts.map +1 -0
- package/dist/core/compaction.js +2 -0
- package/dist/core/compaction.js.map +1 -0
- package/dist/core/context-budget.d.ts +28 -0
- package/dist/core/context-budget.d.ts.map +1 -0
- package/dist/core/context-budget.js +108 -0
- package/dist/core/context-budget.js.map +1 -0
- package/dist/core/context.d.ts +16 -0
- package/dist/core/context.d.ts.map +1 -0
- package/dist/core/context.js +13 -0
- package/dist/core/context.js.map +1 -0
- package/dist/core/runtime.d.ts +394 -0
- package/dist/core/runtime.d.ts.map +1 -0
- package/dist/core/runtime.js +507 -0
- package/dist/core/runtime.js.map +1 -0
- package/dist/core/usage.d.ts +26 -0
- package/dist/core/usage.d.ts.map +1 -0
- package/dist/core/usage.js +133 -0
- package/dist/core/usage.js.map +1 -0
- package/dist/extensions/claude-extension-tools.d.ts +16 -0
- package/dist/extensions/claude-extension-tools.d.ts.map +1 -0
- package/dist/extensions/claude-extension-tools.js +78 -0
- package/dist/extensions/claude-extension-tools.js.map +1 -0
- package/dist/extensions/claude-extensions.d.ts +41 -0
- package/dist/extensions/claude-extensions.d.ts.map +1 -0
- package/dist/extensions/claude-extensions.js +270 -0
- package/dist/extensions/claude-extensions.js.map +1 -0
- package/dist/hooks/claude-hook-tools.d.ts +26 -0
- package/dist/hooks/claude-hook-tools.d.ts.map +1 -0
- package/dist/hooks/claude-hook-tools.js +129 -0
- package/dist/hooks/claude-hook-tools.js.map +1 -0
- package/dist/hooks/claude-hooks.d.ts +87 -0
- package/dist/hooks/claude-hooks.d.ts.map +1 -0
- package/dist/hooks/claude-hooks.js +411 -0
- package/dist/hooks/claude-hooks.js.map +1 -0
- package/dist/maintenance/doctor.d.ts +33 -0
- package/dist/maintenance/doctor.d.ts.map +1 -0
- package/dist/maintenance/doctor.js +391 -0
- package/dist/maintenance/doctor.js.map +1 -0
- package/dist/maintenance/self-update.d.ts +31 -0
- package/dist/maintenance/self-update.d.ts.map +1 -0
- package/dist/maintenance/self-update.js +81 -0
- package/dist/maintenance/self-update.js.map +1 -0
- package/dist/mcp/claude-mcp-management.d.ts +28 -0
- package/dist/mcp/claude-mcp-management.d.ts.map +1 -0
- package/dist/mcp/claude-mcp-management.js +169 -0
- package/dist/mcp/claude-mcp-management.js.map +1 -0
- package/dist/mcp/claude-mcp-oauth.d.ts +113 -0
- package/dist/mcp/claude-mcp-oauth.d.ts.map +1 -0
- package/dist/mcp/claude-mcp-oauth.js +1053 -0
- package/dist/mcp/claude-mcp-oauth.js.map +1 -0
- package/dist/mcp/claude-mcp-tools.d.ts +98 -0
- package/dist/mcp/claude-mcp-tools.d.ts.map +1 -0
- package/dist/mcp/claude-mcp-tools.js +1246 -0
- package/dist/mcp/claude-mcp-tools.js.map +1 -0
- package/dist/mcp/praxis-mcp-server.d.ts +35 -0
- package/dist/mcp/praxis-mcp-server.d.ts.map +1 -0
- package/dist/mcp/praxis-mcp-server.js +520 -0
- package/dist/mcp/praxis-mcp-server.js.map +1 -0
- package/dist/permissions/claude-auto-classifier.d.ts +20 -0
- package/dist/permissions/claude-auto-classifier.d.ts.map +1 -0
- package/dist/permissions/claude-auto-classifier.js +140 -0
- package/dist/permissions/claude-auto-classifier.js.map +1 -0
- package/dist/permissions/claude-permission-resolver.d.ts +30 -0
- package/dist/permissions/claude-permission-resolver.d.ts.map +1 -0
- package/dist/permissions/claude-permission-resolver.js +334 -0
- package/dist/permissions/claude-permission-resolver.js.map +1 -0
- package/dist/persistence/claude-job-store.d.ts +68 -0
- package/dist/persistence/claude-job-store.d.ts.map +1 -0
- package/dist/persistence/claude-job-store.js +311 -0
- package/dist/persistence/claude-job-store.js.map +1 -0
- package/dist/persistence/claude-scheduled-task-store.d.ts +29 -0
- package/dist/persistence/claude-scheduled-task-store.d.ts.map +1 -0
- package/dist/persistence/claude-scheduled-task-store.js +163 -0
- package/dist/persistence/claude-scheduled-task-store.js.map +1 -0
- package/dist/persistence/claude-sidechain-store.d.ts +18 -0
- package/dist/persistence/claude-sidechain-store.d.ts.map +1 -0
- package/dist/persistence/claude-sidechain-store.js +107 -0
- package/dist/persistence/claude-sidechain-store.js.map +1 -0
- package/dist/persistence/claude-task-store.d.ts +72 -0
- package/dist/persistence/claude-task-store.d.ts.map +1 -0
- package/dist/persistence/claude-task-store.js +440 -0
- package/dist/persistence/claude-task-store.js.map +1 -0
- package/dist/persistence/claude-transcript-store.d.ts +85 -0
- package/dist/persistence/claude-transcript-store.d.ts.map +1 -0
- package/dist/persistence/claude-transcript-store.js +385 -0
- package/dist/persistence/claude-transcript-store.js.map +1 -0
- package/dist/persistence/claude-workflow-store.d.ts +41 -0
- package/dist/persistence/claude-workflow-store.d.ts.map +1 -0
- package/dist/persistence/claude-workflow-store.js +263 -0
- package/dist/persistence/claude-workflow-store.js.map +1 -0
- package/dist/persistence/in-memory-sidechain-store.d.ts +12 -0
- package/dist/persistence/in-memory-sidechain-store.d.ts.map +1 -0
- package/dist/persistence/in-memory-sidechain-store.js +30 -0
- package/dist/persistence/in-memory-sidechain-store.js.map +1 -0
- package/dist/persistence/in-memory-transcript-store.d.ts +15 -0
- package/dist/persistence/in-memory-transcript-store.d.ts.map +1 -0
- package/dist/persistence/in-memory-transcript-store.js +95 -0
- package/dist/persistence/in-memory-transcript-store.js.map +1 -0
- package/dist/platform/atomic-write.d.ts +5 -0
- package/dist/platform/atomic-write.d.ts.map +1 -0
- package/dist/platform/atomic-write.js +25 -0
- package/dist/platform/atomic-write.js.map +1 -0
- package/dist/platform/bounded-process-runner.d.ts +28 -0
- package/dist/platform/bounded-process-runner.d.ts.map +1 -0
- package/dist/platform/bounded-process-runner.js +209 -0
- package/dist/platform/bounded-process-runner.js.map +1 -0
- package/dist/platform/claude-version.d.ts +5 -0
- package/dist/platform/claude-version.d.ts.map +1 -0
- package/dist/platform/claude-version.js +17 -0
- package/dist/platform/claude-version.js.map +1 -0
- package/dist/platform/command-shell.d.ts +3 -0
- package/dist/platform/command-shell.d.ts.map +1 -0
- package/dist/platform/command-shell.js +15 -0
- package/dist/platform/command-shell.js.map +1 -0
- package/dist/platform/exclusive-file-lease.d.ts +15 -0
- package/dist/platform/exclusive-file-lease.d.ts.map +1 -0
- package/dist/platform/exclusive-file-lease.js +265 -0
- package/dist/platform/exclusive-file-lease.js.map +1 -0
- package/dist/platform/sensitive-data.d.ts +8 -0
- package/dist/platform/sensitive-data.d.ts.map +1 -0
- package/dist/platform/sensitive-data.js +138 -0
- package/dist/platform/sensitive-data.js.map +1 -0
- package/dist/platform/tmux-worktree.d.ts +30 -0
- package/dist/platform/tmux-worktree.d.ts.map +1 -0
- package/dist/platform/tmux-worktree.js +138 -0
- package/dist/platform/tmux-worktree.js.map +1 -0
- package/dist/plugins/claude-plugin-eval-discovery.d.ts +20 -0
- package/dist/plugins/claude-plugin-eval-discovery.d.ts.map +1 -0
- package/dist/plugins/claude-plugin-eval-discovery.js +128 -0
- package/dist/plugins/claude-plugin-eval-discovery.js.map +1 -0
- package/dist/plugins/claude-plugin-eval-graders.d.ts +49 -0
- package/dist/plugins/claude-plugin-eval-graders.d.ts.map +1 -0
- package/dist/plugins/claude-plugin-eval-graders.js +137 -0
- package/dist/plugins/claude-plugin-eval-graders.js.map +1 -0
- package/dist/plugins/claude-plugin-eval-runner.d.ts +50 -0
- package/dist/plugins/claude-plugin-eval-runner.d.ts.map +1 -0
- package/dist/plugins/claude-plugin-eval-runner.js +189 -0
- package/dist/plugins/claude-plugin-eval-runner.js.map +1 -0
- package/dist/plugins/claude-plugin-eval-schema.d.ts +80 -0
- package/dist/plugins/claude-plugin-eval-schema.d.ts.map +1 -0
- package/dist/plugins/claude-plugin-eval-schema.js +436 -0
- package/dist/plugins/claude-plugin-eval-schema.js.map +1 -0
- package/dist/plugins/claude-plugin-eval.d.ts +66 -0
- package/dist/plugins/claude-plugin-eval.d.ts.map +1 -0
- package/dist/plugins/claude-plugin-eval.js +438 -0
- package/dist/plugins/claude-plugin-eval.js.map +1 -0
- package/dist/plugins/claude-plugin-maintenance.d.ts +39 -0
- package/dist/plugins/claude-plugin-maintenance.d.ts.map +1 -0
- package/dist/plugins/claude-plugin-maintenance.js +336 -0
- package/dist/plugins/claude-plugin-maintenance.js.map +1 -0
- package/dist/plugins/claude-plugin-marketplace.d.ts +97 -0
- package/dist/plugins/claude-plugin-marketplace.d.ts.map +1 -0
- package/dist/plugins/claude-plugin-marketplace.js +1433 -0
- package/dist/plugins/claude-plugin-marketplace.js.map +1 -0
- package/dist/plugins/claude-plugin-mcpb.d.ts +71 -0
- package/dist/plugins/claude-plugin-mcpb.d.ts.map +1 -0
- package/dist/plugins/claude-plugin-mcpb.js +885 -0
- package/dist/plugins/claude-plugin-mcpb.js.map +1 -0
- package/dist/plugins/claude-plugin-runtime.d.ts +120 -0
- package/dist/plugins/claude-plugin-runtime.d.ts.map +1 -0
- package/dist/plugins/claude-plugin-runtime.js +1263 -0
- package/dist/plugins/claude-plugin-runtime.js.map +1 -0
- package/dist/providers/anthropic-compatible.d.ts +35 -0
- package/dist/providers/anthropic-compatible.d.ts.map +1 -0
- package/dist/providers/anthropic-compatible.js +733 -0
- package/dist/providers/anthropic-compatible.js.map +1 -0
- package/dist/providers/environment.d.ts +14 -0
- package/dist/providers/environment.d.ts.map +1 -0
- package/dist/providers/environment.js +67 -0
- package/dist/providers/environment.js.map +1 -0
- package/dist/providers/fallback-provider.d.ts +17 -0
- package/dist/providers/fallback-provider.d.ts.map +1 -0
- package/dist/providers/fallback-provider.js +70 -0
- package/dist/providers/fallback-provider.js.map +1 -0
- package/dist/providers/openai-compatible.d.ts +30 -0
- package/dist/providers/openai-compatible.d.ts.map +1 -0
- package/dist/providers/openai-compatible.js +409 -0
- package/dist/providers/openai-compatible.js.map +1 -0
- package/dist/tools/claude-interactive-tools.d.ts +58 -0
- package/dist/tools/claude-interactive-tools.d.ts.map +1 -0
- package/dist/tools/claude-interactive-tools.js +390 -0
- package/dist/tools/claude-interactive-tools.js.map +1 -0
- package/dist/tools/claude-lsp-formatters.d.ts +2 -0
- package/dist/tools/claude-lsp-formatters.d.ts.map +1 -0
- package/dist/tools/claude-lsp-formatters.js +291 -0
- package/dist/tools/claude-lsp-formatters.js.map +1 -0
- package/dist/tools/claude-lsp-tool.d.ts +22 -0
- package/dist/tools/claude-lsp-tool.d.ts.map +1 -0
- package/dist/tools/claude-lsp-tool.js +744 -0
- package/dist/tools/claude-lsp-tool.js.map +1 -0
- package/dist/tools/claude-scheduled-tools.d.ts +20 -0
- package/dist/tools/claude-scheduled-tools.d.ts.map +1 -0
- package/dist/tools/claude-scheduled-tools.js +344 -0
- package/dist/tools/claude-scheduled-tools.js.map +1 -0
- package/dist/tools/claude-task-tools.d.ts +30 -0
- package/dist/tools/claude-task-tools.d.ts.map +1 -0
- package/dist/tools/claude-task-tools.js +503 -0
- package/dist/tools/claude-task-tools.js.map +1 -0
- package/dist/tools/claude-user-message.d.ts +16 -0
- package/dist/tools/claude-user-message.d.ts.map +1 -0
- package/dist/tools/claude-user-message.js +98 -0
- package/dist/tools/claude-user-message.js.map +1 -0
- package/dist/tools/claude-workflow-tools.d.ts +27 -0
- package/dist/tools/claude-workflow-tools.d.ts.map +1 -0
- package/dist/tools/claude-workflow-tools.js +257 -0
- package/dist/tools/claude-workflow-tools.js.map +1 -0
- package/dist/tools/claude-worktree-tools.d.ts +17 -0
- package/dist/tools/claude-worktree-tools.d.ts.map +1 -0
- package/dist/tools/claude-worktree-tools.js +135 -0
- package/dist/tools/claude-worktree-tools.js.map +1 -0
- package/dist/tools/filtered-tool-registry.d.ts +16 -0
- package/dist/tools/filtered-tool-registry.d.ts.map +1 -0
- package/dist/tools/filtered-tool-registry.js +51 -0
- package/dist/tools/filtered-tool-registry.js.map +1 -0
- package/dist/tools/glob.d.ts +9 -0
- package/dist/tools/glob.d.ts.map +1 -0
- package/dist/tools/glob.js +88 -0
- package/dist/tools/glob.js.map +1 -0
- package/dist/tools/local-tools.d.ts +47 -0
- package/dist/tools/local-tools.d.ts.map +1 -0
- package/dist/tools/local-tools.js +1082 -0
- package/dist/tools/local-tools.js.map +1 -0
- package/dist/tools/notebook.d.ts +22 -0
- package/dist/tools/notebook.d.ts.map +1 -0
- package/dist/tools/notebook.js +154 -0
- package/dist/tools/notebook.js.map +1 -0
- package/dist/tools/web.d.ts +48 -0
- package/dist/tools/web.d.ts.map +1 -0
- package/dist/tools/web.js +726 -0
- package/dist/tools/web.js.map +1 -0
- package/package.json +145 -0
|
@@ -0,0 +1,137 @@
|
|
|
1
|
+
import { readFile } from 'node:fs/promises';
|
|
2
|
+
import { minimatch } from 'minimatch';
|
|
3
|
+
import { resolveContainedPath } from './claude-plugin-eval-schema.js';
|
|
4
|
+
function subset(actual, expected) {
|
|
5
|
+
if (expected && typeof expected === 'object' && !Array.isArray(expected)) {
|
|
6
|
+
if (!actual || typeof actual !== 'object' || Array.isArray(actual))
|
|
7
|
+
return false;
|
|
8
|
+
return Object.entries(expected).every(([key, value]) => subset(actual[key], value));
|
|
9
|
+
}
|
|
10
|
+
return Object.is(actual, expected);
|
|
11
|
+
}
|
|
12
|
+
function matches(event, match) {
|
|
13
|
+
const selected = typeof match === 'string' ? { tool: match } : match;
|
|
14
|
+
return (event.type === 'tool-call' &&
|
|
15
|
+
event.tool === selected.tool &&
|
|
16
|
+
(!selected.input_match || subset(event.input, selected.input_match)));
|
|
17
|
+
}
|
|
18
|
+
async function focusText(focus, artifacts) {
|
|
19
|
+
if (focus === 'last_message')
|
|
20
|
+
return artifacts.lastMessage;
|
|
21
|
+
if (focus === 'trace')
|
|
22
|
+
return artifacts.trace.map((item) => JSON.stringify(item)).join('\n');
|
|
23
|
+
if (focus === 'files') {
|
|
24
|
+
const { glob } = await import('node:fs/promises');
|
|
25
|
+
const names = [];
|
|
26
|
+
for await (const path of glob('**/*', {
|
|
27
|
+
cwd: artifacts.cwd,
|
|
28
|
+
exclude: ['node_modules/**', '.git/**'],
|
|
29
|
+
}))
|
|
30
|
+
names.push(path);
|
|
31
|
+
return names.sort().join('\n');
|
|
32
|
+
}
|
|
33
|
+
return readFile(await resolveContainedPath(artifacts.cwd, focus.path, 'Grader focus'), 'utf8');
|
|
34
|
+
}
|
|
35
|
+
async function freeGrade(grader, artifacts) {
|
|
36
|
+
let passed = false;
|
|
37
|
+
let evidence = '';
|
|
38
|
+
if (grader.type === 'regex') {
|
|
39
|
+
const target = await focusText(grader.target, artifacts);
|
|
40
|
+
const expression = new RegExp(grader.pattern, grader.flags.includes('g') ? grader.flags : `${grader.flags}g`);
|
|
41
|
+
const count = [...target.matchAll(expression)].length;
|
|
42
|
+
passed =
|
|
43
|
+
grader.match === 'contains'
|
|
44
|
+
? count > 0
|
|
45
|
+
: grader.match === 'not_contains'
|
|
46
|
+
? count === 0
|
|
47
|
+
: count === Number(grader.match.slice(6));
|
|
48
|
+
evidence = `matches=${count}`;
|
|
49
|
+
}
|
|
50
|
+
else if (grader.type === 'tool_used') {
|
|
51
|
+
const count = artifacts.trace.filter((event) => matches(event, {
|
|
52
|
+
tool: grader.tool,
|
|
53
|
+
...(grader.input_match ? { input_match: grader.input_match } : {}),
|
|
54
|
+
})).length;
|
|
55
|
+
passed = count >= grader.min && count <= grader.max;
|
|
56
|
+
evidence = `uses=${count}`;
|
|
57
|
+
}
|
|
58
|
+
else if (grader.type === 'tool_order') {
|
|
59
|
+
const before = artifacts.trace.findIndex((event) => matches(event, grader.before));
|
|
60
|
+
const after = artifacts.trace.findIndex((event, index) => index > before && matches(event, grader.after));
|
|
61
|
+
passed = before >= 0 && after > before;
|
|
62
|
+
evidence = `before=${before},after=${after}`;
|
|
63
|
+
}
|
|
64
|
+
else {
|
|
65
|
+
const { glob } = await import('node:fs/promises');
|
|
66
|
+
let count = 0;
|
|
67
|
+
for await (const path of glob('**/*', { cwd: artifacts.cwd })) {
|
|
68
|
+
if (!minimatch(path, grader.path))
|
|
69
|
+
continue;
|
|
70
|
+
await resolveContainedPath(artifacts.cwd, path, 'file_exists match');
|
|
71
|
+
count += 1;
|
|
72
|
+
}
|
|
73
|
+
passed = grader.exists ? count > 0 : count === 0;
|
|
74
|
+
evidence = `files=${count}`;
|
|
75
|
+
}
|
|
76
|
+
return {
|
|
77
|
+
name: grader.name,
|
|
78
|
+
passed,
|
|
79
|
+
weight: grader.weight,
|
|
80
|
+
explanation: passed ? 'passed' : 'failed',
|
|
81
|
+
evidence,
|
|
82
|
+
...(grader.arm === 'with-only' ? { with_only: true } : {}),
|
|
83
|
+
};
|
|
84
|
+
}
|
|
85
|
+
export async function gradeClaudePluginEvalRun(options) {
|
|
86
|
+
const results = [];
|
|
87
|
+
let skippedPaidGraders = false;
|
|
88
|
+
let judgeCostUsd = 0;
|
|
89
|
+
for (const item of options.case.graders) {
|
|
90
|
+
if (options.arm === 'without' && item.arm === 'with-only')
|
|
91
|
+
continue;
|
|
92
|
+
if (item.type !== 'llm' && item.type !== 'baseline') {
|
|
93
|
+
results.push(await freeGrade(item, options.artifacts));
|
|
94
|
+
continue;
|
|
95
|
+
}
|
|
96
|
+
if (options.skipPaid) {
|
|
97
|
+
skippedPaidGraders = true;
|
|
98
|
+
continue;
|
|
99
|
+
}
|
|
100
|
+
if (!options.judge)
|
|
101
|
+
throw new Error(`Paid grader ${item.name} requires an eval judge`);
|
|
102
|
+
const judge = options.judge;
|
|
103
|
+
const focus = item.type === 'llm'
|
|
104
|
+
? await focusText(item.focus, options.artifacts)
|
|
105
|
+
: options.artifacts.lastMessage;
|
|
106
|
+
const baseline = item.type === 'baseline'
|
|
107
|
+
? await readFile(await resolveContainedPath(options.case.dir, item.baseline_file, 'Baseline file'), 'utf8')
|
|
108
|
+
: undefined;
|
|
109
|
+
const votes = await Promise.all(Array.from({ length: 3 }, () => judge.vote({
|
|
110
|
+
criteria: item.criteria,
|
|
111
|
+
focus,
|
|
112
|
+
...(baseline === undefined ? {} : { baseline }),
|
|
113
|
+
model: options.judgeModel,
|
|
114
|
+
...(options.signal ? { signal: options.signal } : {}),
|
|
115
|
+
})));
|
|
116
|
+
const passed = votes.filter((vote) => vote.passed).length >= 2;
|
|
117
|
+
judgeCostUsd += votes.reduce((sum, vote) => sum + vote.costUsd, 0);
|
|
118
|
+
results.push({
|
|
119
|
+
name: item.name,
|
|
120
|
+
passed,
|
|
121
|
+
weight: item.weight,
|
|
122
|
+
explanation: votes.find((vote) => vote.explanation)?.explanation ??
|
|
123
|
+
(passed ? 'passed' : 'failed'),
|
|
124
|
+
judge_votes: votes.map((vote) => vote.passed),
|
|
125
|
+
...(item.arm === 'with-only' ? { with_only: true } : {}),
|
|
126
|
+
});
|
|
127
|
+
}
|
|
128
|
+
const scored = results;
|
|
129
|
+
const total = scored.reduce((sum, item) => sum + item.weight, 0);
|
|
130
|
+
const score = total === 0
|
|
131
|
+
? 0
|
|
132
|
+
: scored
|
|
133
|
+
.filter((item) => item.passed)
|
|
134
|
+
.reduce((sum, item) => sum + item.weight, 0) / total;
|
|
135
|
+
return { score, graders: results, skippedPaidGraders, judgeCostUsd };
|
|
136
|
+
}
|
|
137
|
+
//# sourceMappingURL=claude-plugin-eval-graders.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"claude-plugin-eval-graders.js","sourceRoot":"","sources":["../../src/plugins/claude-plugin-eval-graders.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAA;AAE3C,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAA;AAQrC,OAAO,EAAE,oBAAoB,EAAE,MAAM,gCAAgC,CAAA;AAgCrE,SAAS,MAAM,CAAC,MAAe,EAAE,QAAiB;IAChD,IAAI,QAAQ,IAAI,OAAO,QAAQ,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;QACzE,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;YAChE,OAAO,KAAK,CAAA;QACd,OAAO,MAAM,CAAC,OAAO,CAAC,QAAmC,CAAC,CAAC,KAAK,CAC9D,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,MAAM,CAAE,MAAkC,CAAC,GAAG,CAAC,EAAE,KAAK,CAAC,CAC1E,CAAA;IACH,CAAC;IACD,OAAO,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAA;AACpC,CAAC;AAED,SAAS,OAAO,CAAC,KAAqB,EAAE,KAAoB;IAC1D,MAAM,QAAQ,GAAG,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,KAAK,CAAA;IACpE,OAAO,CACL,KAAK,CAAC,IAAI,KAAK,WAAW;QAC1B,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,IAAI;QAC5B,CAAC,CAAC,QAAQ,CAAC,WAAW,IAAI,MAAM,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,WAAW,CAAC,CAAC,CACrE,CAAA;AACH,CAAC;AAED,KAAK,UAAU,SAAS,CACtB,KAAgB,EAChB,SAA2B;IAE3B,IAAI,KAAK,KAAK,cAAc;QAAE,OAAO,SAAS,CAAC,WAAW,CAAA;IAC1D,IAAI,KAAK,KAAK,OAAO;QACnB,OAAO,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;IACvE,IAAI,KAAK,KAAK,OAAO,EAAE,CAAC;QACtB,MAAM,EAAE,IAAI,EAAE,GAAG,MAAM,MAAM,CAAC,kBAAkB,CAAC,CAAA;QACjD,MAAM,KAAK,GAAa,EAAE,CAAA;QAC1B,IAAI,KAAK,EAAE,MAAM,IAAI,IAAI,IAAI,CAAC,MAAM,EAAE;YACpC,GAAG,EAAE,SAAS,CAAC,GAAG;YAClB,OAAO,EAAE,CAAC,iBAAiB,EAAE,SAAS,CAAC;SACxC,CAAC;YACA,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QAClB,OAAO,KAAK,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;IAChC,CAAC;IACD,OAAO,QAAQ,CACb,MAAM,oBAAoB,CAAC,SAAS,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,EAAE,cAAc,CAAC,EACrE,MAAM,CACP,CAAA;AACH,CAAC;AAED,KAAK,UAAU,SAAS,CACtB,MAAqE,EACrE,SAA2B;IAE3B,IAAI,MAAM,GAAG,KAAK,CAAA;IAClB,IAAI,QAAQ,GAAG,EAAE,CAAA;IACjB,IAAI,MAAM,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;QAC5B,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAA;QACxD,MAAM,UAAU,GAAG,IAAI,MAAM,CAC3B,MAAM,CAAC,OAAO,EACd,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,KAAK,GAAG,CAC/D,CAAA;QACD,MAAM,KAAK,GAAG,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAA;QACrD,MAAM;YACJ,MAAM,CAAC,KAAK,KAAK,UAAU;gBACzB,CAAC,CAAC,KAAK,GAAG,CAAC;gBACX,CAAC,CAAC,MAAM,CAAC,KAAK,KAAK,cAAc;oBAC/B,CAAC,CAAC,KAAK,KAAK,CAAC;oBACb,CAAC,CAAC,KAAK,KAAK,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAA;QAC/C,QAAQ,GAAG,WAAW,KAAK,EAAE,CAAA;IAC/B,CAAC;SAAM,IAAI,MAAM,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;QACvC,MAAM,KAAK,GAAG,SAAS,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAC7C,OAAO,CAAC,KAAK,EAAE;YACb,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,GAAG,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACnE,CAAC,CACH,CAAC,MAAM,CAAA;QACR,MAAM,GAAG,KAAK,IAAI,MAAM,CAAC,GAAG,IAAI,KAAK,IAAI,MAAM,CAAC,GAAG,CAAA;QACnD,QAAQ,GAAG,QAAQ,KAAK,EAAE,CAAA;IAC5B,CAAC;SAAM,IAAI,MAAM,CAAC,IAAI,KAAK,YAAY,EAAE,CAAC;QACxC,MAAM,MAAM,GAAG,SAAS,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,KAAK,EAAE,EAAE,CACjD,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,CAAC,CAC9B,CAAA;QACD,MAAM,KAAK,GAAG,SAAS,CAAC,KAAK,CAAC,SAAS,CACrC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,IAAI,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,CACjE,CAAA;QACD,MAAM,GAAG,MAAM,IAAI,CAAC,IAAI,KAAK,GAAG,MAAM,CAAA;QACtC,QAAQ,GAAG,UAAU,MAAM,UAAU,KAAK,EAAE,CAAA;IAC9C,CAAC;SAAM,CAAC;QACN,MAAM,EAAE,IAAI,EAAE,GAAG,MAAM,MAAM,CAAC,kBAAkB,CAAC,CAAA;QACjD,IAAI,KAAK,GAAG,CAAC,CAAA;QACb,IAAI,KAAK,EAAE,MAAM,IAAI,IAAI,IAAI,CAAC,MAAM,EAAE,EAAE,GAAG,EAAE,SAAS,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC;YAC9D,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC;gBAAE,SAAQ;YAC3C,MAAM,oBAAoB,CAAC,SAAS,CAAC,GAAG,EAAE,IAAI,EAAE,mBAAmB,CAAC,CAAA;YACpE,KAAK,IAAI,CAAC,CAAA;QACZ,CAAC;QACD,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,KAAK,CAAC,CAAA;QAChD,QAAQ,GAAG,SAAS,KAAK,EAAE,CAAA;IAC7B,CAAC;IACD,OAAO;QACL,IAAI,EAAE,MAAM,CAAC,IAAI;QACjB,MAAM;QACN,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,WAAW,EAAE,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ;QACzC,QAAQ;QACR,GAAG,CAAC,MAAM,CAAC,GAAG,KAAK,WAAW,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC3D,CAAA;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAAC,OAQ9C;IAMC,MAAM,OAAO,GAAuB,EAAE,CAAA;IACtC,IAAI,kBAAkB,GAAG,KAAK,CAAA;IAC9B,IAAI,YAAY,GAAG,CAAC,CAAA;IACpB,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;QACxC,IAAI,OAAO,CAAC,GAAG,KAAK,SAAS,IAAI,IAAI,CAAC,GAAG,KAAK,WAAW;YAAE,SAAQ;QACnE,IAAI,IAAI,CAAC,IAAI,KAAK,KAAK,IAAI,IAAI,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;YACpD,OAAO,CAAC,IAAI,CAAC,MAAM,SAAS,CAAC,IAAI,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC,CAAA;YACtD,SAAQ;QACV,CAAC;QACD,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;YACrB,kBAAkB,GAAG,IAAI,CAAA;YACzB,SAAQ;QACV,CAAC;QACD,IAAI,CAAC,OAAO,CAAC,KAAK;YAChB,MAAM,IAAI,KAAK,CAAC,eAAe,IAAI,CAAC,IAAI,yBAAyB,CAAC,CAAA;QACpE,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAA;QAC3B,MAAM,KAAK,GACT,IAAI,CAAC,IAAI,KAAK,KAAK;YACjB,CAAC,CAAC,MAAM,SAAS,CAAC,IAAI,CAAC,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC;YAChD,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC,WAAW,CAAA;QACnC,MAAM,QAAQ,GACZ,IAAI,CAAC,IAAI,KAAK,UAAU;YACtB,CAAC,CAAC,MAAM,QAAQ,CACZ,MAAM,oBAAoB,CACxB,OAAO,CAAC,IAAI,CAAC,GAAG,EAChB,IAAI,CAAC,aAAa,EAClB,eAAe,CAChB,EACD,MAAM,CACP;YACH,CAAC,CAAC,SAAS,CAAA;QACf,MAAM,KAAK,GAAG,MAAM,OAAO,CAAC,GAAG,CAC7B,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,EAAE,GAAG,EAAE,CAC7B,KAAK,CAAC,IAAI,CAAC;YACT,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK;YACL,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;YAC/C,KAAK,EAAE,OAAO,CAAC,UAAU;YACzB,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACtD,CAAC,CACH,CACF,CAAA;QACD,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,MAAM,IAAI,CAAC,CAAA;QAC9D,YAAY,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAA;QAClE,OAAO,CAAC,IAAI,CAAC;YACX,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,MAAM;YACN,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,WAAW,EACT,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,EAAE,WAAW;gBACnD,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;YAChC,WAAW,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC;YAC7C,GAAG,CAAC,IAAI,CAAC,GAAG,KAAK,WAAW,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACzD,CAAC,CAAA;IACJ,CAAC;IACD,MAAM,MAAM,GAAG,OAAO,CAAA;IACtB,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC,CAAA;IAChE,MAAM,KAAK,GACT,KAAK,KAAK,CAAC;QACT,CAAC,CAAC,CAAC;QACH,CAAC,CAAC,MAAM;aACH,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC;aAC7B,MAAM,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,KAAK,CAAA;IAC5D,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,kBAAkB,EAAE,YAAY,EAAE,CAAA;AACtE,CAAC"}
|
|
@@ -0,0 +1,50 @@
|
|
|
1
|
+
import type { RuntimeEvent } from '../core/runtime.js';
|
|
2
|
+
import type { ClaudePluginEvalCase } from './claude-plugin-eval-schema.js';
|
|
3
|
+
import type { EvalRunArtifacts } from './claude-plugin-eval-graders.js';
|
|
4
|
+
export declare const DEFAULT_EVAL_ALLOWED_TOOLS: readonly ["Read", "Glob", "Grep", "Skill"];
|
|
5
|
+
export interface PluginEvalRuntime {
|
|
6
|
+
run(prompt: string, signal: AbortSignal): Promise<{
|
|
7
|
+
text: string;
|
|
8
|
+
turns: number;
|
|
9
|
+
costUsd?: number;
|
|
10
|
+
}>;
|
|
11
|
+
close?(): Promise<void>;
|
|
12
|
+
}
|
|
13
|
+
export interface PluginEvalRuntimeFactory {
|
|
14
|
+
create(options: {
|
|
15
|
+
cwd: string;
|
|
16
|
+
configRoot: string;
|
|
17
|
+
home: string;
|
|
18
|
+
model?: string;
|
|
19
|
+
maxTurns: number;
|
|
20
|
+
pluginDirectories: readonly string[];
|
|
21
|
+
allowedTools: readonly string[];
|
|
22
|
+
appendSystemPrompt?: string;
|
|
23
|
+
historyFile?: string;
|
|
24
|
+
addDirs: readonly string[];
|
|
25
|
+
env: Readonly<Record<string, string>>;
|
|
26
|
+
eventSink(event: RuntimeEvent): void;
|
|
27
|
+
}): Promise<PluginEvalRuntime>;
|
|
28
|
+
}
|
|
29
|
+
export interface EvalSingleRunResult {
|
|
30
|
+
text: string;
|
|
31
|
+
turns: number;
|
|
32
|
+
costUsd: number;
|
|
33
|
+
costKnown: boolean;
|
|
34
|
+
artifacts: EvalRunArtifacts;
|
|
35
|
+
tracePath: string;
|
|
36
|
+
error: string | null;
|
|
37
|
+
termination: 'timeout' | 'interrupted' | null;
|
|
38
|
+
tempRoot?: string;
|
|
39
|
+
}
|
|
40
|
+
export declare function resolveEvalAllowedTools(requested: readonly string[], grants: readonly string[]): string[];
|
|
41
|
+
export declare function runClaudePluginEvalOnce(options: {
|
|
42
|
+
case: ClaudePluginEvalCase;
|
|
43
|
+
factory: PluginEvalRuntimeFactory;
|
|
44
|
+
pluginDirectories: readonly string[];
|
|
45
|
+
model?: string;
|
|
46
|
+
allowTools?: readonly string[];
|
|
47
|
+
scaffold: boolean;
|
|
48
|
+
signal?: AbortSignal;
|
|
49
|
+
}): Promise<EvalSingleRunResult>;
|
|
50
|
+
//# sourceMappingURL=claude-plugin-eval-runner.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"claude-plugin-eval-runner.d.ts","sourceRoot":"","sources":["../../src/plugins/claude-plugin-eval-runner.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAA;AAKtD,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,gCAAgC,CAAA;AAE1E,OAAO,KAAK,EACV,gBAAgB,EAEjB,MAAM,iCAAiC,CAAA;AAExC,eAAO,MAAM,0BAA0B,4CAK7B,CAAA;AAEV,MAAM,WAAW,iBAAiB;IAChC,GAAG,CACD,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,WAAW,GAClB,OAAO,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAC7D,KAAK,CAAC,IAAI,OAAO,CAAC,IAAI,CAAC,CAAA;CACxB;AACD,MAAM,WAAW,wBAAwB;IACvC,MAAM,CAAC,OAAO,EAAE;QACd,GAAG,EAAE,MAAM,CAAA;QACX,UAAU,EAAE,MAAM,CAAA;QAClB,IAAI,EAAE,MAAM,CAAA;QACZ,KAAK,CAAC,EAAE,MAAM,CAAA;QACd,QAAQ,EAAE,MAAM,CAAA;QAChB,iBAAiB,EAAE,SAAS,MAAM,EAAE,CAAA;QACpC,YAAY,EAAE,SAAS,MAAM,EAAE,CAAA;QAC/B,kBAAkB,CAAC,EAAE,MAAM,CAAA;QAC3B,WAAW,CAAC,EAAE,MAAM,CAAA;QACpB,OAAO,EAAE,SAAS,MAAM,EAAE,CAAA;QAC1B,GAAG,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAA;QACrC,SAAS,CAAC,KAAK,EAAE,YAAY,GAAG,IAAI,CAAA;KACrC,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAAA;CAC/B;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,MAAM,CAAA;IACf,SAAS,EAAE,OAAO,CAAA;IAClB,SAAS,EAAE,gBAAgB,CAAA;IAC3B,SAAS,EAAE,MAAM,CAAA;IACjB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,WAAW,EAAE,SAAS,GAAG,aAAa,GAAG,IAAI,CAAA;IAC7C,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAwCD,wBAAgB,uBAAuB,CACrC,SAAS,EAAE,SAAS,MAAM,EAAE,EAC5B,MAAM,EAAE,SAAS,MAAM,EAAE,GACxB,MAAM,EAAE,CAQV;AAqDD,wBAAsB,uBAAuB,CAAC,OAAO,EAAE;IACrD,IAAI,EAAE,oBAAoB,CAAA;IAC1B,OAAO,EAAE,wBAAwB,CAAA;IACjC,iBAAiB,EAAE,SAAS,MAAM,EAAE,CAAA;IACpC,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,UAAU,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;IAC9B,QAAQ,EAAE,OAAO,CAAA;IACjB,MAAM,CAAC,EAAE,WAAW,CAAA;CACrB,GAAG,OAAO,CAAC,mBAAmB,CAAC,CAiH/B"}
|
|
@@ -0,0 +1,189 @@
|
|
|
1
|
+
import { mkdtemp, mkdir, rm, writeFile } from 'node:fs/promises';
|
|
2
|
+
import { tmpdir } from 'node:os';
|
|
3
|
+
import { join } from 'node:path';
|
|
4
|
+
import { BoundedProcessRunner, joinedProcessOutput, } from '../platform/bounded-process-runner.js';
|
|
5
|
+
import { resolveContainedPath } from './claude-plugin-eval-schema.js';
|
|
6
|
+
export const DEFAULT_EVAL_ALLOWED_TOOLS = [
|
|
7
|
+
'Read',
|
|
8
|
+
'Glob',
|
|
9
|
+
'Grep',
|
|
10
|
+
'Skill',
|
|
11
|
+
];
|
|
12
|
+
function abortError() {
|
|
13
|
+
return new DOMException('Eval run timed out', 'AbortError');
|
|
14
|
+
}
|
|
15
|
+
function toolName(rule) {
|
|
16
|
+
const separator = rule.indexOf('(');
|
|
17
|
+
return separator < 0 ? rule : rule.slice(0, separator);
|
|
18
|
+
}
|
|
19
|
+
function gatedTool(rule) {
|
|
20
|
+
const name = toolName(rule);
|
|
21
|
+
return (['Bash', 'Write', 'Edit', 'NotebookEdit', 'WebFetch', 'WebSearch'].includes(name) || name.startsWith('mcp__'));
|
|
22
|
+
}
|
|
23
|
+
function operatorGrants(requested, grants) {
|
|
24
|
+
const name = toolName(requested);
|
|
25
|
+
const wildcardMatch = (pattern) => {
|
|
26
|
+
const parts = pattern.split('*');
|
|
27
|
+
if (parts.length === 1)
|
|
28
|
+
return false;
|
|
29
|
+
if (!requested.startsWith(parts[0] ?? ''))
|
|
30
|
+
return false;
|
|
31
|
+
let cursor = parts[0]?.length ?? 0;
|
|
32
|
+
for (const part of parts.slice(1, -1)) {
|
|
33
|
+
const index = requested.indexOf(part, cursor);
|
|
34
|
+
if (index < 0)
|
|
35
|
+
return false;
|
|
36
|
+
cursor = index + part.length;
|
|
37
|
+
}
|
|
38
|
+
const last = parts.at(-1) ?? '';
|
|
39
|
+
return last.length === 0 || requested.slice(cursor).endsWith(last);
|
|
40
|
+
};
|
|
41
|
+
return grants.some((grant) => grant === requested || grant === name || wildcardMatch(grant));
|
|
42
|
+
}
|
|
43
|
+
export function resolveEvalAllowedTools(requested, grants) {
|
|
44
|
+
const selected = requested.length ? requested : DEFAULT_EVAL_ALLOWED_TOOLS;
|
|
45
|
+
for (const rule of selected)
|
|
46
|
+
if (gatedTool(rule) && !operatorGrants(rule, grants))
|
|
47
|
+
throw new Error(`Eval case requests gated tool ${rule}; grant it with --allow-tools`);
|
|
48
|
+
return [...new Set(selected)];
|
|
49
|
+
}
|
|
50
|
+
async function scaffold(caseDef, cwd, home, signal) {
|
|
51
|
+
if (!caseDef.context.scaffoldScript)
|
|
52
|
+
return;
|
|
53
|
+
const script = await resolveContainedPath(caseDef.dir, caseDef.context.scaffoldScript, 'Scaffold script');
|
|
54
|
+
const result = await new BoundedProcessRunner({
|
|
55
|
+
cwd,
|
|
56
|
+
maxOutputBytes: 16 * 1024,
|
|
57
|
+
}).run({
|
|
58
|
+
command: '/bin/bash',
|
|
59
|
+
args: [script],
|
|
60
|
+
timeoutMs: caseDef.execution.timeoutSeconds * 1000,
|
|
61
|
+
signal,
|
|
62
|
+
env: {
|
|
63
|
+
...caseDef.execution.env,
|
|
64
|
+
HOME: home,
|
|
65
|
+
USERPROFILE: home,
|
|
66
|
+
},
|
|
67
|
+
});
|
|
68
|
+
if (result.timedOut)
|
|
69
|
+
throw new Error('Scaffold timed out');
|
|
70
|
+
if (result.code !== 0)
|
|
71
|
+
throw new Error(`Scaffold failed (${result.code}): ${joinedProcessOutput(result)}`);
|
|
72
|
+
}
|
|
73
|
+
function traceEvent(event) {
|
|
74
|
+
if (event.type === 'tool-call')
|
|
75
|
+
return {
|
|
76
|
+
type: 'tool-call',
|
|
77
|
+
tool: event.call.name,
|
|
78
|
+
input: event.call.input,
|
|
79
|
+
id: event.call.id,
|
|
80
|
+
};
|
|
81
|
+
if (event.type === 'tool-result')
|
|
82
|
+
return {
|
|
83
|
+
type: 'tool-result',
|
|
84
|
+
callId: event.callId,
|
|
85
|
+
content: event.content,
|
|
86
|
+
isError: event.isError,
|
|
87
|
+
};
|
|
88
|
+
return event;
|
|
89
|
+
}
|
|
90
|
+
export async function runClaudePluginEvalOnce(options) {
|
|
91
|
+
const tempRoot = await mkdtemp(join(tmpdir(), 'praxis-eval-'));
|
|
92
|
+
const cwd = join(tempRoot, 'cwd');
|
|
93
|
+
const configRoot = join(tempRoot, 'config');
|
|
94
|
+
const home = join(tempRoot, 'home');
|
|
95
|
+
const out = join(tempRoot, 'out');
|
|
96
|
+
try {
|
|
97
|
+
await Promise.all([mkdir(cwd), mkdir(configRoot), mkdir(home), mkdir(out)]);
|
|
98
|
+
}
|
|
99
|
+
catch (error) {
|
|
100
|
+
await rm(tempRoot, { recursive: true, force: true });
|
|
101
|
+
throw error;
|
|
102
|
+
}
|
|
103
|
+
const trace = [];
|
|
104
|
+
const tracePath = join(out, 'trace.jsonl');
|
|
105
|
+
const controller = new AbortController();
|
|
106
|
+
const forwardAbort = () => controller.abort(options.signal?.reason);
|
|
107
|
+
options.signal?.addEventListener('abort', forwardAbort, { once: true });
|
|
108
|
+
if (options.signal?.aborted)
|
|
109
|
+
forwardAbort();
|
|
110
|
+
const timer = setTimeout(() => controller.abort(abortError()), options.case.execution.timeoutSeconds * 1000);
|
|
111
|
+
let runtime;
|
|
112
|
+
let text = '';
|
|
113
|
+
let turns = 0;
|
|
114
|
+
let costUsd = 0;
|
|
115
|
+
let costKnown = false;
|
|
116
|
+
let error = null;
|
|
117
|
+
let termination = null;
|
|
118
|
+
try {
|
|
119
|
+
if (options.scaffold)
|
|
120
|
+
await scaffold(options.case, cwd, home, controller.signal);
|
|
121
|
+
const addDirs = await Promise.all(options.case.context.addDirs.map((path) => resolveContainedPath(options.case.dir, path, 'Add directory')));
|
|
122
|
+
const historyFile = options.case.context.historyFile
|
|
123
|
+
? await resolveContainedPath(options.case.dir, options.case.context.historyFile, 'History file')
|
|
124
|
+
: undefined;
|
|
125
|
+
runtime = await options.factory.create({
|
|
126
|
+
cwd,
|
|
127
|
+
configRoot,
|
|
128
|
+
home,
|
|
129
|
+
...((options.model ?? options.case.execution.model)
|
|
130
|
+
? { model: options.model ?? options.case.execution.model }
|
|
131
|
+
: {}),
|
|
132
|
+
maxTurns: options.case.execution.maxTurns,
|
|
133
|
+
pluginDirectories: options.pluginDirectories,
|
|
134
|
+
allowedTools: resolveEvalAllowedTools(options.case.execution.allowedTools, options.allowTools ?? []),
|
|
135
|
+
...(options.case.execution.appendSystemPrompt
|
|
136
|
+
? { appendSystemPrompt: options.case.execution.appendSystemPrompt }
|
|
137
|
+
: {}),
|
|
138
|
+
...(historyFile ? { historyFile } : {}),
|
|
139
|
+
addDirs,
|
|
140
|
+
env: options.case.execution.env,
|
|
141
|
+
eventSink: (event) => trace.push(traceEvent(event)),
|
|
142
|
+
});
|
|
143
|
+
const result = await runtime.run(options.case.execution.prompt ?? '', controller.signal);
|
|
144
|
+
text = result.text;
|
|
145
|
+
turns = result.turns;
|
|
146
|
+
costKnown = result.costUsd !== undefined;
|
|
147
|
+
costUsd = result.costUsd ?? 0;
|
|
148
|
+
}
|
|
149
|
+
catch (caught) {
|
|
150
|
+
if (options.signal?.aborted) {
|
|
151
|
+
termination = 'interrupted';
|
|
152
|
+
error = 'Eval run interrupted';
|
|
153
|
+
}
|
|
154
|
+
else if (controller.signal.aborted) {
|
|
155
|
+
termination = 'timeout';
|
|
156
|
+
error = 'Eval run timed out';
|
|
157
|
+
}
|
|
158
|
+
else
|
|
159
|
+
error = caught instanceof Error ? caught.message : String(caught);
|
|
160
|
+
}
|
|
161
|
+
finally {
|
|
162
|
+
clearTimeout(timer);
|
|
163
|
+
options.signal?.removeEventListener('abort', forwardAbort);
|
|
164
|
+
try {
|
|
165
|
+
await runtime?.close?.();
|
|
166
|
+
}
|
|
167
|
+
catch (caught) {
|
|
168
|
+
error ??= `Runtime cleanup failed: ${caught instanceof Error ? caught.message : String(caught)}`;
|
|
169
|
+
}
|
|
170
|
+
try {
|
|
171
|
+
await writeFile(tracePath, trace.map((item) => JSON.stringify(item)).join('\n'));
|
|
172
|
+
}
|
|
173
|
+
catch (caught) {
|
|
174
|
+
error ??= `Trace write failed: ${caught instanceof Error ? caught.message : String(caught)}`;
|
|
175
|
+
}
|
|
176
|
+
}
|
|
177
|
+
return {
|
|
178
|
+
text,
|
|
179
|
+
turns,
|
|
180
|
+
costUsd,
|
|
181
|
+
costKnown,
|
|
182
|
+
artifacts: { lastMessage: text, trace, cwd },
|
|
183
|
+
tracePath,
|
|
184
|
+
error,
|
|
185
|
+
termination,
|
|
186
|
+
tempRoot,
|
|
187
|
+
};
|
|
188
|
+
}
|
|
189
|
+
//# sourceMappingURL=claude-plugin-eval-runner.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"claude-plugin-eval-runner.js","sourceRoot":"","sources":["../../src/plugins/claude-plugin-eval-runner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAA;AAChE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAA;AAChC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAA;AAGhC,OAAO,EACL,oBAAoB,EACpB,mBAAmB,GACpB,MAAM,uCAAuC,CAAA;AAE9C,OAAO,EAAE,oBAAoB,EAAE,MAAM,gCAAgC,CAAA;AAMrE,MAAM,CAAC,MAAM,0BAA0B,GAAG;IACxC,MAAM;IACN,MAAM;IACN,MAAM;IACN,OAAO;CACC,CAAA;AAsCV,SAAS,UAAU;IACjB,OAAO,IAAI,YAAY,CAAC,oBAAoB,EAAE,YAAY,CAAC,CAAA;AAC7D,CAAC;AAED,SAAS,QAAQ,CAAC,IAAY;IAC5B,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;IACnC,OAAO,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAA;AACxD,CAAC;AAED,SAAS,SAAS,CAAC,IAAY;IAC7B,MAAM,IAAI,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAA;IAC3B,OAAO,CACL,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,UAAU,EAAE,WAAW,CAAC,CAAC,QAAQ,CACzE,IAAI,CACL,IAAI,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,CAC9B,CAAA;AACH,CAAC;AAED,SAAS,cAAc,CAAC,SAAiB,EAAE,MAAyB;IAClE,MAAM,IAAI,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAA;IAChC,MAAM,aAAa,GAAG,CAAC,OAAe,EAAW,EAAE;QACjD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAA;QAChC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,KAAK,CAAA;QACpC,IAAI,CAAC,SAAS,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;YAAE,OAAO,KAAK,CAAA;QACvD,IAAI,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,IAAI,CAAC,CAAA;QAClC,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACtC,MAAM,KAAK,GAAG,SAAS,CAAC,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC,CAAA;YAC7C,IAAI,KAAK,GAAG,CAAC;gBAAE,OAAO,KAAK,CAAA;YAC3B,MAAM,GAAG,KAAK,GAAG,IAAI,CAAC,MAAM,CAAA;QAC9B,CAAC;QACD,MAAM,IAAI,GAAG,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAA;QAC/B,OAAO,IAAI,CAAC,MAAM,KAAK,CAAC,IAAI,SAAS,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAA;IACpE,CAAC,CAAA;IACD,OAAO,MAAM,CAAC,IAAI,CAChB,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,SAAS,IAAI,KAAK,KAAK,IAAI,IAAI,aAAa,CAAC,KAAK,CAAC,CACzE,CAAA;AACH,CAAC;AAED,MAAM,UAAU,uBAAuB,CACrC,SAA4B,EAC5B,MAAyB;IAEzB,MAAM,QAAQ,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,0BAA0B,CAAA;IAC1E,KAAK,MAAM,IAAI,IAAI,QAAQ;QACzB,IAAI,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,CAAC,IAAI,EAAE,MAAM,CAAC;YAClD,MAAM,IAAI,KAAK,CACb,iCAAiC,IAAI,+BAA+B,CACrE,CAAA;IACL,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAA;AAC/B,CAAC;AAED,KAAK,UAAU,QAAQ,CACrB,OAA6B,EAC7B,GAAW,EACX,IAAY,EACZ,MAAmB;IAEnB,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,cAAc;QAAE,OAAM;IAC3C,MAAM,MAAM,GAAG,MAAM,oBAAoB,CACvC,OAAO,CAAC,GAAG,EACX,OAAO,CAAC,OAAO,CAAC,cAAc,EAC9B,iBAAiB,CAClB,CAAA;IACD,MAAM,MAAM,GAAG,MAAM,IAAI,oBAAoB,CAAC;QAC5C,GAAG;QACH,cAAc,EAAE,EAAE,GAAG,IAAI;KAC1B,CAAC,CAAC,GAAG,CAAC;QACL,OAAO,EAAE,WAAW;QACpB,IAAI,EAAE,CAAC,MAAM,CAAC;QACd,SAAS,EAAE,OAAO,CAAC,SAAS,CAAC,cAAc,GAAG,IAAI;QAClD,MAAM;QACN,GAAG,EAAE;YACH,GAAG,OAAO,CAAC,SAAS,CAAC,GAAG;YACxB,IAAI,EAAE,IAAI;YACV,WAAW,EAAE,IAAI;SAClB;KACF,CAAC,CAAA;IACF,IAAI,MAAM,CAAC,QAAQ;QAAE,MAAM,IAAI,KAAK,CAAC,oBAAoB,CAAC,CAAA;IAC1D,IAAI,MAAM,CAAC,IAAI,KAAK,CAAC;QACnB,MAAM,IAAI,KAAK,CACb,oBAAoB,MAAM,CAAC,IAAI,MAAM,mBAAmB,CAAC,MAAM,CAAC,EAAE,CACnE,CAAA;AACL,CAAC;AAED,SAAS,UAAU,CAAC,KAAmB;IACrC,IAAI,KAAK,CAAC,IAAI,KAAK,WAAW;QAC5B,OAAO;YACL,IAAI,EAAE,WAAW;YACjB,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI;YACrB,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,KAAK;YACvB,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,EAAE;SAClB,CAAA;IACH,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa;QAC9B,OAAO;YACL,IAAI,EAAE,aAAa;YACnB,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,OAAO,EAAE,KAAK,CAAC,OAAO;SACvB,CAAA;IACH,OAAO,KAAkC,CAAA;AAC3C,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAAC,OAQ7C;IACC,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,cAAc,CAAC,CAAC,CAAA;IAC9D,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAA;IACjC,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAA;IAC3C,MAAM,IAAI,GAAG,IAAI,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAA;IACnC,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAA;IACjC,IAAI,CAAC;QACH,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,KAAK,CAAC,UAAU,CAAC,EAAE,KAAK,CAAC,IAAI,CAAC,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAA;IAC7E,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,EAAE,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAA;QACpD,MAAM,KAAK,CAAA;IACb,CAAC;IACD,MAAM,KAAK,GAAqB,EAAE,CAAA;IAClC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,aAAa,CAAC,CAAA;IAC1C,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAA;IACxC,MAAM,YAAY,GAAG,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACnE,OAAO,CAAC,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAA;IACvE,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO;QAAE,YAAY,EAAE,CAAA;IAC3C,MAAM,KAAK,GAAG,UAAU,CACtB,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,CAAC,UAAU,EAAE,CAAC,EACpC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,cAAc,GAAG,IAAI,CAC7C,CAAA;IACD,IAAI,OAAsC,CAAA;IAC1C,IAAI,IAAI,GAAG,EAAE,CAAA;IACb,IAAI,KAAK,GAAG,CAAC,CAAA;IACb,IAAI,OAAO,GAAG,CAAC,CAAA;IACf,IAAI,SAAS,GAAG,KAAK,CAAA;IACrB,IAAI,KAAK,GAAkB,IAAI,CAAA;IAC/B,IAAI,WAAW,GAAuC,IAAI,CAAA;IAC1D,IAAI,CAAC;QACH,IAAI,OAAO,CAAC,QAAQ;YAClB,MAAM,QAAQ,CAAC,OAAO,CAAC,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,UAAU,CAAC,MAAM,CAAC,CAAA;QAC5D,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,GAAG,CAC/B,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CACxC,oBAAoB,CAAC,OAAO,CAAC,IAAI,CAAC,GAAG,EAAE,IAAI,EAAE,eAAe,CAAC,CAC9D,CACF,CAAA;QACD,MAAM,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW;YAClD,CAAC,CAAC,MAAM,oBAAoB,CACxB,OAAO,CAAC,IAAI,CAAC,GAAG,EAChB,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW,EAChC,cAAc,CACf;YACH,CAAC,CAAC,SAAS,CAAA;QACb,OAAO,GAAG,MAAM,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC;YACrC,GAAG;YACH,UAAU;YACV,IAAI;YACJ,GAAG,CAAC,CAAC,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC;gBACjD,CAAC,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE;gBAC1D,CAAC,CAAC,EAAE,CAAC;YACP,QAAQ,EAAE,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ;YACzC,iBAAiB,EAAE,OAAO,CAAC,iBAAiB;YAC5C,YAAY,EAAE,uBAAuB,CACnC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,YAAY,EACnC,OAAO,CAAC,UAAU,IAAI,EAAE,CACzB;YACD,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,kBAAkB;gBAC3C,CAAC,CAAC,EAAE,kBAAkB,EAAE,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,kBAAkB,EAAE;gBACnE,CAAC,CAAC,EAAE,CAAC;YACP,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACvC,OAAO;YACP,GAAG,EAAE,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG;YAC/B,SAAS,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC;SACpD,CAAC,CAAA;QACF,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,CAC9B,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,IAAI,EAAE,EACnC,UAAU,CAAC,MAAM,CAClB,CAAA;QACD,IAAI,GAAG,MAAM,CAAC,IAAI,CAAA;QAClB,KAAK,GAAG,MAAM,CAAC,KAAK,CAAA;QACpB,SAAS,GAAG,MAAM,CAAC,OAAO,KAAK,SAAS,CAAA;QACxC,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,CAAC,CAAA;IAC/B,CAAC;IAAC,OAAO,MAAM,EAAE,CAAC;QAChB,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,EAAE,CAAC;YAC5B,WAAW,GAAG,aAAa,CAAA;YAC3B,KAAK,GAAG,sBAAsB,CAAA;QAChC,CAAC;aAAM,IAAI,UAAU,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACrC,WAAW,GAAG,SAAS,CAAA;YACvB,KAAK,GAAG,oBAAoB,CAAA;QAC9B,CAAC;;YAAM,KAAK,GAAG,MAAM,YAAY,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAA;IAC1E,CAAC;YAAS,CAAC;QACT,YAAY,CAAC,KAAK,CAAC,CAAA;QACnB,OAAO,CAAC,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAA;QAC1D,IAAI,CAAC;YACH,MAAM,OAAO,EAAE,KAAK,EAAE,EAAE,CAAA;QAC1B,CAAC;QAAC,OAAO,MAAM,EAAE,CAAC;YAChB,KAAK,KAAK,2BACR,MAAM,YAAY,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAC1D,EAAE,CAAA;QACJ,CAAC;QACD,IAAI,CAAC;YACH,MAAM,SAAS,CACb,SAAS,EACT,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CACrD,CAAA;QACH,CAAC;QAAC,OAAO,MAAM,EAAE,CAAC;YAChB,KAAK,KAAK,uBACR,MAAM,YAAY,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAC1D,EAAE,CAAA;QACJ,CAAC;IACH,CAAC;IACD,OAAO;QACL,IAAI;QACJ,KAAK;QACL,OAAO;QACP,SAAS;QACT,SAAS,EAAE,EAAE,WAAW,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,EAAE;QAC5C,SAAS;QACT,KAAK;QACL,WAAW;QACX,QAAQ;KACT,CAAA;AACH,CAAC"}
|
|
@@ -0,0 +1,80 @@
|
|
|
1
|
+
export declare const EVAL_SCHEMA_VERSION = "1.0";
|
|
2
|
+
export declare const MAX_EVAL_FILE_BYTES: number;
|
|
3
|
+
export declare const MAX_EVAL_GRADERS = 256;
|
|
4
|
+
export declare const MAX_EVAL_ARRAY_ITEMS = 256;
|
|
5
|
+
export declare const MAX_EVAL_STRING_LENGTH: number;
|
|
6
|
+
export declare const MAX_EVAL_OBJECT_DEPTH = 16;
|
|
7
|
+
export declare const MAX_EVAL_OBJECT_NODES = 4096;
|
|
8
|
+
export type EvalFocus = 'last_message' | 'trace' | 'files' | {
|
|
9
|
+
source: 'file';
|
|
10
|
+
path: string;
|
|
11
|
+
};
|
|
12
|
+
export type EvalArm = 'with-only' | 'both';
|
|
13
|
+
interface EvalGraderBase {
|
|
14
|
+
name: string;
|
|
15
|
+
weight: number;
|
|
16
|
+
arm?: EvalArm;
|
|
17
|
+
}
|
|
18
|
+
export type ClaudePluginEvalGrader = (EvalGraderBase & {
|
|
19
|
+
type: 'regex';
|
|
20
|
+
target: EvalFocus;
|
|
21
|
+
pattern: string;
|
|
22
|
+
flags: string;
|
|
23
|
+
match: 'contains' | 'not_contains' | `count:${number}`;
|
|
24
|
+
}) | (EvalGraderBase & {
|
|
25
|
+
type: 'tool_order';
|
|
26
|
+
before: EvalToolMatch;
|
|
27
|
+
after: EvalToolMatch;
|
|
28
|
+
}) | (EvalGraderBase & {
|
|
29
|
+
type: 'tool_used';
|
|
30
|
+
tool: string;
|
|
31
|
+
input_match?: Record<string, unknown>;
|
|
32
|
+
min: number;
|
|
33
|
+
max: number;
|
|
34
|
+
}) | (EvalGraderBase & {
|
|
35
|
+
type: 'file_exists';
|
|
36
|
+
path: string;
|
|
37
|
+
exists: boolean;
|
|
38
|
+
}) | (EvalGraderBase & {
|
|
39
|
+
type: 'llm';
|
|
40
|
+
criteria: string;
|
|
41
|
+
focus: EvalFocus;
|
|
42
|
+
}) | (EvalGraderBase & {
|
|
43
|
+
type: 'baseline';
|
|
44
|
+
baseline_file: string;
|
|
45
|
+
criteria: string;
|
|
46
|
+
});
|
|
47
|
+
export type EvalToolMatch = string | {
|
|
48
|
+
tool: string;
|
|
49
|
+
input_match?: Record<string, unknown>;
|
|
50
|
+
};
|
|
51
|
+
export interface ClaudePluginEvalCase {
|
|
52
|
+
schemaVersion: string;
|
|
53
|
+
name: string;
|
|
54
|
+
description?: string;
|
|
55
|
+
tags: readonly string[];
|
|
56
|
+
plugins?: readonly string[];
|
|
57
|
+
context: {
|
|
58
|
+
scaffoldScript?: string;
|
|
59
|
+
historyFile?: string;
|
|
60
|
+
addDirs: readonly string[];
|
|
61
|
+
};
|
|
62
|
+
execution: {
|
|
63
|
+
prompt?: string;
|
|
64
|
+
maxTurns: number;
|
|
65
|
+
timeoutSeconds: number;
|
|
66
|
+
model?: string;
|
|
67
|
+
allowedTools: readonly string[];
|
|
68
|
+
appendSystemPrompt?: string;
|
|
69
|
+
env: Readonly<Record<string, string>>;
|
|
70
|
+
};
|
|
71
|
+
runs: number;
|
|
72
|
+
graders: readonly ClaudePluginEvalGrader[];
|
|
73
|
+
expectedOutcome?: string;
|
|
74
|
+
dir: string;
|
|
75
|
+
source: 'case_yaml' | 'prose' | 'mixed';
|
|
76
|
+
}
|
|
77
|
+
export declare function loadClaudePluginEvalCase(caseDir: string): Promise<ClaudePluginEvalCase>;
|
|
78
|
+
export declare function resolveContainedPath(root: string, candidate: string, label: string): Promise<string>;
|
|
79
|
+
export {};
|
|
80
|
+
//# sourceMappingURL=claude-plugin-eval-schema.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"claude-plugin-eval-schema.d.ts","sourceRoot":"","sources":["../../src/plugins/claude-plugin-eval-schema.ts"],"names":[],"mappings":"AAKA,eAAO,MAAM,mBAAmB,QAAQ,CAAA;AACxC,eAAO,MAAM,mBAAmB,QAAc,CAAA;AAC9C,eAAO,MAAM,gBAAgB,MAAM,CAAA;AACnC,eAAO,MAAM,oBAAoB,MAAM,CAAA;AACvC,eAAO,MAAM,sBAAsB,QAAY,CAAA;AAC/C,eAAO,MAAM,qBAAqB,KAAK,CAAA;AACvC,eAAO,MAAM,qBAAqB,OAAO,CAAA;AAIzC,MAAM,MAAM,SAAS,GACnB,cAAc,GAAG,OAAO,GAAG,OAAO,GAAG;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAA;AACvE,MAAM,MAAM,OAAO,GAAG,WAAW,GAAG,MAAM,CAAA;AAE1C,UAAU,cAAc;IACtB,IAAI,EAAE,MAAM,CAAA;IACZ,MAAM,EAAE,MAAM,CAAA;IACd,GAAG,CAAC,EAAE,OAAO,CAAA;CACd;AAED,MAAM,MAAM,sBAAsB,GAC9B,CAAC,cAAc,GAAG;IAChB,IAAI,EAAE,OAAO,CAAA;IACb,MAAM,EAAE,SAAS,CAAA;IACjB,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,UAAU,GAAG,cAAc,GAAG,SAAS,MAAM,EAAE,CAAA;CACvD,CAAC,GACF,CAAC,cAAc,GAAG;IAChB,IAAI,EAAE,YAAY,CAAA;IAClB,MAAM,EAAE,aAAa,CAAA;IACrB,KAAK,EAAE,aAAa,CAAA;CACrB,CAAC,GACF,CAAC,cAAc,GAAG;IAChB,IAAI,EAAE,WAAW,CAAA;IACjB,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IACrC,GAAG,EAAE,MAAM,CAAA;IACX,GAAG,EAAE,MAAM,CAAA;CACZ,CAAC,GACF,CAAC,cAAc,GAAG;IAAE,IAAI,EAAE,aAAa,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,OAAO,CAAA;CAAE,CAAC,GACzE,CAAC,cAAc,GAAG;IAAE,IAAI,EAAE,KAAK,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,SAAS,CAAA;CAAE,CAAC,GACtE,CAAC,cAAc,GAAG;IAChB,IAAI,EAAE,UAAU,CAAA;IAChB,aAAa,EAAE,MAAM,CAAA;IACrB,QAAQ,EAAE,MAAM,CAAA;CACjB,CAAC,CAAA;AAEN,MAAM,MAAM,aAAa,GACrB,MAAM,GACN;IACE,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACtC,CAAA;AAEL,MAAM,WAAW,oBAAoB;IACnC,aAAa,EAAE,MAAM,CAAA;IACrB,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,IAAI,EAAE,SAAS,MAAM,EAAE,CAAA;IACvB,OAAO,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;IAC3B,OAAO,EAAE;QACP,cAAc,CAAC,EAAE,MAAM,CAAA;QACvB,WAAW,CAAC,EAAE,MAAM,CAAA;QACpB,OAAO,EAAE,SAAS,MAAM,EAAE,CAAA;KAC3B,CAAA;IACD,SAAS,EAAE;QACT,MAAM,CAAC,EAAE,MAAM,CAAA;QACf,QAAQ,EAAE,MAAM,CAAA;QAChB,cAAc,EAAE,MAAM,CAAA;QACtB,KAAK,CAAC,EAAE,MAAM,CAAA;QACd,YAAY,EAAE,SAAS,MAAM,EAAE,CAAA;QAC/B,kBAAkB,CAAC,EAAE,MAAM,CAAA;QAC3B,GAAG,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAA;KACtC,CAAA;IACD,IAAI,EAAE,MAAM,CAAA;IACZ,OAAO,EAAE,SAAS,sBAAsB,EAAE,CAAA;IAC1C,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,GAAG,EAAE,MAAM,CAAA;IACX,MAAM,EAAE,WAAW,GAAG,OAAO,GAAG,OAAO,CAAA;CACxC;AAiaD,wBAAsB,wBAAwB,CAC5C,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CAiF/B;AAED,wBAAsB,oBAAoB,CACxC,IAAI,EAAE,MAAM,EACZ,SAAS,EAAE,MAAM,EACjB,KAAK,EAAE,MAAM,GACZ,OAAO,CAAC,MAAM,CAAC,CAMjB"}
|