@rulemetric/cli 0.1.1
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/bin/dev.js +3 -0
- package/bin/run.js +3 -0
- package/dist/base-command.d.ts +12 -0
- package/dist/base-command.d.ts.map +1 -0
- package/dist/base-command.js +37 -0
- package/dist/base-command.js.map +1 -0
- package/dist/commands/auth/create-key.d.ts +11 -0
- package/dist/commands/auth/create-key.d.ts.map +1 -0
- package/dist/commands/auth/create-key.js +34 -0
- package/dist/commands/auth/create-key.js.map +1 -0
- package/dist/commands/auth/login.d.ts +11 -0
- package/dist/commands/auth/login.d.ts.map +1 -0
- package/dist/commands/auth/login.js +40 -0
- package/dist/commands/auth/login.js.map +1 -0
- package/dist/commands/auth/logout.d.ts +7 -0
- package/dist/commands/auth/logout.d.ts.map +1 -0
- package/dist/commands/auth/logout.js +13 -0
- package/dist/commands/auth/logout.js.map +1 -0
- package/dist/commands/auth/revoke-key.d.ts +10 -0
- package/dist/commands/auth/revoke-key.d.ts.map +1 -0
- package/dist/commands/auth/revoke-key.js +18 -0
- package/dist/commands/auth/revoke-key.js.map +1 -0
- package/dist/commands/auth/status.d.ts +7 -0
- package/dist/commands/auth/status.d.ts.map +1 -0
- package/dist/commands/auth/status.js +23 -0
- package/dist/commands/auth/status.js.map +1 -0
- package/dist/commands/catalog/index.d.ts +11 -0
- package/dist/commands/catalog/index.d.ts.map +1 -0
- package/dist/commands/catalog/index.js +35 -0
- package/dist/commands/catalog/index.js.map +1 -0
- package/dist/commands/catalog/recommend.d.ts +9 -0
- package/dist/commands/catalog/recommend.d.ts.map +1 -0
- package/dist/commands/catalog/recommend.js +27 -0
- package/dist/commands/catalog/recommend.js.map +1 -0
- package/dist/commands/convert.d.ts +15 -0
- package/dist/commands/convert.d.ts.map +1 -0
- package/dist/commands/convert.js +55 -0
- package/dist/commands/convert.js.map +1 -0
- package/dist/commands/converters.d.ts +12 -0
- package/dist/commands/converters.d.ts.map +1 -0
- package/dist/commands/converters.js +39 -0
- package/dist/commands/converters.js.map +1 -0
- package/dist/commands/dashboard.d.ts +10 -0
- package/dist/commands/dashboard.d.ts.map +1 -0
- package/dist/commands/dashboard.js +24 -0
- package/dist/commands/dashboard.js.map +1 -0
- package/dist/commands/eval-targets/create.d.ts +14 -0
- package/dist/commands/eval-targets/create.d.ts.map +1 -0
- package/dist/commands/eval-targets/create.js +40 -0
- package/dist/commands/eval-targets/create.js.map +1 -0
- package/dist/commands/eval-targets/list.d.ts +10 -0
- package/dist/commands/eval-targets/list.d.ts.map +1 -0
- package/dist/commands/eval-targets/list.js +36 -0
- package/dist/commands/eval-targets/list.js.map +1 -0
- package/dist/commands/evals/benchmark.d.ts +13 -0
- package/dist/commands/evals/benchmark.d.ts.map +1 -0
- package/dist/commands/evals/benchmark.js +53 -0
- package/dist/commands/evals/benchmark.js.map +1 -0
- package/dist/commands/evals/compare.d.ts +15 -0
- package/dist/commands/evals/compare.d.ts.map +1 -0
- package/dist/commands/evals/compare.js +101 -0
- package/dist/commands/evals/compare.js.map +1 -0
- package/dist/commands/evals/create.d.ts +17 -0
- package/dist/commands/evals/create.d.ts.map +1 -0
- package/dist/commands/evals/create.js +52 -0
- package/dist/commands/evals/create.js.map +1 -0
- package/dist/commands/evals/listen.d.ts +11 -0
- package/dist/commands/evals/listen.d.ts.map +1 -0
- package/dist/commands/evals/listen.js +54 -0
- package/dist/commands/evals/listen.js.map +1 -0
- package/dist/commands/evals/optimize-description.d.ts +22 -0
- package/dist/commands/evals/optimize-description.d.ts.map +1 -0
- package/dist/commands/evals/optimize-description.js +269 -0
- package/dist/commands/evals/optimize-description.js.map +1 -0
- package/dist/commands/evals/optimize.d.ts +22 -0
- package/dist/commands/evals/optimize.d.ts.map +1 -0
- package/dist/commands/evals/optimize.js +200 -0
- package/dist/commands/evals/optimize.js.map +1 -0
- package/dist/commands/evals/run.d.ts +21 -0
- package/dist/commands/evals/run.d.ts.map +1 -0
- package/dist/commands/evals/run.js +168 -0
- package/dist/commands/evals/run.js.map +1 -0
- package/dist/commands/gateway/ensure.d.ts +7 -0
- package/dist/commands/gateway/ensure.d.ts.map +1 -0
- package/dist/commands/gateway/ensure.js +20 -0
- package/dist/commands/gateway/ensure.js.map +1 -0
- package/dist/commands/hooks/install.d.ts +16 -0
- package/dist/commands/hooks/install.d.ts.map +1 -0
- package/dist/commands/hooks/install.js +248 -0
- package/dist/commands/hooks/install.js.map +1 -0
- package/dist/commands/hooks/run.d.ts +11 -0
- package/dist/commands/hooks/run.d.ts.map +1 -0
- package/dist/commands/hooks/run.js +56 -0
- package/dist/commands/hooks/run.js.map +1 -0
- package/dist/commands/hooks/uninstall.d.ts +7 -0
- package/dist/commands/hooks/uninstall.d.ts.map +1 -0
- package/dist/commands/hooks/uninstall.js +88 -0
- package/dist/commands/hooks/uninstall.js.map +1 -0
- package/dist/commands/import.d.ts +17 -0
- package/dist/commands/import.d.ts.map +1 -0
- package/dist/commands/import.js +55 -0
- package/dist/commands/import.js.map +1 -0
- package/dist/commands/insights/sync.d.ts +17 -0
- package/dist/commands/insights/sync.d.ts.map +1 -0
- package/dist/commands/insights/sync.js +477 -0
- package/dist/commands/insights/sync.js.map +1 -0
- package/dist/commands/instructions/create.d.ts +17 -0
- package/dist/commands/instructions/create.d.ts.map +1 -0
- package/dist/commands/instructions/create.js +60 -0
- package/dist/commands/instructions/create.js.map +1 -0
- package/dist/commands/instructions/delete.d.ts +13 -0
- package/dist/commands/instructions/delete.d.ts.map +1 -0
- package/dist/commands/instructions/delete.js +37 -0
- package/dist/commands/instructions/delete.js.map +1 -0
- package/dist/commands/instructions/fork.d.ts +10 -0
- package/dist/commands/instructions/fork.d.ts.map +1 -0
- package/dist/commands/instructions/fork.js +19 -0
- package/dist/commands/instructions/fork.js.map +1 -0
- package/dist/commands/instructions/get.d.ts +15 -0
- package/dist/commands/instructions/get.d.ts.map +1 -0
- package/dist/commands/instructions/get.js +50 -0
- package/dist/commands/instructions/get.js.map +1 -0
- package/dist/commands/instructions/list.d.ts +12 -0
- package/dist/commands/instructions/list.d.ts.map +1 -0
- package/dist/commands/instructions/list.js +47 -0
- package/dist/commands/instructions/list.js.map +1 -0
- package/dist/commands/instructions/promote.d.ts +14 -0
- package/dist/commands/instructions/promote.d.ts.map +1 -0
- package/dist/commands/instructions/promote.js +27 -0
- package/dist/commands/instructions/promote.js.map +1 -0
- package/dist/commands/instructions/pull.d.ts +9 -0
- package/dist/commands/instructions/pull.d.ts.map +1 -0
- package/dist/commands/instructions/pull.js +22 -0
- package/dist/commands/instructions/pull.js.map +1 -0
- package/dist/commands/instructions/upstream.d.ts +9 -0
- package/dist/commands/instructions/upstream.d.ts.map +1 -0
- package/dist/commands/instructions/upstream.js +22 -0
- package/dist/commands/instructions/upstream.js.map +1 -0
- package/dist/commands/instructions/versions.d.ts +12 -0
- package/dist/commands/instructions/versions.d.ts.map +1 -0
- package/dist/commands/instructions/versions.js +34 -0
- package/dist/commands/instructions/versions.js.map +1 -0
- package/dist/commands/packs/add-instruction.d.ts +11 -0
- package/dist/commands/packs/add-instruction.d.ts.map +1 -0
- package/dist/commands/packs/add-instruction.js +22 -0
- package/dist/commands/packs/add-instruction.js.map +1 -0
- package/dist/commands/packs/apply.d.ts +13 -0
- package/dist/commands/packs/apply.d.ts.map +1 -0
- package/dist/commands/packs/apply.js +24 -0
- package/dist/commands/packs/apply.js.map +1 -0
- package/dist/commands/packs/create.d.ts +13 -0
- package/dist/commands/packs/create.d.ts.map +1 -0
- package/dist/commands/packs/create.js +35 -0
- package/dist/commands/packs/create.js.map +1 -0
- package/dist/commands/packs/fork.d.ts +10 -0
- package/dist/commands/packs/fork.d.ts.map +1 -0
- package/dist/commands/packs/fork.js +19 -0
- package/dist/commands/packs/fork.js.map +1 -0
- package/dist/commands/packs/get.d.ts +10 -0
- package/dist/commands/packs/get.d.ts.map +1 -0
- package/dist/commands/packs/get.js +36 -0
- package/dist/commands/packs/get.js.map +1 -0
- package/dist/commands/packs/list.d.ts +22 -0
- package/dist/commands/packs/list.d.ts.map +1 -0
- package/dist/commands/packs/list.js +48 -0
- package/dist/commands/packs/list.js.map +1 -0
- package/dist/commands/packs/remove-instruction.d.ts +11 -0
- package/dist/commands/packs/remove-instruction.d.ts.map +1 -0
- package/dist/commands/packs/remove-instruction.js +20 -0
- package/dist/commands/packs/remove-instruction.js.map +1 -0
- package/dist/commands/proxy/env.d.ts +14 -0
- package/dist/commands/proxy/env.d.ts.map +1 -0
- package/dist/commands/proxy/env.js +118 -0
- package/dist/commands/proxy/env.js.map +1 -0
- package/dist/commands/proxy/logs.d.ts +11 -0
- package/dist/commands/proxy/logs.d.ts.map +1 -0
- package/dist/commands/proxy/logs.js +75 -0
- package/dist/commands/proxy/logs.js.map +1 -0
- package/dist/commands/proxy/setup.d.ts +23 -0
- package/dist/commands/proxy/setup.d.ts.map +1 -0
- package/dist/commands/proxy/setup.js +391 -0
- package/dist/commands/proxy/setup.js.map +1 -0
- package/dist/commands/proxy/start.d.ts +17 -0
- package/dist/commands/proxy/start.d.ts.map +1 -0
- package/dist/commands/proxy/start.js +306 -0
- package/dist/commands/proxy/start.js.map +1 -0
- package/dist/commands/proxy/status.d.ts +7 -0
- package/dist/commands/proxy/status.d.ts.map +1 -0
- package/dist/commands/proxy/status.js +65 -0
- package/dist/commands/proxy/status.js.map +1 -0
- package/dist/commands/proxy/stop.d.ts +8 -0
- package/dist/commands/proxy/stop.d.ts.map +1 -0
- package/dist/commands/proxy/stop.js +77 -0
- package/dist/commands/proxy/stop.js.map +1 -0
- package/dist/commands/service/install.d.ts +17 -0
- package/dist/commands/service/install.d.ts.map +1 -0
- package/dist/commands/service/install.js +383 -0
- package/dist/commands/service/install.js.map +1 -0
- package/dist/commands/service/install.test.d.ts +2 -0
- package/dist/commands/service/install.test.d.ts.map +1 -0
- package/dist/commands/service/install.test.js +345 -0
- package/dist/commands/service/install.test.js.map +1 -0
- package/dist/commands/service/status.d.ts +6 -0
- package/dist/commands/service/status.d.ts.map +1 -0
- package/dist/commands/service/status.js +112 -0
- package/dist/commands/service/status.js.map +1 -0
- package/dist/commands/service/uninstall.d.ts +6 -0
- package/dist/commands/service/uninstall.d.ts.map +1 -0
- package/dist/commands/service/uninstall.js +84 -0
- package/dist/commands/service/uninstall.js.map +1 -0
- package/dist/commands/sessions/analyze.d.ts +10 -0
- package/dist/commands/sessions/analyze.d.ts.map +1 -0
- package/dist/commands/sessions/analyze.js +26 -0
- package/dist/commands/sessions/analyze.js.map +1 -0
- package/dist/commands/sessions/end.d.ts +9 -0
- package/dist/commands/sessions/end.d.ts.map +1 -0
- package/dist/commands/sessions/end.js +82 -0
- package/dist/commands/sessions/end.js.map +1 -0
- package/dist/commands/sessions/import.d.ts +13 -0
- package/dist/commands/sessions/import.d.ts.map +1 -0
- package/dist/commands/sessions/import.js +28 -0
- package/dist/commands/sessions/import.js.map +1 -0
- package/dist/commands/sessions/list.d.ts +11 -0
- package/dist/commands/sessions/list.d.ts.map +1 -0
- package/dist/commands/sessions/list.js +34 -0
- package/dist/commands/sessions/list.js.map +1 -0
- package/dist/commands/sessions/start.d.ts +11 -0
- package/dist/commands/sessions/start.d.ts.map +1 -0
- package/dist/commands/sessions/start.js +83 -0
- package/dist/commands/sessions/start.js.map +1 -0
- package/dist/commands/skills/install.d.ts +15 -0
- package/dist/commands/skills/install.d.ts.map +1 -0
- package/dist/commands/skills/install.js +91 -0
- package/dist/commands/skills/install.js.map +1 -0
- package/dist/commands/skills/list.d.ts +11 -0
- package/dist/commands/skills/list.d.ts.map +1 -0
- package/dist/commands/skills/list.js +47 -0
- package/dist/commands/skills/list.js.map +1 -0
- package/dist/commands/skills/search.d.ts +17 -0
- package/dist/commands/skills/search.d.ts.map +1 -0
- package/dist/commands/skills/search.js +56 -0
- package/dist/commands/skills/search.js.map +1 -0
- package/dist/commands/skills/update.d.ts +7 -0
- package/dist/commands/skills/update.d.ts.map +1 -0
- package/dist/commands/skills/update.js +30 -0
- package/dist/commands/skills/update.js.map +1 -0
- package/dist/dashboard/Dashboard.d.ts +3 -0
- package/dist/dashboard/Dashboard.d.ts.map +1 -0
- package/dist/dashboard/Dashboard.js +132 -0
- package/dist/dashboard/Dashboard.js.map +1 -0
- package/dist/dashboard/data.d.ts +11 -0
- package/dist/dashboard/data.d.ts.map +1 -0
- package/dist/dashboard/data.js +137 -0
- package/dist/dashboard/data.js.map +1 -0
- package/dist/dashboard/panels/friction.d.ts +7 -0
- package/dist/dashboard/panels/friction.d.ts.map +1 -0
- package/dist/dashboard/panels/friction.js +22 -0
- package/dist/dashboard/panels/friction.js.map +1 -0
- package/dist/dashboard/panels/insights.d.ts +8 -0
- package/dist/dashboard/panels/insights.d.ts.map +1 -0
- package/dist/dashboard/panels/insights.js +109 -0
- package/dist/dashboard/panels/insights.js.map +1 -0
- package/dist/dashboard/panels/instructions.d.ts +9 -0
- package/dist/dashboard/panels/instructions.d.ts.map +1 -0
- package/dist/dashboard/panels/instructions.js +17 -0
- package/dist/dashboard/panels/instructions.js.map +1 -0
- package/dist/dashboard/panels/outcomes.d.ts +7 -0
- package/dist/dashboard/panels/outcomes.d.ts.map +1 -0
- package/dist/dashboard/panels/outcomes.js +29 -0
- package/dist/dashboard/panels/outcomes.js.map +1 -0
- package/dist/dashboard/panels/overview.d.ts +8 -0
- package/dist/dashboard/panels/overview.d.ts.map +1 -0
- package/dist/dashboard/panels/overview.js +20 -0
- package/dist/dashboard/panels/overview.js.map +1 -0
- package/dist/dashboard/panels/shared.d.ts +17 -0
- package/dist/dashboard/panels/shared.d.ts.map +1 -0
- package/dist/dashboard/panels/shared.js +62 -0
- package/dist/dashboard/panels/shared.js.map +1 -0
- package/dist/dashboard/types.d.ts +80 -0
- package/dist/dashboard/types.d.ts.map +1 -0
- package/dist/dashboard/types.js +27 -0
- package/dist/dashboard/types.js.map +1 -0
- package/dist/index.d.ts +2 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +2 -0
- package/dist/index.js.map +1 -0
- package/dist/lib/api-client.d.ts +17 -0
- package/dist/lib/api-client.d.ts.map +1 -0
- package/dist/lib/api-client.js +92 -0
- package/dist/lib/api-client.js.map +1 -0
- package/dist/lib/api-client.test.d.ts +2 -0
- package/dist/lib/api-client.test.d.ts.map +1 -0
- package/dist/lib/api-client.test.js +161 -0
- package/dist/lib/api-client.test.js.map +1 -0
- package/dist/lib/auth.d.ts +27 -0
- package/dist/lib/auth.d.ts.map +1 -0
- package/dist/lib/auth.js +137 -0
- package/dist/lib/auth.js.map +1 -0
- package/dist/lib/auth.test.d.ts +2 -0
- package/dist/lib/auth.test.d.ts.map +1 -0
- package/dist/lib/auth.test.js +110 -0
- package/dist/lib/auth.test.js.map +1 -0
- package/dist/lib/eval-analyzer.d.ts +23 -0
- package/dist/lib/eval-analyzer.d.ts.map +1 -0
- package/dist/lib/eval-analyzer.js +89 -0
- package/dist/lib/eval-analyzer.js.map +1 -0
- package/dist/lib/eval-benchmark.d.ts +17 -0
- package/dist/lib/eval-benchmark.d.ts.map +1 -0
- package/dist/lib/eval-benchmark.js +92 -0
- package/dist/lib/eval-benchmark.js.map +1 -0
- package/dist/lib/eval-conversation-handler.d.ts +9 -0
- package/dist/lib/eval-conversation-handler.d.ts.map +1 -0
- package/dist/lib/eval-conversation-handler.js +200 -0
- package/dist/lib/eval-conversation-handler.js.map +1 -0
- package/dist/lib/eval-executor.d.ts +21 -0
- package/dist/lib/eval-executor.d.ts.map +1 -0
- package/dist/lib/eval-executor.js +80 -0
- package/dist/lib/eval-executor.js.map +1 -0
- package/dist/lib/eval-executor.test.d.ts +2 -0
- package/dist/lib/eval-executor.test.d.ts.map +1 -0
- package/dist/lib/eval-executor.test.js +8 -0
- package/dist/lib/eval-executor.test.js.map +1 -0
- package/dist/lib/eval-grader.d.ts +38 -0
- package/dist/lib/eval-grader.d.ts.map +1 -0
- package/dist/lib/eval-grader.js +123 -0
- package/dist/lib/eval-grader.js.map +1 -0
- package/dist/lib/eval-grader.test.d.ts +2 -0
- package/dist/lib/eval-grader.test.d.ts.map +1 -0
- package/dist/lib/eval-grader.test.js +8 -0
- package/dist/lib/eval-grader.test.js.map +1 -0
- package/dist/lib/eval-meta-judge.d.ts +28 -0
- package/dist/lib/eval-meta-judge.d.ts.map +1 -0
- package/dist/lib/eval-meta-judge.js +116 -0
- package/dist/lib/eval-meta-judge.js.map +1 -0
- package/dist/lib/eval-trigger-tester.d.ts +19 -0
- package/dist/lib/eval-trigger-tester.d.ts.map +1 -0
- package/dist/lib/eval-trigger-tester.js +96 -0
- package/dist/lib/eval-trigger-tester.js.map +1 -0
- package/dist/lib/eval-worker.d.ts +12 -0
- package/dist/lib/eval-worker.d.ts.map +1 -0
- package/dist/lib/eval-worker.js +395 -0
- package/dist/lib/eval-worker.js.map +1 -0
- package/dist/lib/eval-worktree.d.ts +11 -0
- package/dist/lib/eval-worktree.d.ts.map +1 -0
- package/dist/lib/eval-worktree.js +87 -0
- package/dist/lib/eval-worktree.js.map +1 -0
- package/dist/lib/eval-worktree.test.d.ts +2 -0
- package/dist/lib/eval-worktree.test.d.ts.map +1 -0
- package/dist/lib/eval-worktree.test.js +68 -0
- package/dist/lib/eval-worktree.test.js.map +1 -0
- package/dist/lib/gateway-entry.d.ts +3 -0
- package/dist/lib/gateway-entry.d.ts.map +1 -0
- package/dist/lib/gateway-entry.js +19 -0
- package/dist/lib/gateway-entry.js.map +1 -0
- package/dist/lib/gateway-lifecycle.d.ts +9 -0
- package/dist/lib/gateway-lifecycle.d.ts.map +1 -0
- package/dist/lib/gateway-lifecycle.js +86 -0
- package/dist/lib/gateway-lifecycle.js.map +1 -0
- package/dist/lib/gateway.d.ts +12 -0
- package/dist/lib/gateway.d.ts.map +1 -0
- package/dist/lib/gateway.js +150 -0
- package/dist/lib/gateway.js.map +1 -0
- package/dist/lib/hooks-config.d.ts +95 -0
- package/dist/lib/hooks-config.d.ts.map +1 -0
- package/dist/lib/hooks-config.js +416 -0
- package/dist/lib/hooks-config.js.map +1 -0
- package/dist/lib/hooks-config.test.d.ts +2 -0
- package/dist/lib/hooks-config.test.d.ts.map +1 -0
- package/dist/lib/hooks-config.test.js +231 -0
- package/dist/lib/hooks-config.test.js.map +1 -0
- package/dist/lib/insights-prompts.d.ts +13 -0
- package/dist/lib/insights-prompts.d.ts.map +1 -0
- package/dist/lib/insights-prompts.js +145 -0
- package/dist/lib/insights-prompts.js.map +1 -0
- package/dist/lib/insights-worker.d.ts +10 -0
- package/dist/lib/insights-worker.d.ts.map +1 -0
- package/dist/lib/insights-worker.js +329 -0
- package/dist/lib/insights-worker.js.map +1 -0
- package/dist/lib/llm-client.d.ts +26 -0
- package/dist/lib/llm-client.d.ts.map +1 -0
- package/dist/lib/llm-client.js +142 -0
- package/dist/lib/llm-client.js.map +1 -0
- package/dist/lib/llm-client.test.d.ts +2 -0
- package/dist/lib/llm-client.test.d.ts.map +1 -0
- package/dist/lib/llm-client.test.js +62 -0
- package/dist/lib/llm-client.test.js.map +1 -0
- package/dist/lib/output.d.ts +4 -0
- package/dist/lib/output.d.ts.map +1 -0
- package/dist/lib/output.js +36 -0
- package/dist/lib/output.js.map +1 -0
- package/dist/lib/output.test.d.ts +2 -0
- package/dist/lib/output.test.d.ts.map +1 -0
- package/dist/lib/output.test.js +71 -0
- package/dist/lib/output.test.js.map +1 -0
- package/dist/lib/provider-aggregator.d.ts +24 -0
- package/dist/lib/provider-aggregator.d.ts.map +1 -0
- package/dist/lib/provider-aggregator.js +57 -0
- package/dist/lib/provider-aggregator.js.map +1 -0
- package/dist/lib/provider-aggregator.test.d.ts +2 -0
- package/dist/lib/provider-aggregator.test.d.ts.map +1 -0
- package/dist/lib/provider-aggregator.test.js +103 -0
- package/dist/lib/provider-aggregator.test.js.map +1 -0
- package/dist/lib/skills-loader.d.ts +6 -0
- package/dist/lib/skills-loader.d.ts.map +1 -0
- package/dist/lib/skills-loader.js +51 -0
- package/dist/lib/skills-loader.js.map +1 -0
- package/dist/lib/types.d.ts +48 -0
- package/dist/lib/types.d.ts.map +1 -0
- package/dist/lib/types.js +7 -0
- package/dist/lib/types.js.map +1 -0
- package/dist/test-helpers.d.ts +58 -0
- package/dist/test-helpers.d.ts.map +1 -0
- package/dist/test-helpers.js +58 -0
- package/dist/test-helpers.js.map +1 -0
- package/oclif.manifest.json +3197 -0
- package/package.json +76 -0
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"create.d.ts","sourceRoot":"","sources":["../../../src/commands/eval-targets/create.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAGpD,MAAM,CAAC,OAAO,OAAO,gBAAiB,SAAQ,WAAW;IACvD,OAAgB,WAAW,SAAyE;IAEpG,OAAgB,QAAQ,WAGtB;IAEF,OAAgB,KAAK;;;;;;MAMnB;IAEI,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC;CAsB3B"}
|
|
@@ -0,0 +1,40 @@
|
|
|
1
|
+
import { Flags } from '@oclif/core';
|
|
2
|
+
import { readFileSync } from 'node:fs';
|
|
3
|
+
import { BaseCommand } from '../../base-command.js';
|
|
4
|
+
import { apiPost } from '../../lib/api-client.js';
|
|
5
|
+
export default class EvalTargetCreate extends BaseCommand {
|
|
6
|
+
static description = 'Create an eval target (instruction, skill, prompt, or plan to test)';
|
|
7
|
+
static examples = [
|
|
8
|
+
'<%= config.bin %> eval-targets create --name my-rules --type instruction --file ./CLAUDE.md',
|
|
9
|
+
'<%= config.bin %> eval-targets create --name my-prompt --type prompt --content "Always use TypeScript"',
|
|
10
|
+
];
|
|
11
|
+
static flags = {
|
|
12
|
+
name: Flags.string({ char: 'n', description: 'Name for the eval target', required: true }),
|
|
13
|
+
type: Flags.string({ char: 't', description: 'Target type', options: ['instruction', 'skill', 'prompt', 'plan'], required: true }),
|
|
14
|
+
file: Flags.string({ char: 'f', description: 'Path to content file' }),
|
|
15
|
+
content: Flags.string({ char: 'c', description: 'Content string (alternative to --file)' }),
|
|
16
|
+
description: Flags.string({ char: 'd', description: 'Description of the target', default: '' }),
|
|
17
|
+
};
|
|
18
|
+
async run() {
|
|
19
|
+
const { flags } = await this.parse(EvalTargetCreate);
|
|
20
|
+
this.requireAuth();
|
|
21
|
+
let content;
|
|
22
|
+
if (flags.file) {
|
|
23
|
+
content = readFileSync(flags.file, 'utf-8');
|
|
24
|
+
}
|
|
25
|
+
else if (flags.content) {
|
|
26
|
+
content = flags.content;
|
|
27
|
+
}
|
|
28
|
+
else {
|
|
29
|
+
this.error('Provide either --file or --content');
|
|
30
|
+
}
|
|
31
|
+
const result = await apiPost('/api/eval-targets', {
|
|
32
|
+
name: flags.name,
|
|
33
|
+
type: flags.type,
|
|
34
|
+
content,
|
|
35
|
+
description: flags.description,
|
|
36
|
+
});
|
|
37
|
+
this.log(`Created eval target: ${result.name} (${result.id}) v${result.version}`);
|
|
38
|
+
}
|
|
39
|
+
}
|
|
40
|
+
//# sourceMappingURL=create.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"create.js","sourceRoot":"","sources":["../../../src/commands/eval-targets/create.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AACpC,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACvC,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AACpD,OAAO,EAAE,OAAO,EAAE,MAAM,yBAAyB,CAAC;AAElD,MAAM,CAAC,OAAO,OAAO,gBAAiB,SAAQ,WAAW;IACvD,MAAM,CAAU,WAAW,GAAG,qEAAqE,CAAC;IAEpG,MAAM,CAAU,QAAQ,GAAG;QACzB,6FAA6F;QAC7F,wGAAwG;KACzG,CAAC;IAEF,MAAM,CAAU,KAAK,GAAG;QACtB,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,0BAA0B,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;QAC1F,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,aAAa,EAAE,OAAO,EAAE,CAAC,aAAa,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;QAClI,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,sBAAsB,EAAE,CAAC;QACtE,OAAO,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,wCAAwC,EAAE,CAAC;QAC3F,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,2BAA2B,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;KAChG,CAAC;IAEF,KAAK,CAAC,GAAG;QACP,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC;QACrD,IAAI,CAAC,WAAW,EAAE,CAAC;QAEnB,IAAI,OAAe,CAAC;QACpB,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;YACf,OAAO,GAAG,YAAY,CAAC,KAAK,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QAC9C,CAAC;aAAM,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;YACzB,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;QAC1B,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,KAAK,CAAC,oCAAoC,CAAC,CAAC;QACnD,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,OAAO,CAAgD,mBAAmB,EAAE;YAC/F,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,OAAO;YACP,WAAW,EAAE,KAAK,CAAC,WAAW;SAC/B,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CAAC,wBAAwB,MAAM,CAAC,IAAI,KAAK,MAAM,CAAC,EAAE,MAAM,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;IACpF,CAAC"}
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
import { BaseCommand } from '../../base-command.js';
|
|
2
|
+
export default class EvalTargetList extends BaseCommand {
|
|
3
|
+
static description: string;
|
|
4
|
+
static examples: string[];
|
|
5
|
+
static flags: {
|
|
6
|
+
type: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
7
|
+
};
|
|
8
|
+
run(): Promise<void>;
|
|
9
|
+
}
|
|
10
|
+
//# sourceMappingURL=list.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"list.d.ts","sourceRoot":"","sources":["../../../src/commands/eval-targets/list.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAapD,MAAM,CAAC,OAAO,OAAO,cAAe,SAAQ,WAAW;IACrD,OAAgB,WAAW,SAAuB;IAElD,OAAgB,QAAQ,WAGtB;IAEF,OAAgB,KAAK;;MAEnB;IAEI,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC;CAuB3B"}
|
|
@@ -0,0 +1,36 @@
|
|
|
1
|
+
import { Flags } from '@oclif/core';
|
|
2
|
+
import { printTable } from '@oclif/table';
|
|
3
|
+
import { BaseCommand } from '../../base-command.js';
|
|
4
|
+
import { apiGet } from '../../lib/api-client.js';
|
|
5
|
+
import { formatDate, truncate } from '../../lib/output.js';
|
|
6
|
+
export default class EvalTargetList extends BaseCommand {
|
|
7
|
+
static description = 'List eval targets';
|
|
8
|
+
static examples = [
|
|
9
|
+
'<%= config.bin %> eval-targets list',
|
|
10
|
+
'<%= config.bin %> eval-targets list --type instruction',
|
|
11
|
+
];
|
|
12
|
+
static flags = {
|
|
13
|
+
type: Flags.string({ char: 't', description: 'Filter by type', options: ['instruction', 'skill', 'prompt', 'plan'] }),
|
|
14
|
+
};
|
|
15
|
+
async run() {
|
|
16
|
+
const { flags } = await this.parse(EvalTargetList);
|
|
17
|
+
this.requireAuth();
|
|
18
|
+
const query = flags.type ? `?type=${flags.type}` : '';
|
|
19
|
+
const results = await apiGet(`/api/eval-targets${query}`);
|
|
20
|
+
if (results.length === 0) {
|
|
21
|
+
this.log('No eval targets found. Create one with `rulemetric eval-targets create`.');
|
|
22
|
+
return;
|
|
23
|
+
}
|
|
24
|
+
printTable({
|
|
25
|
+
data: results.map((t) => ({
|
|
26
|
+
Name: truncate(t.name, 30),
|
|
27
|
+
Type: t.type,
|
|
28
|
+
Version: `v${t.version}`,
|
|
29
|
+
Description: truncate(t.description, 40),
|
|
30
|
+
Updated: formatDate(t.updatedAt),
|
|
31
|
+
})),
|
|
32
|
+
columns: ['Name', 'Type', 'Version', 'Description', 'Updated'],
|
|
33
|
+
});
|
|
34
|
+
}
|
|
35
|
+
}
|
|
36
|
+
//# sourceMappingURL=list.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"list.js","sourceRoot":"","sources":["../../../src/commands/eval-targets/list.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AACpC,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAC1C,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AACpD,OAAO,EAAE,MAAM,EAAE,MAAM,yBAAyB,CAAC;AACjD,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,qBAAqB,CAAC;AAW3D,MAAM,CAAC,OAAO,OAAO,cAAe,SAAQ,WAAW;IACrD,MAAM,CAAU,WAAW,GAAG,mBAAmB,CAAC;IAElD,MAAM,CAAU,QAAQ,GAAG;QACzB,qCAAqC;QACrC,wDAAwD;KACzD,CAAC;IAEF,MAAM,CAAU,KAAK,GAAG;QACtB,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,gBAAgB,EAAE,OAAO,EAAE,CAAC,aAAa,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,CAAC,EAAE,CAAC;KACtH,CAAC;IAEF,KAAK,CAAC,GAAG;QACP,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC;QACnD,IAAI,CAAC,WAAW,EAAE,CAAC;QAEnB,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACtD,MAAM,OAAO,GAAG,MAAM,MAAM,CAAe,oBAAoB,KAAK,EAAE,CAAC,CAAC;QAExE,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACzB,IAAI,CAAC,GAAG,CAAC,0EAA0E,CAAC,CAAC;YACrF,OAAO;QACT,CAAC;QAED,UAAU,CAAC;YACT,IAAI,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBACxB,IAAI,EAAE,QAAQ,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;gBAC1B,IAAI,EAAE,CAAC,CAAC,IAAI;gBACZ,OAAO,EAAE,IAAI,CAAC,CAAC,OAAO,EAAE;gBACxB,WAAW,EAAE,QAAQ,CAAC,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC;gBACxC,OAAO,EAAE,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;aACjC,CAAC,CAAC;YACH,OAAO,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,aAAa,EAAE,SAAS,CAAC;SAC/D,CAAC,CAAC;IACL,CAAC"}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
import { BaseCommand } from '../../base-command.js';
|
|
2
|
+
export default class EvalBenchmarkCommand extends BaseCommand {
|
|
3
|
+
static description: string;
|
|
4
|
+
static examples: string[];
|
|
5
|
+
static args: {
|
|
6
|
+
targetId: import("@oclif/core/interfaces").Arg<string, Record<string, unknown>>;
|
|
7
|
+
};
|
|
8
|
+
static flags: {
|
|
9
|
+
iteration: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
10
|
+
};
|
|
11
|
+
run(): Promise<void>;
|
|
12
|
+
}
|
|
13
|
+
//# sourceMappingURL=benchmark.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark.d.ts","sourceRoot":"","sources":["../../../src/commands/evals/benchmark.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAIpD,MAAM,CAAC,OAAO,OAAO,oBAAqB,SAAQ,WAAW;IAC3D,OAAgB,WAAW,SAAuD;IAElF,OAAgB,QAAQ,WAGtB;IAEF,OAAgB,IAAI;;MAElB;IAEF,OAAgB,KAAK;;MAEnB;IAEI,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC;CA0C3B"}
|
|
@@ -0,0 +1,53 @@
|
|
|
1
|
+
import { Args, Flags } from '@oclif/core';
|
|
2
|
+
import { BaseCommand } from '../../base-command.js';
|
|
3
|
+
import { apiGet } from '../../lib/api-client.js';
|
|
4
|
+
import { aggregateBenchmark, postBenchmark } from '../../lib/eval-benchmark.js';
|
|
5
|
+
export default class EvalBenchmarkCommand extends BaseCommand {
|
|
6
|
+
static description = 'Aggregate eval runs into a benchmark for a target';
|
|
7
|
+
static examples = [
|
|
8
|
+
'<%= config.bin %> evals benchmark <target-id>',
|
|
9
|
+
'<%= config.bin %> evals benchmark <target-id> --iteration 2',
|
|
10
|
+
];
|
|
11
|
+
static args = {
|
|
12
|
+
targetId: Args.string({ description: 'Eval target ID', required: true }),
|
|
13
|
+
};
|
|
14
|
+
static flags = {
|
|
15
|
+
iteration: Flags.integer({ description: 'Filter runs by iteration', default: 1 }),
|
|
16
|
+
};
|
|
17
|
+
async run() {
|
|
18
|
+
const { args, flags } = await this.parse(EvalBenchmarkCommand);
|
|
19
|
+
this.requireAuth();
|
|
20
|
+
const target = await apiGet(`/api/eval-targets/${args.targetId}`);
|
|
21
|
+
this.log(`Aggregating benchmark for "${target.name}" v${target.version}, iteration ${flags.iteration}`);
|
|
22
|
+
const result = await aggregateBenchmark(args.targetId, flags.iteration);
|
|
23
|
+
if (result.totalRuns === 0) {
|
|
24
|
+
this.error('No completed runs found. Run evals first with `rulemetric evals run`.');
|
|
25
|
+
}
|
|
26
|
+
this.log(`\nAggregated ${result.totalRuns} runs\n`);
|
|
27
|
+
// Display summary
|
|
28
|
+
for (const [config, stats] of Object.entries(result.runSummary)) {
|
|
29
|
+
if (config === 'delta') {
|
|
30
|
+
this.log(`\nDelta (with_target - without_target):`);
|
|
31
|
+
for (const [metric, value] of Object.entries(stats)) {
|
|
32
|
+
this.log(` ${metric}: ${value}`);
|
|
33
|
+
}
|
|
34
|
+
continue;
|
|
35
|
+
}
|
|
36
|
+
this.log(`${config}:`);
|
|
37
|
+
for (const [metric, stat] of Object.entries(stats)) {
|
|
38
|
+
const s = stat;
|
|
39
|
+
this.log(` ${metric}: mean=${s.mean} stddev=${s.stddev} [${s.min}, ${s.max}]`);
|
|
40
|
+
}
|
|
41
|
+
}
|
|
42
|
+
if (result.notes.length > 0) {
|
|
43
|
+
this.log('\nNotes:');
|
|
44
|
+
for (const note of result.notes) {
|
|
45
|
+
this.log(` - ${note}`);
|
|
46
|
+
}
|
|
47
|
+
}
|
|
48
|
+
// Post to API
|
|
49
|
+
const benchmark = await postBenchmark(args.targetId, flags.iteration, result);
|
|
50
|
+
this.log(`\nBenchmark saved: ${benchmark.id}`);
|
|
51
|
+
}
|
|
52
|
+
}
|
|
53
|
+
//# sourceMappingURL=benchmark.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark.js","sourceRoot":"","sources":["../../../src/commands/evals/benchmark.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AAC1C,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AACpD,OAAO,EAAE,MAAM,EAAE,MAAM,yBAAyB,CAAC;AACjD,OAAO,EAAE,kBAAkB,EAAE,aAAa,EAAE,MAAM,6BAA6B,CAAC;AAEhF,MAAM,CAAC,OAAO,OAAO,oBAAqB,SAAQ,WAAW;IAC3D,MAAM,CAAU,WAAW,GAAG,mDAAmD,CAAC;IAElF,MAAM,CAAU,QAAQ,GAAG;QACzB,+CAA+C;QAC/C,6DAA6D;KAC9D,CAAC;IAEF,MAAM,CAAU,IAAI,GAAG;QACrB,QAAQ,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,gBAAgB,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;KACzE,CAAC;IAEF,MAAM,CAAU,KAAK,GAAG;QACtB,SAAS,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,0BAA0B,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;KAClF,CAAC;IAEF,KAAK,CAAC,GAAG;QACP,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,oBAAoB,CAAC,CAAC;QAC/D,IAAI,CAAC,WAAW,EAAE,CAAC;QAEnB,MAAM,MAAM,GAAG,MAAM,MAAM,CAAoC,qBAAqB,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;QACrG,IAAI,CAAC,GAAG,CAAC,8BAA8B,MAAM,CAAC,IAAI,MAAM,MAAM,CAAC,OAAO,eAAe,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC;QAExG,MAAM,MAAM,GAAG,MAAM,kBAAkB,CAAC,IAAI,CAAC,QAAQ,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;QAExE,IAAI,MAAM,CAAC,SAAS,KAAK,CAAC,EAAE,CAAC;YAC3B,IAAI,CAAC,KAAK,CAAC,uEAAuE,CAAC,CAAC;QACtF,CAAC;QAED,IAAI,CAAC,GAAG,CAAC,gBAAgB,MAAM,CAAC,SAAS,SAAS,CAAC,CAAC;QAEpD,kBAAkB;QAClB,KAAK,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,UAAU,CAAC,EAAE,CAAC;YAChE,IAAI,MAAM,KAAK,OAAO,EAAE,CAAC;gBACvB,IAAI,CAAC,GAAG,CAAC,yCAAyC,CAAC,CAAC;gBACpD,KAAK,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,KAA0C,CAAC,EAAE,CAAC;oBACzF,IAAI,CAAC,GAAG,CAAC,KAAK,MAAM,KAAK,KAAK,EAAE,CAAC,CAAC;gBACpC,CAAC;gBACD,SAAS;YACX,CAAC;YACD,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,GAAG,CAAC,CAAC;YACvB,KAAK,MAAM,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;gBACnD,MAAM,CAAC,GAAG,IAAkE,CAAC;gBAC7E,IAAI,CAAC,GAAG,CAAC,KAAK,MAAM,UAAU,CAAC,CAAC,IAAI,WAAW,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,GAAG,KAAK,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC;YAClF,CAAC;QACH,CAAC;QAED,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC5B,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;YACrB,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;gBAChC,IAAI,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;YAC1B,CAAC;QACH,CAAC;QAED,cAAc;QACd,MAAM,SAAS,GAAG,MAAM,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,KAAK,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;QAC9E,IAAI,CAAC,GAAG,CAAC,sBAAsB,SAAS,CAAC,EAAE,EAAE,CAAC,CAAC;IACjD,CAAC"}
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
import { BaseCommand } from '../../base-command.js';
|
|
2
|
+
export default class EvalCompareCommand extends BaseCommand {
|
|
3
|
+
static description: string;
|
|
4
|
+
static examples: string[];
|
|
5
|
+
static args: {
|
|
6
|
+
targetA: import("@oclif/core/interfaces").Arg<string, Record<string, unknown>>;
|
|
7
|
+
targetB: import("@oclif/core/interfaces").Arg<string, Record<string, unknown>>;
|
|
8
|
+
};
|
|
9
|
+
static flags: {
|
|
10
|
+
model: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
11
|
+
timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
12
|
+
};
|
|
13
|
+
run(): Promise<void>;
|
|
14
|
+
}
|
|
15
|
+
//# sourceMappingURL=compare.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"compare.d.ts","sourceRoot":"","sources":["../../../src/commands/evals/compare.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAkBpD,MAAM,CAAC,OAAO,OAAO,kBAAmB,SAAQ,WAAW;IACzD,OAAgB,WAAW,SAA2D;IAEtF,OAAgB,QAAQ,WAEtB;IAEF,OAAgB,IAAI;;;MAGlB;IAEF,OAAgB,KAAK;;;MAGnB;IAEI,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC;CAiG3B"}
|
|
@@ -0,0 +1,101 @@
|
|
|
1
|
+
import { Args, Flags } from '@oclif/core';
|
|
2
|
+
import { BaseCommand } from '../../base-command.js';
|
|
3
|
+
import { apiGet } from '../../lib/api-client.js';
|
|
4
|
+
import { executeEval } from '../../lib/eval-executor.js';
|
|
5
|
+
export default class EvalCompareCommand extends BaseCommand {
|
|
6
|
+
static description = 'Blind A/B comparison between two eval target versions';
|
|
7
|
+
static examples = [
|
|
8
|
+
'<%= config.bin %> evals compare <target-a> <target-b>',
|
|
9
|
+
];
|
|
10
|
+
static args = {
|
|
11
|
+
targetA: Args.string({ description: 'First eval target ID', required: true }),
|
|
12
|
+
targetB: Args.string({ description: 'Second eval target ID', required: true }),
|
|
13
|
+
};
|
|
14
|
+
static flags = {
|
|
15
|
+
model: Flags.string({ description: 'Model for execution and comparison' }),
|
|
16
|
+
timeout: Flags.integer({ description: 'Timeout per eval in ms', default: 120_000 }),
|
|
17
|
+
};
|
|
18
|
+
async run() {
|
|
19
|
+
const { args, flags } = await this.parse(EvalCompareCommand);
|
|
20
|
+
this.requireAuth();
|
|
21
|
+
const [targetA, targetB] = await Promise.all([
|
|
22
|
+
apiGet(`/api/eval-targets/${args.targetA}`),
|
|
23
|
+
apiGet(`/api/eval-targets/${args.targetB}`),
|
|
24
|
+
]);
|
|
25
|
+
// Use evals from target A (they should be testing the same thing)
|
|
26
|
+
const evals = await apiGet(`/api/eval-targets/${args.targetA}/evals`);
|
|
27
|
+
if (evals.length === 0) {
|
|
28
|
+
this.error('No evals found for target A. Create some first.');
|
|
29
|
+
}
|
|
30
|
+
this.log(`Comparing "${targetA.name}" v${targetA.version} vs "${targetB.name}" v${targetB.version}`);
|
|
31
|
+
this.log(`Running ${evals.length} eval(s)...\n`);
|
|
32
|
+
// Randomize A/B assignment for blindness
|
|
33
|
+
const isSwapped = Math.random() > 0.5;
|
|
34
|
+
const contentA = isSwapped ? targetB.content : targetA.content;
|
|
35
|
+
const contentB = isSwapped ? targetA.content : targetB.content;
|
|
36
|
+
const labelA = isSwapped ? targetB.name : targetA.name;
|
|
37
|
+
const labelB = isSwapped ? targetA.name : targetB.name;
|
|
38
|
+
const comparisons = [];
|
|
39
|
+
for (const eval_ of evals) {
|
|
40
|
+
this.log(`[${eval_.name}] Running A...`);
|
|
41
|
+
const resultA = await executeEval({
|
|
42
|
+
prompt: eval_.prompt,
|
|
43
|
+
targetContent: contentA,
|
|
44
|
+
timeout: flags.timeout,
|
|
45
|
+
model: flags.model,
|
|
46
|
+
});
|
|
47
|
+
this.log(`[${eval_.name}] Running B...`);
|
|
48
|
+
const resultB = await executeEval({
|
|
49
|
+
prompt: eval_.prompt,
|
|
50
|
+
targetContent: contentB,
|
|
51
|
+
timeout: flags.timeout,
|
|
52
|
+
model: flags.model,
|
|
53
|
+
});
|
|
54
|
+
comparisons.push({
|
|
55
|
+
evalName: eval_.name,
|
|
56
|
+
outputA: resultA.output,
|
|
57
|
+
outputB: resultB.output,
|
|
58
|
+
});
|
|
59
|
+
}
|
|
60
|
+
// Run blind comparison via LLM
|
|
61
|
+
this.log('\nRunning blind comparison...');
|
|
62
|
+
const comparisonPrompt = comparisons.map((c) => `
|
|
63
|
+
## Eval: ${c.evalName}
|
|
64
|
+
|
|
65
|
+
### Output A
|
|
66
|
+
${c.outputA}
|
|
67
|
+
|
|
68
|
+
### Output B
|
|
69
|
+
${c.outputB}
|
|
70
|
+
`).join('\n---\n');
|
|
71
|
+
const comparatorResult = await executeEval({
|
|
72
|
+
prompt: `Compare these two sets of outputs. For each eval, determine which output (A or B) is better.
|
|
73
|
+
Then provide an overall winner with reasoning.
|
|
74
|
+
|
|
75
|
+
${comparisonPrompt}
|
|
76
|
+
|
|
77
|
+
Respond with JSON:
|
|
78
|
+
{
|
|
79
|
+
"per_eval": [
|
|
80
|
+
{ "eval_name": "...", "winner": "A|B|TIE", "reasoning": "..." }
|
|
81
|
+
],
|
|
82
|
+
"overall": {
|
|
83
|
+
"winner": "A|B|TIE",
|
|
84
|
+
"reasoning": "...",
|
|
85
|
+
"strengths_winner": ["..."],
|
|
86
|
+
"weaknesses_loser": ["..."]
|
|
87
|
+
}
|
|
88
|
+
}`,
|
|
89
|
+
timeout: flags.timeout,
|
|
90
|
+
model: flags.model,
|
|
91
|
+
});
|
|
92
|
+
// Unblind and display results
|
|
93
|
+
this.log('\n=== Comparison Results ===\n');
|
|
94
|
+
this.log(`A = "${labelA}" | B = "${labelB}"`);
|
|
95
|
+
if (isSwapped)
|
|
96
|
+
this.log('(Labels were swapped for blind comparison)');
|
|
97
|
+
this.log('');
|
|
98
|
+
this.log(comparatorResult.output);
|
|
99
|
+
}
|
|
100
|
+
}
|
|
101
|
+
//# sourceMappingURL=compare.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"compare.js","sourceRoot":"","sources":["../../../src/commands/evals/compare.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AAC1C,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AACpD,OAAO,EAAE,MAAM,EAAE,MAAM,yBAAyB,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,MAAM,4BAA4B,CAAC;AAgBzD,MAAM,CAAC,OAAO,OAAO,kBAAmB,SAAQ,WAAW;IACzD,MAAM,CAAU,WAAW,GAAG,uDAAuD,CAAC;IAEtF,MAAM,CAAU,QAAQ,GAAG;QACzB,uDAAuD;KACxD,CAAC;IAEF,MAAM,CAAU,IAAI,GAAG;QACrB,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,sBAAsB,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;QAC7E,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,uBAAuB,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;KAC/E,CAAC;IAEF,MAAM,CAAU,KAAK,GAAG;QACtB,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,oCAAoC,EAAE,CAAC;QAC1E,OAAO,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,wBAAwB,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC;KACpF,CAAC;IAEF,KAAK,CAAC,GAAG;QACP,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC;QAC7D,IAAI,CAAC,WAAW,EAAE,CAAC;QAEnB,MAAM,CAAC,OAAO,EAAE,OAAO,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;YAC3C,MAAM,CAAa,qBAAqB,IAAI,CAAC,OAAO,EAAE,CAAC;YACvD,MAAM,CAAa,qBAAqB,IAAI,CAAC,OAAO,EAAE,CAAC;SACxD,CAAC,CAAC;QAEH,kEAAkE;QAClE,MAAM,KAAK,GAAG,MAAM,MAAM,CAAS,qBAAqB,IAAI,CAAC,OAAO,QAAQ,CAAC,CAAC;QAC9E,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACvB,IAAI,CAAC,KAAK,CAAC,iDAAiD,CAAC,CAAC;QAChE,CAAC;QAED,IAAI,CAAC,GAAG,CAAC,cAAc,OAAO,CAAC,IAAI,MAAM,OAAO,CAAC,OAAO,QAAQ,OAAO,CAAC,IAAI,MAAM,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC;QACrG,IAAI,CAAC,GAAG,CAAC,WAAW,KAAK,CAAC,MAAM,eAAe,CAAC,CAAC;QAEjD,yCAAyC;QACzC,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,CAAC;QACtC,MAAM,QAAQ,GAAG,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC;QAC/D,MAAM,QAAQ,GAAG,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC;QAC/D,MAAM,MAAM,GAAG,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC;QACvD,MAAM,MAAM,GAAG,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC;QAEvD,MAAM,WAAW,GAIZ,EAAE,CAAC;QAER,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;YAC1B,IAAI,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,IAAI,gBAAgB,CAAC,CAAC;YACzC,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC;gBAChC,MAAM,EAAE,KAAK,CAAC,MAAM;gBACpB,aAAa,EAAE,QAAQ;gBACvB,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,KAAK,EAAE,KAAK,CAAC,KAAK;aACnB,CAAC,CAAC;YAEH,IAAI,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,IAAI,gBAAgB,CAAC,CAAC;YACzC,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC;gBAChC,MAAM,EAAE,KAAK,CAAC,MAAM;gBACpB,aAAa,EAAE,QAAQ;gBACvB,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,KAAK,EAAE,KAAK,CAAC,KAAK;aACnB,CAAC,CAAC;YAEH,WAAW,CAAC,IAAI,CAAC;gBACf,QAAQ,EAAE,KAAK,CAAC,IAAI;gBACpB,OAAO,EAAE,OAAO,CAAC,MAAM;gBACvB,OAAO,EAAE,OAAO,CAAC,MAAM;aACxB,CAAC,CAAC;QACL,CAAC;QAED,+BAA+B;QAC/B,IAAI,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;QAE1C,MAAM,gBAAgB,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;WACzC,CAAC,CAAC,QAAQ;;;EAGnB,CAAC,CAAC,OAAO;;;EAGT,CAAC,CAAC,OAAO;CACV,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QAEf,MAAM,gBAAgB,GAAG,MAAM,WAAW,CAAC;YACzC,MAAM,EAAE;;;EAGZ,gBAAgB;;;;;;;;;;;;;EAahB;YACI,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,KAAK,EAAE,KAAK,CAAC,KAAK;SACnB,CAAC,CAAC;QAEH,8BAA8B;QAC9B,IAAI,CAAC,GAAG,CAAC,gCAAgC,CAAC,CAAC;QAC3C,IAAI,CAAC,GAAG,CAAC,QAAQ,MAAM,YAAY,MAAM,GAAG,CAAC,CAAC;QAC9C,IAAI,SAAS;YAAE,IAAI,CAAC,GAAG,CAAC,4CAA4C,CAAC,CAAC;QACtE,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QACb,IAAI,CAAC,GAAG,CAAC,gBAAgB,CAAC,MAAM,CAAC,CAAC;IACpC,CAAC"}
|
|
@@ -0,0 +1,17 @@
|
|
|
1
|
+
import { BaseCommand } from '../../base-command.js';
|
|
2
|
+
export default class EvalCreate extends BaseCommand {
|
|
3
|
+
static description: string;
|
|
4
|
+
static examples: string[];
|
|
5
|
+
static args: {
|
|
6
|
+
targetId: import("@oclif/core/interfaces").Arg<string, Record<string, unknown>>;
|
|
7
|
+
};
|
|
8
|
+
static flags: {
|
|
9
|
+
name: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
10
|
+
prompt: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
11
|
+
'expected-output': import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
12
|
+
expectations: import("@oclif/core/interfaces").OptionFlag<string[] | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
13
|
+
'from-file': import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
14
|
+
};
|
|
15
|
+
run(): Promise<void>;
|
|
16
|
+
}
|
|
17
|
+
//# sourceMappingURL=create.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"create.d.ts","sourceRoot":"","sources":["../../../src/commands/evals/create.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAWpD,MAAM,CAAC,OAAO,OAAO,UAAW,SAAQ,WAAW;IACjD,OAAgB,WAAW,SAAkD;IAE7E,OAAgB,QAAQ,WAGtB;IAEF,OAAgB,IAAI;;MAElB;IAEF,OAAgB,KAAK;;;;;;MAMnB;IAEI,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC;CAoC3B"}
|
|
@@ -0,0 +1,52 @@
|
|
|
1
|
+
import { Args, Flags } from '@oclif/core';
|
|
2
|
+
import { readFileSync } from 'node:fs';
|
|
3
|
+
import { BaseCommand } from '../../base-command.js';
|
|
4
|
+
import { apiPost } from '../../lib/api-client.js';
|
|
5
|
+
export default class EvalCreate extends BaseCommand {
|
|
6
|
+
static description = 'Create test cases (evals) for an eval target';
|
|
7
|
+
static examples = [
|
|
8
|
+
'<%= config.bin %> evals create <target-id> --name "ts-check" --prompt "Write a function to add two numbers"',
|
|
9
|
+
'<%= config.bin %> evals create <target-id> --from-file evals.json',
|
|
10
|
+
];
|
|
11
|
+
static args = {
|
|
12
|
+
targetId: Args.string({ description: 'Eval target ID', required: true }),
|
|
13
|
+
};
|
|
14
|
+
static flags = {
|
|
15
|
+
name: Flags.string({ char: 'n', description: 'Eval name' }),
|
|
16
|
+
prompt: Flags.string({ char: 'p', description: 'Test prompt' }),
|
|
17
|
+
'expected-output': Flags.string({ description: 'Expected output description' }),
|
|
18
|
+
expectations: Flags.string({ char: 'e', description: 'Assertion (repeatable)', multiple: true }),
|
|
19
|
+
'from-file': Flags.string({ description: 'Bulk create from JSON file' }),
|
|
20
|
+
};
|
|
21
|
+
async run() {
|
|
22
|
+
const { args, flags } = await this.parse(EvalCreate);
|
|
23
|
+
this.requireAuth();
|
|
24
|
+
if (flags['from-file']) {
|
|
25
|
+
const data = JSON.parse(readFileSync(flags['from-file'], 'utf-8'));
|
|
26
|
+
const evalsArray = Array.isArray(data) ? data : data.evals;
|
|
27
|
+
for (const evalItem of evalsArray) {
|
|
28
|
+
const result = await apiPost(`/api/eval-targets/${args.targetId}/evals`, {
|
|
29
|
+
name: evalItem.name,
|
|
30
|
+
prompt: evalItem.prompt,
|
|
31
|
+
expectedOutput: evalItem.expectedOutput ?? '',
|
|
32
|
+
expectations: evalItem.expectations ?? [],
|
|
33
|
+
files: evalItem.files ?? [],
|
|
34
|
+
});
|
|
35
|
+
this.log(`Created eval: ${result.name} (${result.id})`);
|
|
36
|
+
}
|
|
37
|
+
this.log(`Created ${evalsArray.length} evals.`);
|
|
38
|
+
return;
|
|
39
|
+
}
|
|
40
|
+
if (!flags.name || !flags.prompt) {
|
|
41
|
+
this.error('Provide --name and --prompt, or use --from-file for bulk creation');
|
|
42
|
+
}
|
|
43
|
+
const result = await apiPost(`/api/eval-targets/${args.targetId}/evals`, {
|
|
44
|
+
name: flags.name,
|
|
45
|
+
prompt: flags.prompt,
|
|
46
|
+
expectedOutput: flags['expected-output'] ?? '',
|
|
47
|
+
expectations: flags.expectations ?? [],
|
|
48
|
+
});
|
|
49
|
+
this.log(`Created eval: ${result.name} (${result.id})`);
|
|
50
|
+
}
|
|
51
|
+
}
|
|
52
|
+
//# sourceMappingURL=create.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"create.js","sourceRoot":"","sources":["../../../src/commands/evals/create.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AAC1C,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACvC,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AACpD,OAAO,EAAE,OAAO,EAAE,MAAM,yBAAyB,CAAC;AAUlD,MAAM,CAAC,OAAO,OAAO,UAAW,SAAQ,WAAW;IACjD,MAAM,CAAU,WAAW,GAAG,8CAA8C,CAAC;IAE7E,MAAM,CAAU,QAAQ,GAAG;QACzB,6GAA6G;QAC7G,mEAAmE;KACpE,CAAC;IAEF,MAAM,CAAU,IAAI,GAAG;QACrB,QAAQ,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,gBAAgB,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;KACzE,CAAC;IAEF,MAAM,CAAU,KAAK,GAAG;QACtB,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,WAAW,EAAE,CAAC;QAC3D,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,aAAa,EAAE,CAAC;QAC/D,iBAAiB,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,6BAA6B,EAAE,CAAC;QAC/E,YAAY,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,wBAAwB,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;QAChG,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,4BAA4B,EAAE,CAAC;KACzE,CAAC;IAEF,KAAK,CAAC,GAAG;QACP,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;QACrD,IAAI,CAAC,WAAW,EAAE,CAAC;QAEnB,IAAI,KAAK,CAAC,WAAW,CAAC,EAAE,CAAC;YACvB,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,KAAK,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC;YACnE,MAAM,UAAU,GAAgB,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC;YAExE,KAAK,MAAM,QAAQ,IAAI,UAAU,EAAE,CAAC;gBAClC,MAAM,MAAM,GAAG,MAAM,OAAO,CAA+B,qBAAqB,IAAI,CAAC,QAAQ,QAAQ,EAAE;oBACrG,IAAI,EAAE,QAAQ,CAAC,IAAI;oBACnB,MAAM,EAAE,QAAQ,CAAC,MAAM;oBACvB,cAAc,EAAE,QAAQ,CAAC,cAAc,IAAI,EAAE;oBAC7C,YAAY,EAAE,QAAQ,CAAC,YAAY,IAAI,EAAE;oBACzC,KAAK,EAAE,QAAQ,CAAC,KAAK,IAAI,EAAE;iBAC5B,CAAC,CAAC;gBACH,IAAI,CAAC,GAAG,CAAC,iBAAiB,MAAM,CAAC,IAAI,KAAK,MAAM,CAAC,EAAE,GAAG,CAAC,CAAC;YAC1D,CAAC;YAED,IAAI,CAAC,GAAG,CAAC,WAAW,UAAU,CAAC,MAAM,SAAS,CAAC,CAAC;YAChD,OAAO;QACT,CAAC;QAED,IAAI,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC;YACjC,IAAI,CAAC,KAAK,CAAC,mEAAmE,CAAC,CAAC;QAClF,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,OAAO,CAA+B,qBAAqB,IAAI,CAAC,QAAQ,QAAQ,EAAE;YACrG,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,cAAc,EAAE,KAAK,CAAC,iBAAiB,CAAC,IAAI,EAAE;YAC9C,YAAY,EAAE,KAAK,CAAC,YAAY,IAAI,EAAE;SACvC,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CAAC,iBAAiB,MAAM,CAAC,IAAI,KAAK,MAAM,CAAC,EAAE,GAAG,CAAC,CAAC;IAC1D,CAAC"}
|
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
import { BaseCommand } from '../../base-command.js';
|
|
2
|
+
export default class EvalListenCommand extends BaseCommand {
|
|
3
|
+
static description: string;
|
|
4
|
+
static examples: string[];
|
|
5
|
+
static flags: {
|
|
6
|
+
'supabase-url': import("@oclif/core/interfaces").OptionFlag<string, import("@oclif/core/interfaces").CustomOptions>;
|
|
7
|
+
'supabase-anon-key': import("@oclif/core/interfaces").OptionFlag<string, import("@oclif/core/interfaces").CustomOptions>;
|
|
8
|
+
};
|
|
9
|
+
run(): Promise<void>;
|
|
10
|
+
}
|
|
11
|
+
//# sourceMappingURL=listen.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"listen.d.ts","sourceRoot":"","sources":["../../../src/commands/evals/listen.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAIpD,MAAM,CAAC,OAAO,OAAO,iBAAkB,SAAQ,WAAW;IACxD,OAAgB,WAAW,SAAmE;IAE9F,OAAgB,QAAQ,WAEtB;IAEF,OAAgB,KAAK;;;MAWnB;IAEI,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC;CAqC3B"}
|
|
@@ -0,0 +1,54 @@
|
|
|
1
|
+
import { Flags } from '@oclif/core';
|
|
2
|
+
import { BaseCommand } from '../../base-command.js';
|
|
3
|
+
import { startWorker, createWorkerId, getUserIdFromJwt } from '../../lib/eval-worker.js';
|
|
4
|
+
import { loadToken } from '../../lib/auth.js';
|
|
5
|
+
export default class EvalListenCommand extends BaseCommand {
|
|
6
|
+
static description = 'Listen for eval jobs from the web UI and execute them locally';
|
|
7
|
+
static examples = [
|
|
8
|
+
'<%= config.bin %> evals listen',
|
|
9
|
+
];
|
|
10
|
+
static flags = {
|
|
11
|
+
'supabase-url': Flags.string({
|
|
12
|
+
description: 'Supabase project URL',
|
|
13
|
+
env: 'SUPABASE_URL',
|
|
14
|
+
required: true,
|
|
15
|
+
}),
|
|
16
|
+
'supabase-anon-key': Flags.string({
|
|
17
|
+
description: 'Supabase anon key',
|
|
18
|
+
env: 'SUPABASE_ANON_KEY',
|
|
19
|
+
required: true,
|
|
20
|
+
}),
|
|
21
|
+
};
|
|
22
|
+
async run() {
|
|
23
|
+
const { flags } = await this.parse(EvalListenCommand);
|
|
24
|
+
this.requireAuth();
|
|
25
|
+
const token = loadToken();
|
|
26
|
+
const accessToken = token?.accessToken ?? flags['supabase-anon-key'];
|
|
27
|
+
const userId = token ? getUserIdFromJwt(token.accessToken) : null;
|
|
28
|
+
if (!userId) {
|
|
29
|
+
this.error('Could not determine user ID. Run `rulemetric auth login`.');
|
|
30
|
+
}
|
|
31
|
+
this.log(`Starting eval worker (${createWorkerId()})...`);
|
|
32
|
+
this.log('Press Ctrl+C to stop.\n');
|
|
33
|
+
const cleanup = await startWorker({
|
|
34
|
+
supabaseUrl: flags['supabase-url'],
|
|
35
|
+
supabaseAnonKey: flags['supabase-anon-key'],
|
|
36
|
+
accessToken,
|
|
37
|
+
userId,
|
|
38
|
+
log: (msg) => this.log(msg),
|
|
39
|
+
});
|
|
40
|
+
// Keep alive until SIGINT
|
|
41
|
+
await new Promise((resolve) => {
|
|
42
|
+
process.on('SIGINT', () => {
|
|
43
|
+
this.log('\nShutting down worker...');
|
|
44
|
+
cleanup();
|
|
45
|
+
resolve();
|
|
46
|
+
});
|
|
47
|
+
process.on('SIGTERM', () => {
|
|
48
|
+
cleanup();
|
|
49
|
+
resolve();
|
|
50
|
+
});
|
|
51
|
+
});
|
|
52
|
+
}
|
|
53
|
+
}
|
|
54
|
+
//# sourceMappingURL=listen.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"listen.js","sourceRoot":"","sources":["../../../src/commands/evals/listen.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AACpC,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AACpD,OAAO,EAAE,WAAW,EAAE,cAAc,EAAE,gBAAgB,EAAE,MAAM,0BAA0B,CAAC;AACzF,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,MAAM,CAAC,OAAO,OAAO,iBAAkB,SAAQ,WAAW;IACxD,MAAM,CAAU,WAAW,GAAG,+DAA+D,CAAC;IAE9F,MAAM,CAAU,QAAQ,GAAG;QACzB,gCAAgC;KACjC,CAAC;IAEF,MAAM,CAAU,KAAK,GAAG;QACtB,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC;YAC3B,WAAW,EAAE,sBAAsB;YACnC,GAAG,EAAE,cAAc;YACnB,QAAQ,EAAE,IAAI;SACf,CAAC;QACF,mBAAmB,EAAE,KAAK,CAAC,MAAM,CAAC;YAChC,WAAW,EAAE,mBAAmB;YAChC,GAAG,EAAE,mBAAmB;YACxB,QAAQ,EAAE,IAAI;SACf,CAAC;KACH,CAAC;IAEF,KAAK,CAAC,GAAG;QACP,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC;QACtD,IAAI,CAAC,WAAW,EAAE,CAAC;QAEnB,MAAM,KAAK,GAAG,SAAS,EAAE,CAAC;QAE1B,MAAM,WAAW,GAAG,KAAK,EAAE,WAAW,IAAI,KAAK,CAAC,mBAAmB,CAAC,CAAC;QACrE,MAAM,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,gBAAgB,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QAElE,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,IAAI,CAAC,KAAK,CAAC,2DAA2D,CAAC,CAAC;QAC1E,CAAC;QAED,IAAI,CAAC,GAAG,CAAC,yBAAyB,cAAc,EAAE,MAAM,CAAC,CAAC;QAC1D,IAAI,CAAC,GAAG,CAAC,yBAAyB,CAAC,CAAC;QAEpC,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC;YAChC,WAAW,EAAE,KAAK,CAAC,cAAc,CAAC;YAClC,eAAe,EAAE,KAAK,CAAC,mBAAmB,CAAC;YAC3C,WAAW;YACX,MAAM;YACN,GAAG,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC;SAC5B,CAAC,CAAC;QAEH,0BAA0B;QAC1B,MAAM,IAAI,OAAO,CAAO,CAAC,OAAO,EAAE,EAAE;YAClC,OAAO,CAAC,EAAE,CAAC,QAAQ,EAAE,GAAG,EAAE;gBACxB,IAAI,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC;gBACtC,OAAO,EAAE,CAAC;gBACV,OAAO,EAAE,CAAC;YACZ,CAAC,CAAC,CAAC;YACH,OAAO,CAAC,EAAE,CAAC,SAAS,EAAE,GAAG,EAAE;gBACzB,OAAO,EAAE,CAAC;gBACV,OAAO,EAAE,CAAC;YACZ,CAAC,CAAC,CAAC;QACL,CAAC,CAAC,CAAC;IACL,CAAC"}
|
|
@@ -0,0 +1,22 @@
|
|
|
1
|
+
import { BaseCommand } from '../../base-command.js';
|
|
2
|
+
export default class EvalOptimizeDescriptionCommand extends BaseCommand {
|
|
3
|
+
static description: string;
|
|
4
|
+
static examples: string[];
|
|
5
|
+
static args: {
|
|
6
|
+
targetId: import("@oclif/core/interfaces").Arg<string, Record<string, unknown>>;
|
|
7
|
+
};
|
|
8
|
+
static flags: {
|
|
9
|
+
'max-iterations': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
10
|
+
model: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
11
|
+
queries: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
12
|
+
'runs-per-query': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
13
|
+
'poll-interval': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
14
|
+
timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
15
|
+
};
|
|
16
|
+
run(): Promise<void>;
|
|
17
|
+
private generateTriggerQueries;
|
|
18
|
+
private testQuerySet;
|
|
19
|
+
private printTriggerResults;
|
|
20
|
+
private improveDescription;
|
|
21
|
+
}
|
|
22
|
+
//# sourceMappingURL=optimize-description.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"optimize-description.d.ts","sourceRoot":"","sources":["../../../src/commands/evals/optimize-description.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AA4BpD,MAAM,CAAC,OAAO,OAAO,8BAA+B,SAAQ,WAAW;IACrE,OAAgB,WAAW,SAAqE;IAEhG,OAAgB,QAAQ,WAGtB;IAEF,OAAgB,IAAI;;MAElB;IAEF,OAAgB,KAAK;;;;;;;MAOnB;IAEI,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC;YAoIZ,sBAAsB;YAsEtB,YAAY;IAyC1B,OAAO,CAAC,mBAAmB;YASb,kBAAkB;CAyCjC"}
|