@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 +1,269 @@
|
|
|
1
|
+
import { Args, Flags } from '@oclif/core';
|
|
2
|
+
import { BaseCommand } from '../../base-command.js';
|
|
3
|
+
import { apiGet, apiPost, apiPut } from '../../lib/api-client.js';
|
|
4
|
+
import { executeEval } from '../../lib/eval-executor.js';
|
|
5
|
+
import { testTrigger, computeTriggerScore } from '../../lib/eval-trigger-tester.js';
|
|
6
|
+
export default class EvalOptimizeDescriptionCommand extends BaseCommand {
|
|
7
|
+
static description = 'Optimize eval target description for better triggering accuracy';
|
|
8
|
+
static examples = [
|
|
9
|
+
'<%= config.bin %> evals optimize-description <target-id>',
|
|
10
|
+
'<%= config.bin %> evals optimize-description <target-id> --queries 10',
|
|
11
|
+
];
|
|
12
|
+
static args = {
|
|
13
|
+
targetId: Args.string({ description: 'Eval target ID', required: true }),
|
|
14
|
+
};
|
|
15
|
+
static flags = {
|
|
16
|
+
'max-iterations': Flags.integer({ description: 'Maximum optimization iterations', default: 5 }),
|
|
17
|
+
model: Flags.string({ description: 'Model for trigger testing and improvement' }),
|
|
18
|
+
queries: Flags.integer({ description: 'Number of trigger queries to generate', default: 20 }),
|
|
19
|
+
'runs-per-query': Flags.integer({ description: 'Runs per query for reliability', default: 3 }),
|
|
20
|
+
'poll-interval': Flags.integer({ description: 'Ms between approval polls', default: 5000 }),
|
|
21
|
+
timeout: Flags.integer({ description: 'Timeout per eval in ms', default: 120_000 }),
|
|
22
|
+
};
|
|
23
|
+
async run() {
|
|
24
|
+
const { args, flags } = await this.parse(EvalOptimizeDescriptionCommand);
|
|
25
|
+
this.requireAuth();
|
|
26
|
+
const target = await apiGet(`/api/eval-targets/${args.targetId}`);
|
|
27
|
+
this.log(`Optimizing description for "${target.name}" v${target.version}`);
|
|
28
|
+
this.log(`Current description: "${target.description}"`);
|
|
29
|
+
this.log('');
|
|
30
|
+
// Step 1: Generate trigger queries
|
|
31
|
+
this.log('Generating trigger queries...');
|
|
32
|
+
const queries = await this.generateTriggerQueries(target, flags.queries, flags);
|
|
33
|
+
// Step 2: Post to API for review
|
|
34
|
+
const triggerEval = await apiPost(`/api/eval-targets/${args.targetId}/trigger-evals`, {
|
|
35
|
+
queries,
|
|
36
|
+
});
|
|
37
|
+
this.log(`\nGenerated ${queries.length} trigger queries.`);
|
|
38
|
+
this.log(`Review and approve them at:`);
|
|
39
|
+
this.log(` http://localhost:5174/eval-targets/${args.targetId}`);
|
|
40
|
+
this.log(` (Triggers tab → click to review)`);
|
|
41
|
+
this.log('\nWaiting for approval...');
|
|
42
|
+
// Step 3: Poll for approval
|
|
43
|
+
let approved = false;
|
|
44
|
+
while (!approved) {
|
|
45
|
+
await new Promise((resolve) => setTimeout(resolve, flags['poll-interval']));
|
|
46
|
+
const sets = await apiGet(`/api/eval-targets/${args.targetId}/trigger-evals?status=approved`);
|
|
47
|
+
const current = sets.find((s) => s.id === triggerEval.id);
|
|
48
|
+
if (current) {
|
|
49
|
+
approved = true;
|
|
50
|
+
// Use potentially edited queries
|
|
51
|
+
queries.length = 0;
|
|
52
|
+
queries.push(...current.queries);
|
|
53
|
+
this.log('Trigger queries approved!');
|
|
54
|
+
}
|
|
55
|
+
}
|
|
56
|
+
// Step 4: Split train/test (60/40)
|
|
57
|
+
const indices = queries.map((_, i) => i).sort(() => Math.random() - 0.5);
|
|
58
|
+
const splitIdx = Math.ceil(indices.length * 0.6);
|
|
59
|
+
const trainIndices = indices.slice(0, splitIdx);
|
|
60
|
+
const testIndices = indices.slice(splitIdx);
|
|
61
|
+
// Save split to API
|
|
62
|
+
await apiPut(`/api/trigger-evals/${triggerEval.id}`, {
|
|
63
|
+
trainIndices,
|
|
64
|
+
testIndices,
|
|
65
|
+
});
|
|
66
|
+
const trainQueries = trainIndices.map((i) => queries[i]);
|
|
67
|
+
const testQueries = testIndices.map((i) => queries[i]);
|
|
68
|
+
this.log(`\nSplit: ${trainQueries.length} train, ${testQueries.length} test`);
|
|
69
|
+
// Step 5: Optimization loop
|
|
70
|
+
let currentDescription = target.description;
|
|
71
|
+
let bestDescription = currentDescription;
|
|
72
|
+
let bestTestScore = 0;
|
|
73
|
+
for (let iteration = 1; iteration <= flags['max-iterations']; iteration++) {
|
|
74
|
+
this.log(`\n=== Iteration ${iteration} ===`);
|
|
75
|
+
this.log(`Description: "${currentDescription}"`);
|
|
76
|
+
// Test on train set (multiple runs per query)
|
|
77
|
+
this.log('Testing train set...');
|
|
78
|
+
const trainResults = await this.testQuerySet(trainQueries, target, currentDescription, flags);
|
|
79
|
+
const trainScore = computeTriggerScore(trainResults);
|
|
80
|
+
this.log(`Train accuracy: ${(trainScore * 100).toFixed(1)}%`);
|
|
81
|
+
this.printTriggerResults(trainResults);
|
|
82
|
+
// Test on test set
|
|
83
|
+
this.log('Testing test set...');
|
|
84
|
+
const testResults = await this.testQuerySet(testQueries, target, currentDescription, flags);
|
|
85
|
+
const testScore = computeTriggerScore(testResults);
|
|
86
|
+
this.log(`Test accuracy: ${(testScore * 100).toFixed(1)}%`);
|
|
87
|
+
// Record optimization run
|
|
88
|
+
await apiPost('/api/optimization-runs', {
|
|
89
|
+
evalTargetId: target.id,
|
|
90
|
+
type: 'description',
|
|
91
|
+
iteration,
|
|
92
|
+
inputVersion: target.version,
|
|
93
|
+
trainScore,
|
|
94
|
+
testScore,
|
|
95
|
+
analysis: {
|
|
96
|
+
description: currentDescription,
|
|
97
|
+
trainResults: trainResults.map((r) => ({ query: r.query, shouldTrigger: r.shouldTrigger, didTrigger: r.didTrigger, correct: r.correct })),
|
|
98
|
+
testResults: testResults.map((r) => ({ query: r.query, shouldTrigger: r.shouldTrigger, didTrigger: r.didTrigger, correct: r.correct })),
|
|
99
|
+
},
|
|
100
|
+
status: 'completed',
|
|
101
|
+
});
|
|
102
|
+
// Track best by test score
|
|
103
|
+
if (testScore > bestTestScore) {
|
|
104
|
+
bestTestScore = testScore;
|
|
105
|
+
bestDescription = currentDescription;
|
|
106
|
+
}
|
|
107
|
+
// Stop if perfect
|
|
108
|
+
if (trainScore >= 1.0) {
|
|
109
|
+
this.log('Perfect train score. Stopping.');
|
|
110
|
+
break;
|
|
111
|
+
}
|
|
112
|
+
// Propose improved description
|
|
113
|
+
this.log('Generating improved description...');
|
|
114
|
+
const failures = trainResults.filter((r) => !r.correct);
|
|
115
|
+
currentDescription = await this.improveDescription(target, currentDescription, failures, flags);
|
|
116
|
+
}
|
|
117
|
+
// Step 6: Apply best description
|
|
118
|
+
this.log(`\n=== Optimization Complete ===`);
|
|
119
|
+
this.log(`Best test accuracy: ${(bestTestScore * 100).toFixed(1)}%`);
|
|
120
|
+
this.log(`\nBefore: "${target.description}"`);
|
|
121
|
+
this.log(`After: "${bestDescription}"`);
|
|
122
|
+
if (bestDescription !== target.description) {
|
|
123
|
+
await apiPut(`/api/eval-targets/${target.id}`, {
|
|
124
|
+
description: bestDescription,
|
|
125
|
+
});
|
|
126
|
+
this.log('\nDescription updated.');
|
|
127
|
+
}
|
|
128
|
+
else {
|
|
129
|
+
this.log('\nNo improvement found. Description unchanged.');
|
|
130
|
+
}
|
|
131
|
+
}
|
|
132
|
+
async generateTriggerQueries(target, count, flags) {
|
|
133
|
+
// Try to seed with real session data
|
|
134
|
+
let sessionSeed = '';
|
|
135
|
+
try {
|
|
136
|
+
const sessions = await apiGet('/api/sessions?limit=10');
|
|
137
|
+
if (sessions.length > 0) {
|
|
138
|
+
sessionSeed = '\n\nReal user prompts from recent sessions (use as inspiration for realistic queries):\n' +
|
|
139
|
+
sessions.slice(0, 5).map((s) => `- Session ${s.id}`).join('\n');
|
|
140
|
+
}
|
|
141
|
+
}
|
|
142
|
+
catch {
|
|
143
|
+
// No sessions available, that's fine
|
|
144
|
+
}
|
|
145
|
+
const shouldTriggerCount = Math.ceil(count / 2);
|
|
146
|
+
const shouldNotTriggerCount = count - shouldTriggerCount;
|
|
147
|
+
const prompt = `Generate ${count} trigger eval queries for a skill/instruction.
|
|
148
|
+
|
|
149
|
+
## Skill Name
|
|
150
|
+
${target.name}
|
|
151
|
+
|
|
152
|
+
## Skill Description
|
|
153
|
+
${target.description}
|
|
154
|
+
|
|
155
|
+
## Skill Content (first 1000 chars)
|
|
156
|
+
${target.content.slice(0, 1000)}
|
|
157
|
+
${sessionSeed}
|
|
158
|
+
|
|
159
|
+
Generate ${shouldTriggerCount} should-trigger queries and ${shouldNotTriggerCount} should-not-trigger queries.
|
|
160
|
+
|
|
161
|
+
Rules:
|
|
162
|
+
- Queries must be realistic — what a real user would actually type
|
|
163
|
+
- Include concrete details (file paths, project names, specific problems)
|
|
164
|
+
- Vary length and formality
|
|
165
|
+
- Should-not-trigger queries should be near-misses (share keywords but need something different)
|
|
166
|
+
- Do NOT make should-not-trigger queries obviously irrelevant
|
|
167
|
+
|
|
168
|
+
Respond with ONLY a valid JSON array:
|
|
169
|
+
[
|
|
170
|
+
{"query": "realistic user prompt...", "should_trigger": true},
|
|
171
|
+
{"query": "another prompt...", "should_trigger": false}
|
|
172
|
+
]`;
|
|
173
|
+
const result = await executeEval({
|
|
174
|
+
prompt,
|
|
175
|
+
timeout: flags.timeout,
|
|
176
|
+
model: flags.model,
|
|
177
|
+
});
|
|
178
|
+
try {
|
|
179
|
+
let jsonStr = result.output.trim();
|
|
180
|
+
const jsonMatch = jsonStr.match(/```(?:json)?\s*([\s\S]*?)```/);
|
|
181
|
+
if (jsonMatch)
|
|
182
|
+
jsonStr = jsonMatch[1].trim();
|
|
183
|
+
const arrMatch = jsonStr.match(/\[[\s\S]*\]/);
|
|
184
|
+
if (arrMatch)
|
|
185
|
+
jsonStr = arrMatch[0];
|
|
186
|
+
return JSON.parse(jsonStr);
|
|
187
|
+
}
|
|
188
|
+
catch {
|
|
189
|
+
// Fallback: generate minimal set
|
|
190
|
+
return [
|
|
191
|
+
{ query: `Help me use ${target.name}`, should_trigger: true },
|
|
192
|
+
{ query: 'Write a hello world program', should_trigger: false },
|
|
193
|
+
];
|
|
194
|
+
}
|
|
195
|
+
}
|
|
196
|
+
async testQuerySet(queries, target, description, flags) {
|
|
197
|
+
const results = [];
|
|
198
|
+
for (const q of queries) {
|
|
199
|
+
// Run multiple times and take majority vote
|
|
200
|
+
const runs = [];
|
|
201
|
+
for (let run = 0; run < flags['runs-per-query']; run++) {
|
|
202
|
+
const result = await testTrigger({
|
|
203
|
+
query: q.query,
|
|
204
|
+
shouldTrigger: q.should_trigger,
|
|
205
|
+
skillName: target.name,
|
|
206
|
+
skillDescription: description,
|
|
207
|
+
skillContent: target.content,
|
|
208
|
+
model: flags.model,
|
|
209
|
+
timeout: flags.timeout,
|
|
210
|
+
});
|
|
211
|
+
runs.push(result);
|
|
212
|
+
}
|
|
213
|
+
// Majority vote
|
|
214
|
+
const triggerCount = runs.filter((r) => r.didTrigger).length;
|
|
215
|
+
const didTrigger = triggerCount > flags['runs-per-query'] / 2;
|
|
216
|
+
results.push({
|
|
217
|
+
query: q.query,
|
|
218
|
+
shouldTrigger: q.should_trigger,
|
|
219
|
+
didTrigger,
|
|
220
|
+
correct: didTrigger === q.should_trigger,
|
|
221
|
+
confidence: runs.reduce((sum, r) => sum + r.confidence, 0) / runs.length,
|
|
222
|
+
evidence: runs[0].evidence,
|
|
223
|
+
});
|
|
224
|
+
}
|
|
225
|
+
return results;
|
|
226
|
+
}
|
|
227
|
+
printTriggerResults(results) {
|
|
228
|
+
for (const r of results) {
|
|
229
|
+
const status = r.correct ? 'OK' : 'FAIL';
|
|
230
|
+
const expected = r.shouldTrigger ? 'should trigger' : 'should NOT trigger';
|
|
231
|
+
const actual = r.didTrigger ? 'triggered' : 'did NOT trigger';
|
|
232
|
+
this.log(` [${status}] ${expected} / ${actual}: "${r.query.slice(0, 60)}..."`);
|
|
233
|
+
}
|
|
234
|
+
}
|
|
235
|
+
async improveDescription(target, currentDescription, failures, flags) {
|
|
236
|
+
const failureText = failures.map((f) => {
|
|
237
|
+
const expected = f.shouldTrigger ? 'SHOULD trigger but DID NOT' : 'SHOULD NOT trigger but DID';
|
|
238
|
+
return `- ${expected}: "${f.query}"`;
|
|
239
|
+
}).join('\n');
|
|
240
|
+
const prompt = `You are optimizing a skill description for better triggering accuracy.
|
|
241
|
+
|
|
242
|
+
## Skill Name
|
|
243
|
+
${target.name}
|
|
244
|
+
|
|
245
|
+
## Current Description
|
|
246
|
+
${currentDescription}
|
|
247
|
+
|
|
248
|
+
## Skill Content (first 500 chars)
|
|
249
|
+
${target.content.slice(0, 500)}
|
|
250
|
+
|
|
251
|
+
## Failures
|
|
252
|
+
${failureText}
|
|
253
|
+
|
|
254
|
+
Propose an improved description that:
|
|
255
|
+
- Fixes the trigger failures above
|
|
256
|
+
- Keeps the description concise but comprehensive
|
|
257
|
+
- Uses specific keywords and contexts for when to trigger
|
|
258
|
+
- Includes "Use PROACTIVELY" phrasing for important trigger contexts
|
|
259
|
+
|
|
260
|
+
Respond with ONLY the new description text (no JSON, no explanation, just the description string).`;
|
|
261
|
+
const result = await executeEval({
|
|
262
|
+
prompt,
|
|
263
|
+
timeout: flags.timeout,
|
|
264
|
+
model: flags.model,
|
|
265
|
+
});
|
|
266
|
+
return result.output.trim().replace(/^["']|["']$/g, '');
|
|
267
|
+
}
|
|
268
|
+
}
|
|
269
|
+
//# sourceMappingURL=optimize-description.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"optimize-description.js","sourceRoot":"","sources":["../../../src/commands/evals/optimize-description.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,OAAO,EAAE,MAAM,EAAE,MAAM,yBAAyB,CAAC;AAClE,OAAO,EAAE,WAAW,EAAE,MAAM,4BAA4B,CAAC;AACzD,OAAO,EAAE,WAAW,EAAE,mBAAmB,EAAE,MAAM,kCAAkC,CAAC;AAyBpF,MAAM,CAAC,OAAO,OAAO,8BAA+B,SAAQ,WAAW;IACrE,MAAM,CAAU,WAAW,GAAG,iEAAiE,CAAC;IAEhG,MAAM,CAAU,QAAQ,GAAG;QACzB,0DAA0D;QAC1D,uEAAuE;KACxE,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,gBAAgB,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,iCAAiC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;QAC/F,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,2CAA2C,EAAE,CAAC;QACjF,OAAO,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,uCAAuC,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;QAC7F,gBAAgB,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,gCAAgC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;QAC9F,eAAe,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,2BAA2B,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;QAC3F,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,8BAA8B,CAAC,CAAC;QACzE,IAAI,CAAC,WAAW,EAAE,CAAC;QAEnB,MAAM,MAAM,GAAG,MAAM,MAAM,CAAa,qBAAqB,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;QAE9E,IAAI,CAAC,GAAG,CAAC,+BAA+B,MAAM,CAAC,IAAI,MAAM,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3E,IAAI,CAAC,GAAG,CAAC,yBAAyB,MAAM,CAAC,WAAW,GAAG,CAAC,CAAC;QACzD,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAEb,mCAAmC;QACnC,IAAI,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;QAC1C,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,sBAAsB,CAAC,MAAM,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAEhF,iCAAiC;QACjC,MAAM,WAAW,GAAG,MAAM,OAAO,CAAiB,qBAAqB,IAAI,CAAC,QAAQ,gBAAgB,EAAE;YACpG,OAAO;SACR,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CAAC,eAAe,OAAO,CAAC,MAAM,mBAAmB,CAAC,CAAC;QAC3D,IAAI,CAAC,GAAG,CAAC,6BAA6B,CAAC,CAAC;QACxC,IAAI,CAAC,GAAG,CAAC,wCAAwC,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;QAClE,IAAI,CAAC,GAAG,CAAC,oCAAoC,CAAC,CAAC;QAC/C,IAAI,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC;QAEtC,4BAA4B;QAC5B,IAAI,QAAQ,GAAG,KAAK,CAAC;QACrB,OAAO,CAAC,QAAQ,EAAE,CAAC;YACjB,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC;YAC5E,MAAM,IAAI,GAAG,MAAM,MAAM,CACvB,qBAAqB,IAAI,CAAC,QAAQ,gCAAgC,CACnE,CAAC;YACF,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,WAAW,CAAC,EAAE,CAAC,CAAC;YAC1D,IAAI,OAAO,EAAE,CAAC;gBACZ,QAAQ,GAAG,IAAI,CAAC;gBAChB,iCAAiC;gBACjC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC;gBACnB,OAAO,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;gBACjC,IAAI,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC;YACxC,CAAC;QACH,CAAC;QAED,mCAAmC;QACnC,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,CAAC,CAAC;QACzE,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC;QACjD,MAAM,YAAY,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC;QAChD,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QAE5C,oBAAoB;QACpB,MAAM,MAAM,CAAC,sBAAsB,WAAW,CAAC,EAAE,EAAE,EAAE;YACnD,YAAY;YACZ,WAAW;SACZ,CAAC,CAAC;QAEH,MAAM,YAAY,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;QACzD,MAAM,WAAW,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;QAEvD,IAAI,CAAC,GAAG,CAAC,YAAY,YAAY,CAAC,MAAM,WAAW,WAAW,CAAC,MAAM,OAAO,CAAC,CAAC;QAE9E,4BAA4B;QAC5B,IAAI,kBAAkB,GAAG,MAAM,CAAC,WAAW,CAAC;QAC5C,IAAI,eAAe,GAAG,kBAAkB,CAAC;QACzC,IAAI,aAAa,GAAG,CAAC,CAAC;QAEtB,KAAK,IAAI,SAAS,GAAG,CAAC,EAAE,SAAS,IAAI,KAAK,CAAC,gBAAgB,CAAC,EAAE,SAAS,EAAE,EAAE,CAAC;YAC1E,IAAI,CAAC,GAAG,CAAC,mBAAmB,SAAS,MAAM,CAAC,CAAC;YAC7C,IAAI,CAAC,GAAG,CAAC,iBAAiB,kBAAkB,GAAG,CAAC,CAAC;YAEjD,8CAA8C;YAC9C,IAAI,CAAC,GAAG,CAAC,sBAAsB,CAAC,CAAC;YACjC,MAAM,YAAY,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,YAAY,EAAE,MAAM,EAAE,kBAAkB,EAAE,KAAK,CAAC,CAAC;YAC9F,MAAM,UAAU,GAAG,mBAAmB,CAAC,YAAY,CAAC,CAAC;YACrD,IAAI,CAAC,GAAG,CAAC,mBAAmB,CAAC,UAAU,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;YAE9D,IAAI,CAAC,mBAAmB,CAAC,YAAY,CAAC,CAAC;YAEvC,mBAAmB;YACnB,IAAI,CAAC,GAAG,CAAC,qBAAqB,CAAC,CAAC;YAChC,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,WAAW,EAAE,MAAM,EAAE,kBAAkB,EAAE,KAAK,CAAC,CAAC;YAC5F,MAAM,SAAS,GAAG,mBAAmB,CAAC,WAAW,CAAC,CAAC;YACnD,IAAI,CAAC,GAAG,CAAC,kBAAkB,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;YAE5D,0BAA0B;YAC1B,MAAM,OAAO,CAAC,wBAAwB,EAAE;gBACtC,YAAY,EAAE,MAAM,CAAC,EAAE;gBACvB,IAAI,EAAE,aAAa;gBACnB,SAAS;gBACT,YAAY,EAAE,MAAM,CAAC,OAAO;gBAC5B,UAAU;gBACV,SAAS;gBACT,QAAQ,EAAE;oBACR,WAAW,EAAE,kBAAkB;oBAC/B,YAAY,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,aAAa,EAAE,CAAC,CAAC,aAAa,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;oBACzI,WAAW,EAAE,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,aAAa,EAAE,CAAC,CAAC,aAAa,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;iBACxI;gBACD,MAAM,EAAE,WAAW;aACpB,CAAC,CAAC;YAEH,2BAA2B;YAC3B,IAAI,SAAS,GAAG,aAAa,EAAE,CAAC;gBAC9B,aAAa,GAAG,SAAS,CAAC;gBAC1B,eAAe,GAAG,kBAAkB,CAAC;YACvC,CAAC;YAED,kBAAkB;YAClB,IAAI,UAAU,IAAI,GAAG,EAAE,CAAC;gBACtB,IAAI,CAAC,GAAG,CAAC,gCAAgC,CAAC,CAAC;gBAC3C,MAAM;YACR,CAAC;YAED,+BAA+B;YAC/B,IAAI,CAAC,GAAG,CAAC,oCAAoC,CAAC,CAAC;YAC/C,MAAM,QAAQ,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;YACxD,kBAAkB,GAAG,MAAM,IAAI,CAAC,kBAAkB,CAAC,MAAM,EAAE,kBAAkB,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC;QAClG,CAAC;QAED,iCAAiC;QACjC,IAAI,CAAC,GAAG,CAAC,iCAAiC,CAAC,CAAC;QAC5C,IAAI,CAAC,GAAG,CAAC,uBAAuB,CAAC,aAAa,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACrE,IAAI,CAAC,GAAG,CAAC,cAAc,MAAM,CAAC,WAAW,GAAG,CAAC,CAAC;QAC9C,IAAI,CAAC,GAAG,CAAC,YAAY,eAAe,GAAG,CAAC,CAAC;QAEzC,IAAI,eAAe,KAAK,MAAM,CAAC,WAAW,EAAE,CAAC;YAC3C,MAAM,MAAM,CAAC,qBAAqB,MAAM,CAAC,EAAE,EAAE,EAAE;gBAC7C,WAAW,EAAE,eAAe;aAC7B,CAAC,CAAC;YACH,IAAI,CAAC,GAAG,CAAC,wBAAwB,CAAC,CAAC;QACrC,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,GAAG,CAAC,gDAAgD,CAAC,CAAC;QAC7D,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,sBAAsB,CAClC,MAAkB,EAClB,KAAa,EACb,KAA0C;QAE1C,qCAAqC;QACrC,IAAI,WAAW,GAAG,EAAE,CAAC;QACrB,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAwB,wBAAwB,CAAC,CAAC;YAC/E,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACxB,WAAW,GAAG,0FAA0F;oBACtG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,aAAa,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACpE,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,qCAAqC;QACvC,CAAC;QAED,MAAM,kBAAkB,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QAChD,MAAM,qBAAqB,GAAG,KAAK,GAAG,kBAAkB,CAAC;QAEzD,MAAM,MAAM,GAAG,YAAY,KAAK;;;EAGlC,MAAM,CAAC,IAAI;;;EAGX,MAAM,CAAC,WAAW;;;EAGlB,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC;EAC7B,WAAW;;WAEF,kBAAkB,+BAA+B,qBAAqB;;;;;;;;;;;;;EAa/E,CAAC;QAEC,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC;YAC/B,MAAM;YACN,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,KAAK,EAAE,KAAK,CAAC,KAAK;SACnB,CAAC,CAAC;QAEH,IAAI,CAAC;YACH,IAAI,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;YACnC,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,8BAA8B,CAAC,CAAC;YAChE,IAAI,SAAS;gBAAE,OAAO,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;YAC7C,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;YAC9C,IAAI,QAAQ;gBAAE,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YAEpC,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAmB,CAAC;QAC/C,CAAC;QAAC,MAAM,CAAC;YACP,iCAAiC;YACjC,OAAO;gBACL,EAAE,KAAK,EAAE,eAAe,MAAM,CAAC,IAAI,EAAE,EAAE,cAAc,EAAE,IAAI,EAAE;gBAC7D,EAAE,KAAK,EAAE,6BAA6B,EAAE,cAAc,EAAE,KAAK,EAAE;aAChE,CAAC;QACJ,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,YAAY,CACxB,OAAuB,EACvB,MAAkB,EAClB,WAAmB,EACnB,KAAoE;QAEpE,MAAM,OAAO,GAAwB,EAAE,CAAC;QAExC,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;YACxB,4CAA4C;YAC5C,MAAM,IAAI,GAAwB,EAAE,CAAC;YACrC,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,KAAK,CAAC,gBAAgB,CAAC,EAAE,GAAG,EAAE,EAAE,CAAC;gBACvD,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC;oBAC/B,KAAK,EAAE,CAAC,CAAC,KAAK;oBACd,aAAa,EAAE,CAAC,CAAC,cAAc;oBAC/B,SAAS,EAAE,MAAM,CAAC,IAAI;oBACtB,gBAAgB,EAAE,WAAW;oBAC7B,YAAY,EAAE,MAAM,CAAC,OAAO;oBAC5B,KAAK,EAAE,KAAK,CAAC,KAAK;oBAClB,OAAO,EAAE,KAAK,CAAC,OAAO;iBACvB,CAAC,CAAC;gBACH,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACpB,CAAC;YAED,gBAAgB;YAChB,MAAM,YAAY,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC;YAC7D,MAAM,UAAU,GAAG,YAAY,GAAG,KAAK,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC;YAE9D,OAAO,CAAC,IAAI,CAAC;gBACX,KAAK,EAAE,CAAC,CAAC,KAAK;gBACd,aAAa,EAAE,CAAC,CAAC,cAAc;gBAC/B,UAAU;gBACV,OAAO,EAAE,UAAU,KAAK,CAAC,CAAC,cAAc;gBACxC,UAAU,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM;gBACxE,QAAQ,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ;aAC3B,CAAC,CAAC;QACL,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAEO,mBAAmB,CAAC,OAA4B;QACtD,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;YACxB,MAAM,MAAM,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC;YACzC,MAAM,QAAQ,GAAG,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,oBAAoB,CAAC;YAC3E,MAAM,MAAM,GAAG,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,iBAAiB,CAAC;YAC9D,IAAI,CAAC,GAAG,CAAC,MAAM,MAAM,KAAK,QAAQ,MAAM,MAAM,MAAM,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC;QAClF,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,kBAAkB,CAC9B,MAAkB,EAClB,kBAA0B,EAC1B,QAA6B,EAC7B,KAA0C;QAE1C,MAAM,WAAW,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;YACrC,MAAM,QAAQ,GAAG,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,4BAA4B,CAAC,CAAC,CAAC,4BAA4B,CAAC;YAC/F,OAAO,KAAK,QAAQ,MAAM,CAAC,CAAC,KAAK,GAAG,CAAC;QACvC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEd,MAAM,MAAM,GAAG;;;EAGjB,MAAM,CAAC,IAAI;;;EAGX,kBAAkB;;;EAGlB,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;;;EAG5B,WAAW;;;;;;;;mGAQsF,CAAC;QAEhG,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC;YAC/B,MAAM;YACN,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,KAAK,EAAE,KAAK,CAAC,KAAK;SACnB,CAAC,CAAC;QAEH,OAAO,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,cAAc,EAAE,EAAE,CAAC,CAAC;IAC1D,CAAC"}
|
|
@@ -0,0 +1,22 @@
|
|
|
1
|
+
import { BaseCommand } from '../../base-command.js';
|
|
2
|
+
export default class EvalOptimizeCommand 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
|
+
auto: import("@oclif/core/interfaces").BooleanFlag<boolean>;
|
|
10
|
+
'max-iterations': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
11
|
+
threshold: import("@oclif/core/interfaces").OptionFlag<string, import("@oclif/core/interfaces").CustomOptions>;
|
|
12
|
+
model: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
13
|
+
'analyzer-model': import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
14
|
+
'runs-per-eval': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
15
|
+
parallel: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
16
|
+
timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
17
|
+
};
|
|
18
|
+
run(): Promise<void>;
|
|
19
|
+
private runEvalsOnTarget;
|
|
20
|
+
private computePassRate;
|
|
21
|
+
}
|
|
22
|
+
//# sourceMappingURL=optimize.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"optimize.d.ts","sourceRoot":"","sources":["../../../src/commands/evals/optimize.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAuBpD,MAAM,CAAC,OAAO,OAAO,mBAAoB,SAAQ,WAAW;IAC1D,OAAgB,WAAW,SAA6E;IAExG,OAAgB,QAAQ,WAGtB;IAEF,OAAgB,IAAI;;MAElB;IAEF,OAAgB,KAAK;;;;;;;;;MASnB;IAEI,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC;YA6JZ,gBAAgB;IAqC9B,OAAO,CAAC,eAAe;CAMxB"}
|
|
@@ -0,0 +1,200 @@
|
|
|
1
|
+
import { Args, Flags } from '@oclif/core';
|
|
2
|
+
import { BaseCommand } from '../../base-command.js';
|
|
3
|
+
import { apiGet, apiPost, apiPut } from '../../lib/api-client.js';
|
|
4
|
+
import { executeEval } from '../../lib/eval-executor.js';
|
|
5
|
+
import { gradeEvalOutput } from '../../lib/eval-grader.js';
|
|
6
|
+
import { analyzeEvalResults } from '../../lib/eval-analyzer.js';
|
|
7
|
+
export default class EvalOptimizeCommand extends BaseCommand {
|
|
8
|
+
static description = 'Iteratively optimize eval target content using eval results as feedback';
|
|
9
|
+
static examples = [
|
|
10
|
+
'<%= config.bin %> evals optimize <target-id>',
|
|
11
|
+
'<%= config.bin %> evals optimize <target-id> --auto --max-iterations 3',
|
|
12
|
+
];
|
|
13
|
+
static args = {
|
|
14
|
+
targetId: Args.string({ description: 'Eval target ID', required: true }),
|
|
15
|
+
};
|
|
16
|
+
static flags = {
|
|
17
|
+
auto: Flags.boolean({ description: 'Skip interactive review, apply suggestions automatically', default: false }),
|
|
18
|
+
'max-iterations': Flags.integer({ description: 'Maximum optimization iterations', default: 5 }),
|
|
19
|
+
threshold: Flags.string({ description: 'Stop if pass rate reaches this threshold (0-1)', default: '0.95' }),
|
|
20
|
+
model: Flags.string({ description: 'Model for eval execution' }),
|
|
21
|
+
'analyzer-model': Flags.string({ description: 'Model for analysis/suggestions' }),
|
|
22
|
+
'runs-per-eval': Flags.integer({ description: 'Runs per eval per iteration', default: 1 }),
|
|
23
|
+
parallel: Flags.integer({ description: 'Number of parallel eval executions', default: 1 }),
|
|
24
|
+
timeout: Flags.integer({ description: 'Timeout per eval in ms', default: 120_000 }),
|
|
25
|
+
};
|
|
26
|
+
async run() {
|
|
27
|
+
const { args, flags } = await this.parse(EvalOptimizeCommand);
|
|
28
|
+
this.requireAuth();
|
|
29
|
+
let target = await apiGet(`/api/eval-targets/${args.targetId}`);
|
|
30
|
+
const allEvals = await apiGet(`/api/eval-targets/${args.targetId}/evals`);
|
|
31
|
+
if (allEvals.length === 0) {
|
|
32
|
+
this.error('No evals found. Create some first with `rulemetric evals create`.');
|
|
33
|
+
}
|
|
34
|
+
// Split evals into train/test (60/40)
|
|
35
|
+
const shuffled = [...allEvals].sort(() => Math.random() - 0.5);
|
|
36
|
+
const splitIdx = Math.ceil(shuffled.length * 0.6);
|
|
37
|
+
const trainEvals = shuffled.slice(0, splitIdx);
|
|
38
|
+
const testEvals = shuffled.slice(splitIdx);
|
|
39
|
+
this.log(`Optimizing "${target.name}" v${target.version}`);
|
|
40
|
+
this.log(`Evals: ${trainEvals.length} train, ${testEvals.length} test`);
|
|
41
|
+
const threshold = parseFloat(flags.threshold);
|
|
42
|
+
this.log(`Config: max ${flags['max-iterations']} iterations, threshold ${(threshold * 100).toFixed(0)}%`);
|
|
43
|
+
if (flags.auto)
|
|
44
|
+
this.log('Mode: auto (no interactive review)');
|
|
45
|
+
this.log('');
|
|
46
|
+
let bestVersion = target.version;
|
|
47
|
+
let bestTestScore = 0;
|
|
48
|
+
let noImprovementCount = 0;
|
|
49
|
+
for (let iteration = 1; iteration <= flags['max-iterations']; iteration++) {
|
|
50
|
+
this.log(`\n=== Iteration ${iteration} ===`);
|
|
51
|
+
// Record optimization run as running
|
|
52
|
+
const optRun = await apiPost('/api/optimization-runs', {
|
|
53
|
+
evalTargetId: target.id,
|
|
54
|
+
type: 'content',
|
|
55
|
+
iteration,
|
|
56
|
+
inputVersion: target.version,
|
|
57
|
+
status: 'running',
|
|
58
|
+
});
|
|
59
|
+
// Run train evals
|
|
60
|
+
this.log('Running train evals...');
|
|
61
|
+
const trainResults = await this.runEvalsOnTarget(target, trainEvals, flags);
|
|
62
|
+
const trainScore = this.computePassRate(trainResults);
|
|
63
|
+
this.log(`Train pass rate: ${(trainScore * 100).toFixed(1)}%`);
|
|
64
|
+
// Check if we've hit the threshold
|
|
65
|
+
if (trainScore >= threshold) {
|
|
66
|
+
this.log(`Train score ${(trainScore * 100).toFixed(1)}% >= threshold ${(threshold * 100).toFixed(0)}%. Stopping.`);
|
|
67
|
+
// Run test evals for final score
|
|
68
|
+
const testResults = await this.runEvalsOnTarget(target, testEvals, flags);
|
|
69
|
+
const testScore = this.computePassRate(testResults);
|
|
70
|
+
this.log(`Final test score: ${(testScore * 100).toFixed(1)}%`);
|
|
71
|
+
await apiPost('/api/optimization-runs', {
|
|
72
|
+
evalTargetId: target.id,
|
|
73
|
+
type: 'content',
|
|
74
|
+
iteration,
|
|
75
|
+
inputVersion: target.version,
|
|
76
|
+
trainScore,
|
|
77
|
+
testScore,
|
|
78
|
+
analysis: { reason: 'threshold_reached' },
|
|
79
|
+
status: 'completed',
|
|
80
|
+
});
|
|
81
|
+
break;
|
|
82
|
+
}
|
|
83
|
+
// Analyze and propose improvements
|
|
84
|
+
this.log('Analyzing results...');
|
|
85
|
+
const evalResults = trainResults.map((r) => ({
|
|
86
|
+
evalName: r.evalName,
|
|
87
|
+
prompt: r.prompt,
|
|
88
|
+
expectations: r.expectations,
|
|
89
|
+
grading: r.grading,
|
|
90
|
+
}));
|
|
91
|
+
const analysis = await analyzeEvalResults({
|
|
92
|
+
content: target.content,
|
|
93
|
+
evalResults,
|
|
94
|
+
model: flags['analyzer-model'] ?? flags.model,
|
|
95
|
+
timeout: flags.timeout,
|
|
96
|
+
});
|
|
97
|
+
this.log(`\nAssessment: ${analysis.overallAssessment}`);
|
|
98
|
+
this.log(`Suggestions: ${analysis.suggestions.length}`);
|
|
99
|
+
for (const s of analysis.suggestions) {
|
|
100
|
+
this.log(` - [${s.section}] ${s.reasoning}`);
|
|
101
|
+
}
|
|
102
|
+
// In interactive mode, ask for approval
|
|
103
|
+
if (!flags.auto) {
|
|
104
|
+
this.log('\nProposed changes above. Applying in auto mode would proceed automatically.');
|
|
105
|
+
this.log('(Interactive approval not yet implemented — use --auto for now)');
|
|
106
|
+
}
|
|
107
|
+
if (analysis.proposedContent === target.content) {
|
|
108
|
+
this.log('No changes proposed. Stopping.');
|
|
109
|
+
await apiPost('/api/optimization-runs', {
|
|
110
|
+
evalTargetId: target.id,
|
|
111
|
+
type: 'content',
|
|
112
|
+
iteration,
|
|
113
|
+
inputVersion: target.version,
|
|
114
|
+
trainScore,
|
|
115
|
+
analysis: { reason: 'no_changes' },
|
|
116
|
+
status: 'completed',
|
|
117
|
+
});
|
|
118
|
+
break;
|
|
119
|
+
}
|
|
120
|
+
// Create new version
|
|
121
|
+
const newTarget = await apiPut(`/api/eval-targets/${target.id}`, {
|
|
122
|
+
content: analysis.proposedContent,
|
|
123
|
+
});
|
|
124
|
+
this.log(`Created v${newTarget.version}`);
|
|
125
|
+
// Run test evals on new version
|
|
126
|
+
this.log('Running test evals...');
|
|
127
|
+
const testResults = await this.runEvalsOnTarget(newTarget, testEvals, flags);
|
|
128
|
+
const testScore = this.computePassRate(testResults);
|
|
129
|
+
this.log(`Test pass rate: ${(testScore * 100).toFixed(1)}%`);
|
|
130
|
+
// Update optimization run
|
|
131
|
+
await apiPost('/api/optimization-runs', {
|
|
132
|
+
evalTargetId: newTarget.id,
|
|
133
|
+
type: 'content',
|
|
134
|
+
iteration,
|
|
135
|
+
inputVersion: target.version,
|
|
136
|
+
outputVersion: newTarget.version,
|
|
137
|
+
trainScore,
|
|
138
|
+
testScore,
|
|
139
|
+
analysis: {
|
|
140
|
+
suggestions: analysis.suggestions,
|
|
141
|
+
overallAssessment: analysis.overallAssessment,
|
|
142
|
+
},
|
|
143
|
+
status: 'completed',
|
|
144
|
+
});
|
|
145
|
+
// Track best
|
|
146
|
+
if (testScore > bestTestScore) {
|
|
147
|
+
bestTestScore = testScore;
|
|
148
|
+
bestVersion = newTarget.version;
|
|
149
|
+
noImprovementCount = 0;
|
|
150
|
+
}
|
|
151
|
+
else {
|
|
152
|
+
noImprovementCount++;
|
|
153
|
+
if (noImprovementCount >= 2) {
|
|
154
|
+
this.log('No improvement for 2 consecutive iterations. Stopping.');
|
|
155
|
+
break;
|
|
156
|
+
}
|
|
157
|
+
}
|
|
158
|
+
target = newTarget;
|
|
159
|
+
}
|
|
160
|
+
this.log(`\n=== Optimization Complete ===`);
|
|
161
|
+
this.log(`Best version: v${bestVersion} (test score: ${(bestTestScore * 100).toFixed(1)}%)`);
|
|
162
|
+
}
|
|
163
|
+
async runEvalsOnTarget(target, evalSet, flags) {
|
|
164
|
+
const results = [];
|
|
165
|
+
for (const eval_ of evalSet) {
|
|
166
|
+
const execResult = await executeEval({
|
|
167
|
+
prompt: eval_.prompt,
|
|
168
|
+
targetContent: target.content,
|
|
169
|
+
timeout: flags.timeout,
|
|
170
|
+
model: flags.model,
|
|
171
|
+
});
|
|
172
|
+
if (execResult.error) {
|
|
173
|
+
results.push({ evalName: eval_.name, prompt: eval_.prompt, expectations: eval_.expectations, grading: {} });
|
|
174
|
+
continue;
|
|
175
|
+
}
|
|
176
|
+
if (eval_.expectations.length > 0) {
|
|
177
|
+
const grading = await gradeEvalOutput({
|
|
178
|
+
prompt: eval_.prompt,
|
|
179
|
+
output: execResult.output,
|
|
180
|
+
expectations: eval_.expectations,
|
|
181
|
+
model: flags.model,
|
|
182
|
+
timeout: flags.timeout,
|
|
183
|
+
});
|
|
184
|
+
results.push({ evalName: eval_.name, prompt: eval_.prompt, expectations: eval_.expectations, grading: grading });
|
|
185
|
+
}
|
|
186
|
+
else {
|
|
187
|
+
results.push({ evalName: eval_.name, prompt: eval_.prompt, expectations: eval_.expectations, grading: {} });
|
|
188
|
+
}
|
|
189
|
+
}
|
|
190
|
+
return results;
|
|
191
|
+
}
|
|
192
|
+
computePassRate(results) {
|
|
193
|
+
const graded = results.filter((r) => Object.keys(r.grading).length > 0);
|
|
194
|
+
if (graded.length === 0)
|
|
195
|
+
return 0;
|
|
196
|
+
const rates = graded.map((r) => r.grading.summary?.pass_rate ?? 0);
|
|
197
|
+
return rates.reduce((a, b) => a + b, 0) / rates.length;
|
|
198
|
+
}
|
|
199
|
+
}
|
|
200
|
+
//# sourceMappingURL=optimize.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"optimize.js","sourceRoot":"","sources":["../../../src/commands/evals/optimize.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,OAAO,EAAE,MAAM,EAAE,MAAM,yBAAyB,CAAC;AAClE,OAAO,EAAE,WAAW,EAAE,MAAM,4BAA4B,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAC3D,OAAO,EAAE,kBAAkB,EAAE,MAAM,4BAA4B,CAAC;AAmBhE,MAAM,CAAC,OAAO,OAAO,mBAAoB,SAAQ,WAAW;IAC1D,MAAM,CAAU,WAAW,GAAG,yEAAyE,CAAC;IAExG,MAAM,CAAU,QAAQ,GAAG;QACzB,8CAA8C;QAC9C,wEAAwE;KACzE,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,OAAO,CAAC,EAAE,WAAW,EAAE,0DAA0D,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QAChH,gBAAgB,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,iCAAiC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;QAC/F,SAAS,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,gDAAgD,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;QAC3G,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,0BAA0B,EAAE,CAAC;QAChE,gBAAgB,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,gCAAgC,EAAE,CAAC;QACjF,eAAe,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,6BAA6B,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;QAC1F,QAAQ,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,oCAAoC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;QAC1F,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,mBAAmB,CAAC,CAAC;QAC9D,IAAI,CAAC,WAAW,EAAE,CAAC;QAEnB,IAAI,MAAM,GAAG,MAAM,MAAM,CAAa,qBAAqB,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;QAC5E,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAS,qBAAqB,IAAI,CAAC,QAAQ,QAAQ,CAAC,CAAC;QAElF,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC1B,IAAI,CAAC,KAAK,CAAC,mEAAmE,CAAC,CAAC;QAClF,CAAC;QAED,sCAAsC;QACtC,MAAM,QAAQ,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,CAAC,CAAC;QAC/D,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC;QAClD,MAAM,UAAU,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC;QAC/C,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QAE3C,IAAI,CAAC,GAAG,CAAC,eAAe,MAAM,CAAC,IAAI,MAAM,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3D,IAAI,CAAC,GAAG,CAAC,UAAU,UAAU,CAAC,MAAM,WAAW,SAAS,CAAC,MAAM,OAAO,CAAC,CAAC;QACxE,MAAM,SAAS,GAAG,UAAU,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAC9C,IAAI,CAAC,GAAG,CAAC,eAAe,KAAK,CAAC,gBAAgB,CAAC,0BAA0B,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QAC1G,IAAI,KAAK,CAAC,IAAI;YAAE,IAAI,CAAC,GAAG,CAAC,oCAAoC,CAAC,CAAC;QAC/D,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAEb,IAAI,WAAW,GAAG,MAAM,CAAC,OAAO,CAAC;QACjC,IAAI,aAAa,GAAG,CAAC,CAAC;QACtB,IAAI,kBAAkB,GAAG,CAAC,CAAC;QAE3B,KAAK,IAAI,SAAS,GAAG,CAAC,EAAE,SAAS,IAAI,KAAK,CAAC,gBAAgB,CAAC,EAAE,SAAS,EAAE,EAAE,CAAC;YAC1E,IAAI,CAAC,GAAG,CAAC,mBAAmB,SAAS,MAAM,CAAC,CAAC;YAE7C,qCAAqC;YACrC,MAAM,MAAM,GAAG,MAAM,OAAO,CAAiB,wBAAwB,EAAE;gBACrE,YAAY,EAAE,MAAM,CAAC,EAAE;gBACvB,IAAI,EAAE,SAAS;gBACf,SAAS;gBACT,YAAY,EAAE,MAAM,CAAC,OAAO;gBAC5B,MAAM,EAAE,SAAS;aAClB,CAAC,CAAC;YAEH,kBAAkB;YAClB,IAAI,CAAC,GAAG,CAAC,wBAAwB,CAAC,CAAC;YACnC,MAAM,YAAY,GAAG,MAAM,IAAI,CAAC,gBAAgB,CAAC,MAAM,EAAE,UAAU,EAAE,KAAK,CAAC,CAAC;YAC5E,MAAM,UAAU,GAAG,IAAI,CAAC,eAAe,CAAC,YAAY,CAAC,CAAC;YACtD,IAAI,CAAC,GAAG,CAAC,oBAAoB,CAAC,UAAU,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;YAE/D,mCAAmC;YACnC,IAAI,UAAU,IAAI,SAAS,EAAE,CAAC;gBAC5B,IAAI,CAAC,GAAG,CAAC,eAAe,CAAC,UAAU,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,kBAAkB,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC;gBACnH,iCAAiC;gBACjC,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,gBAAgB,CAAC,MAAM,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;gBAC1E,MAAM,SAAS,GAAG,IAAI,CAAC,eAAe,CAAC,WAAW,CAAC,CAAC;gBACpD,IAAI,CAAC,GAAG,CAAC,qBAAqB,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;gBAE/D,MAAM,OAAO,CAAC,wBAAwB,EAAE;oBACtC,YAAY,EAAE,MAAM,CAAC,EAAE;oBACvB,IAAI,EAAE,SAAS;oBACf,SAAS;oBACT,YAAY,EAAE,MAAM,CAAC,OAAO;oBAC5B,UAAU;oBACV,SAAS;oBACT,QAAQ,EAAE,EAAE,MAAM,EAAE,mBAAmB,EAAE;oBACzC,MAAM,EAAE,WAAW;iBACpB,CAAC,CAAC;gBACH,MAAM;YACR,CAAC;YAED,mCAAmC;YACnC,IAAI,CAAC,GAAG,CAAC,sBAAsB,CAAC,CAAC;YACjC,MAAM,WAAW,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBAC3C,QAAQ,EAAE,CAAC,CAAC,QAAQ;gBACpB,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,YAAY,EAAE,CAAC,CAAC,YAAY;gBAC5B,OAAO,EAAE,CAAC,CAAC,OAAO;aACnB,CAAC,CAAC,CAAC;YAEJ,MAAM,QAAQ,GAAG,MAAM,kBAAkB,CAAC;gBACxC,OAAO,EAAE,MAAM,CAAC,OAAO;gBACvB,WAAW;gBACX,KAAK,EAAE,KAAK,CAAC,gBAAgB,CAAC,IAAI,KAAK,CAAC,KAAK;gBAC7C,OAAO,EAAE,KAAK,CAAC,OAAO;aACvB,CAAC,CAAC;YAEH,IAAI,CAAC,GAAG,CAAC,iBAAiB,QAAQ,CAAC,iBAAiB,EAAE,CAAC,CAAC;YACxD,IAAI,CAAC,GAAG,CAAC,gBAAgB,QAAQ,CAAC,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC;YACxD,KAAK,MAAM,CAAC,IAAI,QAAQ,CAAC,WAAW,EAAE,CAAC;gBACrC,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,OAAO,KAAK,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;YAChD,CAAC;YAED,wCAAwC;YACxC,IAAI,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC;gBAChB,IAAI,CAAC,GAAG,CAAC,8EAA8E,CAAC,CAAC;gBACzF,IAAI,CAAC,GAAG,CAAC,iEAAiE,CAAC,CAAC;YAC9E,CAAC;YAED,IAAI,QAAQ,CAAC,eAAe,KAAK,MAAM,CAAC,OAAO,EAAE,CAAC;gBAChD,IAAI,CAAC,GAAG,CAAC,gCAAgC,CAAC,CAAC;gBAC3C,MAAM,OAAO,CAAC,wBAAwB,EAAE;oBACtC,YAAY,EAAE,MAAM,CAAC,EAAE;oBACvB,IAAI,EAAE,SAAS;oBACf,SAAS;oBACT,YAAY,EAAE,MAAM,CAAC,OAAO;oBAC5B,UAAU;oBACV,QAAQ,EAAE,EAAE,MAAM,EAAE,YAAY,EAAE;oBAClC,MAAM,EAAE,WAAW;iBACpB,CAAC,CAAC;gBACH,MAAM;YACR,CAAC;YAED,qBAAqB;YACrB,MAAM,SAAS,GAAG,MAAM,MAAM,CAAa,qBAAqB,MAAM,CAAC,EAAE,EAAE,EAAE;gBAC3E,OAAO,EAAE,QAAQ,CAAC,eAAe;aAClC,CAAC,CAAC;YACH,IAAI,CAAC,GAAG,CAAC,YAAY,SAAS,CAAC,OAAO,EAAE,CAAC,CAAC;YAE1C,gCAAgC;YAChC,IAAI,CAAC,GAAG,CAAC,uBAAuB,CAAC,CAAC;YAClC,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,gBAAgB,CAAC,SAAS,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;YAC7E,MAAM,SAAS,GAAG,IAAI,CAAC,eAAe,CAAC,WAAW,CAAC,CAAC;YACpD,IAAI,CAAC,GAAG,CAAC,mBAAmB,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;YAE7D,0BAA0B;YAC1B,MAAM,OAAO,CAAC,wBAAwB,EAAE;gBACtC,YAAY,EAAE,SAAS,CAAC,EAAE;gBAC1B,IAAI,EAAE,SAAS;gBACf,SAAS;gBACT,YAAY,EAAE,MAAM,CAAC,OAAO;gBAC5B,aAAa,EAAE,SAAS,CAAC,OAAO;gBAChC,UAAU;gBACV,SAAS;gBACT,QAAQ,EAAE;oBACR,WAAW,EAAE,QAAQ,CAAC,WAAW;oBACjC,iBAAiB,EAAE,QAAQ,CAAC,iBAAiB;iBAC9C;gBACD,MAAM,EAAE,WAAW;aACpB,CAAC,CAAC;YAEH,aAAa;YACb,IAAI,SAAS,GAAG,aAAa,EAAE,CAAC;gBAC9B,aAAa,GAAG,SAAS,CAAC;gBAC1B,WAAW,GAAG,SAAS,CAAC,OAAO,CAAC;gBAChC,kBAAkB,GAAG,CAAC,CAAC;YACzB,CAAC;iBAAM,CAAC;gBACN,kBAAkB,EAAE,CAAC;gBACrB,IAAI,kBAAkB,IAAI,CAAC,EAAE,CAAC;oBAC5B,IAAI,CAAC,GAAG,CAAC,wDAAwD,CAAC,CAAC;oBACnE,MAAM;gBACR,CAAC;YACH,CAAC;YAED,MAAM,GAAG,SAAS,CAAC;QACrB,CAAC;QAED,IAAI,CAAC,GAAG,CAAC,iCAAiC,CAAC,CAAC;QAC5C,IAAI,CAAC,GAAG,CAAC,kBAAkB,WAAW,iBAAiB,CAAC,aAAa,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IAC/F,CAAC;IAEO,KAAK,CAAC,gBAAgB,CAC5B,MAAkB,EAClB,OAAe,EACf,KAAqF;QAErF,MAAM,OAAO,GAA0G,EAAE,CAAC;QAE1H,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,MAAM,UAAU,GAAG,MAAM,WAAW,CAAC;gBACnC,MAAM,EAAE,KAAK,CAAC,MAAM;gBACpB,aAAa,EAAE,MAAM,CAAC,OAAO;gBAC7B,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,KAAK,EAAE,KAAK,CAAC,KAAK;aACnB,CAAC,CAAC;YAEH,IAAI,UAAU,CAAC,KAAK,EAAE,CAAC;gBACrB,OAAO,CAAC,IAAI,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,YAAY,EAAE,KAAK,CAAC,YAAY,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,CAAC;gBAC5G,SAAS;YACX,CAAC;YAED,IAAI,KAAK,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAClC,MAAM,OAAO,GAAG,MAAM,eAAe,CAAC;oBACpC,MAAM,EAAE,KAAK,CAAC,MAAM;oBACpB,MAAM,EAAE,UAAU,CAAC,MAAM;oBACzB,YAAY,EAAE,KAAK,CAAC,YAAY;oBAChC,KAAK,EAAE,KAAK,CAAC,KAAK;oBAClB,OAAO,EAAE,KAAK,CAAC,OAAO;iBACvB,CAAC,CAAC;gBACH,OAAO,CAAC,IAAI,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,YAAY,EAAE,KAAK,CAAC,YAAY,EAAE,OAAO,EAAE,OAA6C,EAAE,CAAC,CAAC;YACzJ,CAAC;iBAAM,CAAC;gBACN,OAAO,CAAC,IAAI,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,YAAY,EAAE,KAAK,CAAC,YAAY,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,CAAC;YAC9G,CAAC;QACH,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAEO,eAAe,CAAC,OAAoD;QAC1E,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACxE,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QAClC,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAE,CAAC,CAAC,OAAe,CAAC,OAAO,EAAE,SAAS,IAAI,CAAC,CAAC,CAAC;QAC5E,OAAO,KAAK,CAAC,MAAM,CAAC,CAAC,CAAS,EAAE,CAAS,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC;IACzE,CAAC"}
|
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
import { BaseCommand } from '../../base-command.js';
|
|
2
|
+
export default class EvalRunCommand 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
|
+
parallel: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
10
|
+
'runs-per-eval': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
11
|
+
iteration: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
12
|
+
'skip-baseline': import("@oclif/core/interfaces").BooleanFlag<boolean>;
|
|
13
|
+
'skip-grading': import("@oclif/core/interfaces").BooleanFlag<boolean>;
|
|
14
|
+
model: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
15
|
+
'grader-model': import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
|
|
16
|
+
timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
|
|
17
|
+
};
|
|
18
|
+
run(): Promise<void>;
|
|
19
|
+
private executeAndGrade;
|
|
20
|
+
}
|
|
21
|
+
//# sourceMappingURL=run.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/commands/evals/run.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAkCpD,MAAM,CAAC,OAAO,OAAO,cAAe,SAAQ,WAAW;IACrD,OAAgB,WAAW,SAA2E;IAEtG,OAAgB,QAAQ,WAItB;IAEF,OAAgB,IAAI;;MAElB;IAEF,OAAgB,KAAK;;;;;;;;;MASnB;IAEI,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC;YA4GZ,eAAe;CAuE9B"}
|