@hcu-lab.me/mcp-device 1.0.10 → 1.0.12
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/PRIVACY.md +21 -0
- package/README.md +57 -2
- package/SECURITY.md +67 -0
- package/config/revoked-approvals.json +1 -0
- package/control-plane/config/mcp-servers.example.toml +52 -0
- package/control-plane/config/mcp-servers.toml +52 -0
- package/control-plane/contracts/device-public-contract-v1.json +733 -0
- package/control-plane/scripts/GATEWAY-LICENSE.txt +7 -0
- package/control-plane/scripts/account-http.mjs +220 -0
- package/control-plane/scripts/account-store.mjs +368 -0
- package/control-plane/scripts/admin.mjs +68 -0
- package/control-plane/scripts/auth-session.mjs +411 -0
- package/control-plane/scripts/authenticated-mcp-wrapper.mjs +1281 -0
- package/control-plane/scripts/benchmark-command.mjs +9 -0
- package/control-plane/scripts/benchmark-mcp-catalog.mjs +174 -0
- package/control-plane/scripts/catalog-budget.mjs +87 -0
- package/control-plane/scripts/check-device-public-contract.mjs +20 -0
- package/control-plane/scripts/custom-tools/index.mjs +221 -0
- package/control-plane/scripts/custom-tools/path-utils.mjs +117 -0
- package/control-plane/scripts/custom-tools/response-utils.mjs +34 -0
- package/control-plane/scripts/dashboard-http.mjs +598 -0
- package/control-plane/scripts/device-access-policy.mjs +205 -0
- package/control-plane/scripts/device-admin.mjs +43 -0
- package/control-plane/scripts/device-audit.mjs +63 -0
- package/control-plane/scripts/device-broker.mjs +1461 -0
- package/control-plane/scripts/device-execution-tool-definitions.mjs +217 -0
- package/control-plane/scripts/device-inventory.mjs +128 -0
- package/control-plane/scripts/device-pairing-http.mjs +109 -0
- package/control-plane/scripts/device-pairing-store.mjs +245 -0
- package/control-plane/scripts/device-public-contract.mjs +165 -0
- package/control-plane/scripts/device-release.mjs +95 -0
- package/control-plane/scripts/device-secure-transport.mjs +128 -0
- package/control-plane/scripts/device-store.mjs +283 -0
- package/control-plane/scripts/device-usage.mjs +657 -0
- package/control-plane/scripts/embedded-device.mjs +149 -0
- package/control-plane/scripts/enduser-ui.mjs +45 -0
- package/control-plane/scripts/external-tool-broker.mjs +179 -0
- package/control-plane/scripts/oauth-authorization-server.mjs +589 -0
- package/control-plane/scripts/operator-dashboard-http.mjs +81 -0
- package/control-plane/scripts/project-inspection.mjs +227 -0
- package/control-plane/scripts/projects/project-id.mjs +69 -0
- package/control-plane/scripts/projects/trusted-roots-projects.mjs +375 -0
- package/control-plane/scripts/prompts/index.mjs +60 -0
- package/control-plane/scripts/remote-process-sessions.mjs +78 -0
- package/control-plane/scripts/remote-tool-result.mjs +9 -0
- package/control-plane/scripts/resources/index.mjs +166 -0
- package/control-plane/scripts/runtime-profile.mjs +54 -0
- package/control-plane/scripts/shell-tool-descriptor.mjs +17 -0
- package/control-plane/scripts/skill-source-safety.mjs +46 -0
- package/control-plane/scripts/skills/README.md +70 -0
- package/control-plane/scripts/skills/_upstream_licenses/anthropic-THIRD_PARTY_NOTICES.md +405 -0
- package/control-plane/scripts/skills/_upstream_licenses/hallmark-LICENSE.txt +21 -0
- package/control-plane/scripts/skills/_upstream_licenses/ponytail-LICENSE.txt +21 -0
- package/control-plane/scripts/skills/_upstream_licenses/stitch-skills-LICENSE.txt +201 -0
- package/control-plane/scripts/skills/_upstream_licenses/superpowers-LICENSE.txt +21 -0
- package/control-plane/scripts/skills/algorithmic-art/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/algorithmic-art/SKILL.md +405 -0
- package/control-plane/scripts/skills/algorithmic-art/templates/generator_template.js +223 -0
- package/control-plane/scripts/skills/algorithmic-art/templates/viewer.html +599 -0
- package/control-plane/scripts/skills/brainstorming/SKILL.md +285 -0
- package/control-plane/scripts/skills/brainstorming/scripts/frame-template.html +213 -0
- package/control-plane/scripts/skills/brainstorming/scripts/helper.js +167 -0
- package/control-plane/scripts/skills/brainstorming/scripts/server.cjs +723 -0
- package/control-plane/scripts/skills/brainstorming/scripts/start-server.sh +209 -0
- package/control-plane/scripts/skills/brainstorming/scripts/stop-server.sh +120 -0
- package/control-plane/scripts/skills/brainstorming/spec-document-reviewer-prompt.md +49 -0
- package/control-plane/scripts/skills/brainstorming/visual-companion.md +299 -0
- package/control-plane/scripts/skills/brand-guidelines/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/brand-guidelines/SKILL.md +73 -0
- package/control-plane/scripts/skills/claude-api/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/claude-api/SKILL.md +570 -0
- package/control-plane/scripts/skills/claude-api/csharp/claude-api/README.md +362 -0
- package/control-plane/scripts/skills/claude-api/csharp/claude-api/batches.md +14 -0
- package/control-plane/scripts/skills/claude-api/csharp/claude-api/files-api.md +25 -0
- package/control-plane/scripts/skills/claude-api/csharp/claude-api/streaming.md +28 -0
- package/control-plane/scripts/skills/claude-api/csharp/claude-api/tool-use.md +164 -0
- package/control-plane/scripts/skills/claude-api/curl/examples.md +256 -0
- package/control-plane/scripts/skills/claude-api/curl/managed-agents.md +371 -0
- package/control-plane/scripts/skills/claude-api/go/claude-api/README.md +188 -0
- package/control-plane/scripts/skills/claude-api/go/claude-api/files-api.md +23 -0
- package/control-plane/scripts/skills/claude-api/go/claude-api/streaming.md +43 -0
- package/control-plane/scripts/skills/claude-api/go/claude-api/tool-use.md +220 -0
- package/control-plane/scripts/skills/claude-api/go/managed-agents/README.md +564 -0
- package/control-plane/scripts/skills/claude-api/java/claude-api/README.md +238 -0
- package/control-plane/scripts/skills/claude-api/java/claude-api/files-api.md +27 -0
- package/control-plane/scripts/skills/claude-api/java/claude-api/streaming.md +24 -0
- package/control-plane/scripts/skills/claude-api/java/claude-api/tool-use.md +228 -0
- package/control-plane/scripts/skills/claude-api/java/managed-agents/README.md +443 -0
- package/control-plane/scripts/skills/claude-api/php/claude-api/README.md +174 -0
- package/control-plane/scripts/skills/claude-api/php/claude-api/batches.md +15 -0
- package/control-plane/scripts/skills/claude-api/php/claude-api/files-api.md +13 -0
- package/control-plane/scripts/skills/claude-api/php/claude-api/streaming.md +27 -0
- package/control-plane/scripts/skills/claude-api/php/claude-api/tool-use.md +253 -0
- package/control-plane/scripts/skills/claude-api/php/managed-agents/README.md +443 -0
- package/control-plane/scripts/skills/claude-api/python/claude-api/README.md +565 -0
- package/control-plane/scripts/skills/claude-api/python/claude-api/batches.md +198 -0
- package/control-plane/scripts/skills/claude-api/python/claude-api/files-api.md +170 -0
- package/control-plane/scripts/skills/claude-api/python/claude-api/sdk-upgrade.md +286 -0
- package/control-plane/scripts/skills/claude-api/python/claude-api/streaming.md +179 -0
- package/control-plane/scripts/skills/claude-api/python/claude-api/tool-use.md +627 -0
- package/control-plane/scripts/skills/claude-api/python/managed-agents/README.md +335 -0
- package/control-plane/scripts/skills/claude-api/ruby/claude-api/README.md +134 -0
- package/control-plane/scripts/skills/claude-api/ruby/claude-api/streaming.md +16 -0
- package/control-plane/scripts/skills/claude-api/ruby/claude-api/tool-use.md +41 -0
- package/control-plane/scripts/skills/claude-api/ruby/managed-agents/README.md +394 -0
- package/control-plane/scripts/skills/claude-api/shared/admin-api.md +179 -0
- package/control-plane/scripts/skills/claude-api/shared/agent-design.md +101 -0
- package/control-plane/scripts/skills/claude-api/shared/anthropic-cli.md +269 -0
- package/control-plane/scripts/skills/claude-api/shared/claude-platform-on-aws.md +59 -0
- package/control-plane/scripts/skills/claude-api/shared/cost-optimization.md +233 -0
- package/control-plane/scripts/skills/claude-api/shared/error-codes.md +262 -0
- package/control-plane/scripts/skills/claude-api/shared/live-sources.md +166 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-api-reference.md +455 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-client-patterns.md +220 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-core.md +359 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-environments.md +226 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-events.md +290 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-memory.md +201 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-multiagent.md +260 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-onboarding.md +82 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-outcomes.md +108 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-overview.md +77 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-scheduled-deployments.md +156 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-self-hosted-sandboxes.md +282 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-tools.md +504 -0
- package/control-plane/scripts/skills/claude-api/shared/managed-agents-webhooks.md +143 -0
- package/control-plane/scripts/skills/claude-api/shared/model-migration.md +1872 -0
- package/control-plane/scripts/skills/claude-api/shared/models.md +142 -0
- package/control-plane/scripts/skills/claude-api/shared/platform-availability.md +52 -0
- package/control-plane/scripts/skills/claude-api/shared/prompt-audit.md +223 -0
- package/control-plane/scripts/skills/claude-api/shared/prompt-caching.md +292 -0
- package/control-plane/scripts/skills/claude-api/shared/token-counting.md +56 -0
- package/control-plane/scripts/skills/claude-api/shared/tool-use-concepts.md +526 -0
- package/control-plane/scripts/skills/claude-api/typescript/claude-api/README.md +405 -0
- package/control-plane/scripts/skills/claude-api/typescript/claude-api/batches.md +106 -0
- package/control-plane/scripts/skills/claude-api/typescript/claude-api/files-api.md +98 -0
- package/control-plane/scripts/skills/claude-api/typescript/claude-api/streaming.md +178 -0
- package/control-plane/scripts/skills/claude-api/typescript/claude-api/tool-use.md +586 -0
- package/control-plane/scripts/skills/claude-api/typescript/managed-agents/README.md +359 -0
- package/control-plane/scripts/skills/dispatching-parallel-agents/SKILL.md +167 -0
- package/control-plane/scripts/skills/enhance-prompt/README.md +34 -0
- package/control-plane/scripts/skills/enhance-prompt/SKILL.md +204 -0
- package/control-plane/scripts/skills/enhance-prompt/references/KEYWORDS.md +114 -0
- package/control-plane/scripts/skills/executing-plans/SKILL.md +371 -0
- package/control-plane/scripts/skills/executing-plans/scripts/task-done +52 -0
- package/control-plane/scripts/skills/executing-plans/scripts/task-start +28 -0
- package/control-plane/scripts/skills/extract-design-md/SKILL.md +358 -0
- package/control-plane/scripts/skills/extract-design-md/examples/DESIGN.md +210 -0
- package/control-plane/scripts/skills/extract-design-md/references/angular.md +122 -0
- package/control-plane/scripts/skills/extract-design-md/references/plain-css.md +161 -0
- package/control-plane/scripts/skills/extract-design-md/references/react-tailwind.md +157 -0
- package/control-plane/scripts/skills/extract-design-md/references/svelte.md +101 -0
- package/control-plane/scripts/skills/extract-design-md/references/vue.md +107 -0
- package/control-plane/scripts/skills/finishing-a-development-branch/SKILL.md +225 -0
- package/control-plane/scripts/skills/frontend-design/LICENSE.txt +177 -0
- package/control-plane/scripts/skills/frontend-design/SKILL.md +71 -0
- package/control-plane/scripts/skills/hallmark/SKILL.md +558 -0
- package/control-plane/scripts/skills/hallmark/references/anti-patterns.md +418 -0
- package/control-plane/scripts/skills/hallmark/references/assets.md +406 -0
- package/control-plane/scripts/skills/hallmark/references/color.md +95 -0
- package/control-plane/scripts/skills/hallmark/references/component-cookbook.md +263 -0
- package/control-plane/scripts/skills/hallmark/references/components/c1-outlined-chip.md +12 -0
- package/control-plane/scripts/skills/hallmark/references/components/c2-inline-form-as-cta.md +16 -0
- package/control-plane/scripts/skills/hallmark/references/components/c3-typographic-link.md +8 -0
- package/control-plane/scripts/skills/hallmark/references/components/c4-sticky-bottom-bar.md +16 -0
- package/control-plane/scripts/skills/hallmark/references/components/f1-bento-grid.md +20 -0
- package/control-plane/scripts/skills/hallmark/references/components/f2-sticky-scroll-stack.md +20 -0
- package/control-plane/scripts/skills/hallmark/references/components/f3-tabular-spec-sheet.md +11 -0
- package/control-plane/scripts/skills/hallmark/references/components/f4-step-sequence.md +11 -0
- package/control-plane/scripts/skills/hallmark/references/components/f5-annotated-screenshot.md +11 -0
- package/control-plane/scripts/skills/hallmark/references/components/f6-product-card-grid.md +41 -0
- package/control-plane/scripts/skills/hallmark/references/components/ft1-mast-headed.md +13 -0
- package/control-plane/scripts/skills/hallmark/references/components/ft2-inline-rule-single-line.md +10 -0
- package/control-plane/scripts/skills/hallmark/references/components/ft3-index-style-category-list.md +12 -0
- package/control-plane/scripts/skills/hallmark/references/components/ft4-dense-typographic.md +10 -0
- package/control-plane/scripts/skills/hallmark/references/components/ft5-statement.md +21 -0
- package/control-plane/scripts/skills/hallmark/references/components/ft6-letter-close.md +19 -0
- package/control-plane/scripts/skills/hallmark/references/components/ft7-newsletter-first.md +27 -0
- package/control-plane/scripts/skills/hallmark/references/components/ft8-marquee-scroll.md +25 -0
- package/control-plane/scripts/skills/hallmark/references/components/h1-marquee.md +15 -0
- package/control-plane/scripts/skills/hallmark/references/components/h2-split-diptych.md +15 -0
- package/control-plane/scripts/skills/hallmark/references/components/h3-quote-led.md +11 -0
- package/control-plane/scripts/skills/hallmark/references/components/h4-stat-led.md +14 -0
- package/control-plane/scripts/skills/hallmark/references/components/h5-letter-hero.md +11 -0
- package/control-plane/scripts/skills/hallmark/references/components/h6-photographic-fold.md +16 -0
- package/control-plane/scripts/skills/hallmark/references/components/h7-demo-video-clipped-by-viewport-edge.md +27 -0
- package/control-plane/scripts/skills/hallmark/references/components/h8-mockup-split-browser-framed.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/components/h9-custom-illustration-centerpiece.md +27 -0
- package/control-plane/scripts/skills/hallmark/references/components/n1-wordmark-2-links.md +12 -0
- package/control-plane/scripts/skills/hallmark/references/components/n10-floating-on-scroll-morph.md +19 -0
- package/control-plane/scripts/skills/hallmark/references/components/n11-mega-menu.md +40 -0
- package/control-plane/scripts/skills/hallmark/references/components/n12-banner-retract.md +34 -0
- package/control-plane/scripts/skills/hallmark/references/components/n13-inline-cmdk-pill.md +39 -0
- package/control-plane/scripts/skills/hallmark/references/components/n1b-saas-three-section.md +35 -0
- package/control-plane/scripts/skills/hallmark/references/components/n2-floating-chip.md +14 -0
- package/control-plane/scripts/skills/hallmark/references/components/n3-side-rail.md +14 -0
- package/control-plane/scripts/skills/hallmark/references/components/n4-hidden-behind-k.md +9 -0
- package/control-plane/scripts/skills/hallmark/references/components/n5-floating-pill.md +28 -0
- package/control-plane/scripts/skills/hallmark/references/components/n6-newspaper-masthead.md +24 -0
- package/control-plane/scripts/skills/hallmark/references/components/n7-brutal-slab.md +22 -0
- package/control-plane/scripts/skills/hallmark/references/components/n8-terminal-command.md +21 -0
- package/control-plane/scripts/skills/hallmark/references/components/n9-edge-aligned-minimal.md +17 -0
- package/control-plane/scripts/skills/hallmark/references/components/s1-left-margin-numbered.md +15 -0
- package/control-plane/scripts/skills/hallmark/references/components/s2-hanging.md +13 -0
- package/control-plane/scripts/skills/hallmark/references/components/s3-sticky-pinned.md +19 -0
- package/control-plane/scripts/skills/hallmark/references/components/s4-inline-no-break.md +11 -0
- package/control-plane/scripts/skills/hallmark/references/components/s5-bottom-anchored.md +13 -0
- package/control-plane/scripts/skills/hallmark/references/components/t1-pull-quote-with-marginalia.md +12 -0
- package/control-plane/scripts/skills/hallmark/references/components/t2-logo-wall-hairline.md +19 -0
- package/control-plane/scripts/skills/hallmark/references/components/t3-single-huge-quote.md +11 -0
- package/control-plane/scripts/skills/hallmark/references/components/t4-numbered-stat-strip.md +14 -0
- package/control-plane/scripts/skills/hallmark/references/contract.md +25 -0
- package/control-plane/scripts/skills/hallmark/references/copy.md +182 -0
- package/control-plane/scripts/skills/hallmark/references/custom-craft.md +626 -0
- package/control-plane/scripts/skills/hallmark/references/custom-theme.md +367 -0
- package/control-plane/scripts/skills/hallmark/references/design-md.md +116 -0
- package/control-plane/scripts/skills/hallmark/references/export-formats.md +329 -0
- package/control-plane/scripts/skills/hallmark/references/floating-nav.md +89 -0
- package/control-plane/scripts/skills/hallmark/references/genres/atmospheric.md +69 -0
- package/control-plane/scripts/skills/hallmark/references/genres/editorial.md +72 -0
- package/control-plane/scripts/skills/hallmark/references/genres/modern-minimal.md +71 -0
- package/control-plane/scripts/skills/hallmark/references/genres/playful.md +70 -0
- package/control-plane/scripts/skills/hallmark/references/hero-enrichment.md +475 -0
- package/control-plane/scripts/skills/hallmark/references/imagery-kit.md +170 -0
- package/control-plane/scripts/skills/hallmark/references/interaction-and-states.md +208 -0
- package/control-plane/scripts/skills/hallmark/references/layout-and-space.md +112 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/01-bento-grid.md +35 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/02-long-document.md +34 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/03-marquee-hero.md +31 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/04-stat-led.md +34 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/05-workbench.md +32 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/06-conversational-faq.md +33 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/07-manifesto.md +32 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/08-photographic.md +34 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/09-quote-led.md +32 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/10-specimen.md +32 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/11-catalogue.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/12-letter.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/13-index-first.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/14-narrative-workflow.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/15-split-studio.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/16-feature-stack.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/17-type-specimen.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/18-portfolio-grid.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/19-map-diagram.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/20-ecosystem-index.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures/21-component-playground.md +23 -0
- package/control-plane/scripts/skills/hallmark/references/macrostructures.md +89 -0
- package/control-plane/scripts/skills/hallmark/references/microinteractions.md +258 -0
- package/control-plane/scripts/skills/hallmark/references/motion.md +109 -0
- package/control-plane/scripts/skills/hallmark/references/preview-examples.md +49 -0
- package/control-plane/scripts/skills/hallmark/references/responsive.md +138 -0
- package/control-plane/scripts/skills/hallmark/references/slop-test.md +192 -0
- package/control-plane/scripts/skills/hallmark/references/structure.md +160 -0
- package/control-plane/scripts/skills/hallmark/references/study.md +509 -0
- package/control-plane/scripts/skills/hallmark/references/themes/carnival.md +301 -0
- package/control-plane/scripts/skills/hallmark/references/themes/cobalt.md +146 -0
- package/control-plane/scripts/skills/hallmark/references/themes/grid.md +161 -0
- package/control-plane/scripts/skills/hallmark/references/themes/hum.md +403 -0
- package/control-plane/scripts/skills/hallmark/references/themes/lumen.md +478 -0
- package/control-plane/scripts/skills/hallmark/references/typography.md +243 -0
- package/control-plane/scripts/skills/hallmark/references/upstream/recipes.md +186 -0
- package/control-plane/scripts/skills/hallmark/references/upstream/study-examples.md +176 -0
- package/control-plane/scripts/skills/hallmark/references/upstream/tokens.css +1221 -0
- package/control-plane/scripts/skills/hallmark/references/verbs/audit.md +25 -0
- package/control-plane/scripts/skills/hallmark/references/verbs/redesign.md +269 -0
- package/control-plane/scripts/skills/human-comms/SKILL.md +29 -0
- package/control-plane/scripts/skills/index.mjs +432 -0
- package/control-plane/scripts/skills/internal-comms/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/internal-comms/SKILL.md +32 -0
- package/control-plane/scripts/skills/internal-comms/examples/3p-updates.md +47 -0
- package/control-plane/scripts/skills/internal-comms/examples/company-newsletter.md +65 -0
- package/control-plane/scripts/skills/internal-comms/examples/faq-answers.md +30 -0
- package/control-plane/scripts/skills/internal-comms/examples/general-comms.md +16 -0
- package/control-plane/scripts/skills/mcp-adapter.mjs +94 -0
- package/control-plane/scripts/skills/mcp-builder/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/mcp-builder/SKILL.md +236 -0
- package/control-plane/scripts/skills/mcp-builder/reference/evaluation.md +602 -0
- package/control-plane/scripts/skills/mcp-builder/reference/mcp_best_practices.md +249 -0
- package/control-plane/scripts/skills/mcp-builder/reference/node_mcp_server.md +970 -0
- package/control-plane/scripts/skills/mcp-builder/reference/python_mcp_server.md +719 -0
- package/control-plane/scripts/skills/mcp-builder/scripts/connections.py +151 -0
- package/control-plane/scripts/skills/mcp-builder/scripts/evaluation.py +373 -0
- package/control-plane/scripts/skills/mcp-builder/scripts/example_evaluation.xml +22 -0
- package/control-plane/scripts/skills/mcp-builder/scripts/requirements.txt +2 -0
- package/control-plane/scripts/skills/ponytail/SKILL.md +120 -0
- package/control-plane/scripts/skills/ponytail-audit/SKILL.md +41 -0
- package/control-plane/scripts/skills/ponytail-debt/SKILL.md +44 -0
- package/control-plane/scripts/skills/ponytail-gain/SKILL.md +50 -0
- package/control-plane/scripts/skills/ponytail-help/SKILL.md +71 -0
- package/control-plane/scripts/skills/ponytail-review/SKILL.md +57 -0
- package/control-plane/scripts/skills/receiving-code-review/SKILL.md +205 -0
- package/control-plane/scripts/skills/requesting-code-review/SKILL.md +95 -0
- package/control-plane/scripts/skills/requesting-code-review/code-reviewer.md +198 -0
- package/control-plane/scripts/skills/skill-creator/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/skill-creator/SKILL.md +485 -0
- package/control-plane/scripts/skills/skill-creator/agents/analyzer.md +274 -0
- package/control-plane/scripts/skills/skill-creator/agents/comparator.md +202 -0
- package/control-plane/scripts/skills/skill-creator/agents/grader.md +223 -0
- package/control-plane/scripts/skills/skill-creator/assets/eval_review.html +146 -0
- package/control-plane/scripts/skills/skill-creator/eval-viewer/generate_review.py +471 -0
- package/control-plane/scripts/skills/skill-creator/eval-viewer/viewer.html +1325 -0
- package/control-plane/scripts/skills/skill-creator/references/schemas.md +430 -0
- package/control-plane/scripts/skills/skill-creator/scripts/__init__.py +0 -0
- package/control-plane/scripts/skills/skill-creator/scripts/aggregate_benchmark.py +401 -0
- package/control-plane/scripts/skills/skill-creator/scripts/generate_report.py +326 -0
- package/control-plane/scripts/skills/skill-creator/scripts/improve_description.py +247 -0
- package/control-plane/scripts/skills/skill-creator/scripts/package_skill.py +136 -0
- package/control-plane/scripts/skills/skill-creator/scripts/quick_validate.py +103 -0
- package/control-plane/scripts/skills/skill-creator/scripts/run_eval.py +310 -0
- package/control-plane/scripts/skills/skill-creator/scripts/run_loop.py +328 -0
- package/control-plane/scripts/skills/skill-creator/scripts/utils.py +47 -0
- package/control-plane/scripts/skills/slack-gif-creator/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/slack-gif-creator/SKILL.md +254 -0
- package/control-plane/scripts/skills/slack-gif-creator/core/easing.py +234 -0
- package/control-plane/scripts/skills/slack-gif-creator/core/frame_composer.py +176 -0
- package/control-plane/scripts/skills/slack-gif-creator/core/gif_builder.py +269 -0
- package/control-plane/scripts/skills/slack-gif-creator/core/validators.py +136 -0
- package/control-plane/scripts/skills/slack-gif-creator/requirements.txt +4 -0
- package/control-plane/scripts/skills/sources.json +401 -0
- package/control-plane/scripts/skills/sources.lock.json +416 -0
- package/control-plane/scripts/skills/subagent-driven-development/SKILL.md +570 -0
- package/control-plane/scripts/skills/subagent-driven-development/implementer-prompt.md +154 -0
- package/control-plane/scripts/skills/subagent-driven-development/re-review-prompt.md +115 -0
- package/control-plane/scripts/skills/subagent-driven-development/scripts/review-package +53 -0
- package/control-plane/scripts/skills/subagent-driven-development/scripts/sdd-workspace +82 -0
- package/control-plane/scripts/skills/subagent-driven-development/scripts/task-brief +43 -0
- package/control-plane/scripts/skills/subagent-driven-development/task-reviewer-prompt.md +207 -0
- package/control-plane/scripts/skills/systematic-debugging/CREATION-LOG.md +119 -0
- package/control-plane/scripts/skills/systematic-debugging/SKILL.md +283 -0
- package/control-plane/scripts/skills/systematic-debugging/condition-based-waiting-example.ts +158 -0
- package/control-plane/scripts/skills/systematic-debugging/condition-based-waiting.md +115 -0
- package/control-plane/scripts/skills/systematic-debugging/defense-in-depth.md +122 -0
- package/control-plane/scripts/skills/systematic-debugging/find-polluter.sh +72 -0
- package/control-plane/scripts/skills/systematic-debugging/root-cause-tracing.md +169 -0
- package/control-plane/scripts/skills/systematic-debugging/test-academic.md +14 -0
- package/control-plane/scripts/skills/systematic-debugging/test-pressure-1.md +58 -0
- package/control-plane/scripts/skills/systematic-debugging/test-pressure-2.md +68 -0
- package/control-plane/scripts/skills/systematic-debugging/test-pressure-3.md +69 -0
- package/control-plane/scripts/skills/test-driven-development/SKILL.md +330 -0
- package/control-plane/scripts/skills/test-driven-development/writing-good-tests.md +198 -0
- package/control-plane/scripts/skills/theme-factory/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/theme-factory/SKILL.md +59 -0
- package/control-plane/scripts/skills/theme-factory/theme-showcase.pdf +0 -0
- package/control-plane/scripts/skills/theme-factory/themes/arctic-frost.md +19 -0
- package/control-plane/scripts/skills/theme-factory/themes/botanical-garden.md +19 -0
- package/control-plane/scripts/skills/theme-factory/themes/desert-rose.md +19 -0
- package/control-plane/scripts/skills/theme-factory/themes/forest-canopy.md +19 -0
- package/control-plane/scripts/skills/theme-factory/themes/golden-hour.md +19 -0
- package/control-plane/scripts/skills/theme-factory/themes/midnight-galaxy.md +19 -0
- package/control-plane/scripts/skills/theme-factory/themes/modern-minimalist.md +19 -0
- package/control-plane/scripts/skills/theme-factory/themes/ocean-depths.md +19 -0
- package/control-plane/scripts/skills/theme-factory/themes/sunset-boulevard.md +19 -0
- package/control-plane/scripts/skills/theme-factory/themes/tech-innovation.md +19 -0
- package/control-plane/scripts/skills/using-git-worktrees/SKILL.md +167 -0
- package/control-plane/scripts/skills/verification-before-completion/SKILL.md +120 -0
- package/control-plane/scripts/skills/web-artifacts-builder/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/web-artifacts-builder/SKILL.md +74 -0
- package/control-plane/scripts/skills/web-artifacts-builder/scripts/bundle-artifact.sh +54 -0
- package/control-plane/scripts/skills/web-artifacts-builder/scripts/init-artifact.sh +322 -0
- package/control-plane/scripts/skills/web-artifacts-builder/scripts/shadcn-components.tar.gz +0 -0
- package/control-plane/scripts/skills/webapp-testing/LICENSE.txt +202 -0
- package/control-plane/scripts/skills/webapp-testing/SKILL.md +96 -0
- package/control-plane/scripts/skills/webapp-testing/examples/console_logging.py +35 -0
- package/control-plane/scripts/skills/webapp-testing/examples/element_discovery.py +40 -0
- package/control-plane/scripts/skills/webapp-testing/examples/static_html_automation.py +33 -0
- package/control-plane/scripts/skills/webapp-testing/scripts/with_server.py +106 -0
- package/control-plane/scripts/skills/writing-plans/SKILL.md +192 -0
- package/control-plane/scripts/skills/writing-plans/plan-document-reviewer-prompt.md +49 -0
- package/control-plane/scripts/skills/writing-skills/SKILL.md +681 -0
- package/control-plane/scripts/skills/writing-skills/anthropic-best-practices.md +1150 -0
- package/control-plane/scripts/skills/writing-skills/examples/CLAUDE_MD_TESTING.md +189 -0
- package/control-plane/scripts/skills/writing-skills/graphviz-conventions.dot +172 -0
- package/control-plane/scripts/skills/writing-skills/persuasion-principles.md +187 -0
- package/control-plane/scripts/skills/writing-skills/render-graphs.js +169 -0
- package/control-plane/scripts/skills/writing-skills/testing-skills-with-subagents.md +384 -0
- package/control-plane/scripts/smoke-mcp-schemas.mjs +30 -0
- package/control-plane/scripts/surface-config.mjs +45 -0
- package/control-plane/scripts/sync-skills.mjs +370 -0
- package/control-plane/scripts/tool-metrics.mjs +74 -0
- package/control-plane/scripts/tool-risk.mjs +156 -0
- package/control-plane/scripts/tool-surface-stability.mjs +38 -0
- package/control-plane/scripts/upstreams/catalog-cache.mjs +28 -0
- package/control-plane/scripts/upstreams/config.mjs +158 -0
- package/control-plane/scripts/upstreams/diagnostics.mjs +21 -0
- package/control-plane/scripts/upstreams/http-client.mjs +67 -0
- package/control-plane/scripts/upstreams/manager.mjs +535 -0
- package/control-plane/scripts/upstreams/names.mjs +55 -0
- package/control-plane/scripts/upstreams/presets.mjs +60 -0
- package/control-plane/scripts/upstreams/resource-uri.mjs +29 -0
- package/control-plane/scripts/upstreams/stdio-client.mjs +73 -0
- package/control-plane/scripts/workspace-registry.mjs +147 -0
- package/dist/context/authorization.d.ts +5 -0
- package/dist/context/authorization.js +70 -0
- package/dist/context/capture.d.ts +46 -0
- package/dist/context/capture.js +187 -0
- package/dist/context/cli.d.ts +47 -0
- package/dist/context/cli.js +287 -0
- package/dist/context/device-adapter.d.ts +102 -0
- package/dist/context/device-adapter.js +79 -0
- package/dist/context/gateway.d.ts +354 -0
- package/dist/context/gateway.js +50 -0
- package/dist/context/graph.d.ts +63 -0
- package/dist/context/graph.js +214 -0
- package/dist/context/importer.d.ts +8 -0
- package/dist/context/importer.js +69 -0
- package/dist/context/indexer.d.ts +14 -0
- package/dist/context/indexer.js +57 -0
- package/dist/context/migrations/index.d.ts +4 -0
- package/dist/context/migrations/index.js +87 -0
- package/dist/context/redact.d.ts +15 -0
- package/dist/context/redact.js +43 -0
- package/dist/context/repositories.d.ts +9 -0
- package/dist/context/repositories.js +45 -0
- package/dist/context/search.d.ts +63 -0
- package/dist/context/search.js +207 -0
- package/dist/context/service.d.ts +278 -0
- package/dist/context/service.js +160 -0
- package/dist/context/store.d.ts +134 -0
- package/dist/context/store.js +722 -0
- package/dist/context/text.d.ts +4 -0
- package/dist/context/text.js +14 -0
- package/dist/context/tool-contract.d.ts +534 -0
- package/dist/context/tool-contract.js +45 -0
- package/dist/device/core.d.ts +29 -0
- package/dist/device/core.js +66 -0
- package/dist/device/device.d.ts +5 -1
- package/dist/device/device.js +87 -39
- package/dist/device/gateway-channel.d.ts +8 -1
- package/dist/device/gateway-channel.js +51 -3
- package/dist/device/gateway-tool-adapter.d.ts +9 -0
- package/dist/device/gateway-tool-adapter.js +20 -1
- package/dist/device/official-trust.d.ts +12 -0
- package/dist/device/official-trust.js +25 -1
- package/dist/device/runtime-owner.js +9 -1
- package/dist/distribution.d.ts +4 -0
- package/dist/distribution.js +8 -0
- package/dist/log-palace/cli.d.ts +7 -0
- package/dist/log-palace/cli.js +89 -0
- package/dist/log-palace/formats.d.ts +26 -0
- package/dist/log-palace/formats.js +126 -0
- package/dist/log-palace/store.d.ts +129 -0
- package/dist/log-palace/store.js +856 -0
- package/dist/mcp-device.js +27 -5
- package/dist/npm-scripts/remote.js +1 -0
- package/dist/platform/activity-journal.d.ts +16 -0
- package/dist/platform/activity-journal.js +68 -0
- package/dist/platform/cli.d.ts +1 -0
- package/dist/platform/cli.js +300 -0
- package/dist/platform/coordination-lock.d.ts +45 -0
- package/dist/platform/coordination-lock.js +233 -0
- package/dist/platform/device-runtime.d.ts +16 -0
- package/dist/platform/device-runtime.js +83 -0
- package/dist/platform/knowledge-contract.d.ts +46 -0
- package/dist/platform/knowledge-contract.js +1 -0
- package/dist/platform/knowledge-mcp.d.ts +29 -0
- package/dist/platform/knowledge-mcp.js +95 -0
- package/dist/platform/knowledge-router.d.ts +64 -0
- package/dist/platform/knowledge-router.js +320 -0
- package/dist/platform/plugin-child.d.ts +1 -0
- package/dist/platform/plugin-child.js +115 -0
- package/dist/platform/plugin-contract.d.ts +76 -0
- package/dist/platform/plugin-contract.js +53 -0
- package/dist/platform/plugin-driver.d.ts +8 -0
- package/dist/platform/plugin-driver.js +443 -0
- package/dist/platform/plugin-host.d.ts +88 -0
- package/dist/platform/plugin-host.js +817 -0
- package/dist/platform/plugin-policy.d.ts +3 -0
- package/dist/platform/plugin-policy.js +29 -0
- package/dist/platform/plugin-services.d.ts +14 -0
- package/dist/platform/plugin-services.js +32 -0
- package/dist/platform/process-identity.d.ts +9 -0
- package/dist/platform/process-identity.js +68 -0
- package/dist/platform/runtime-lease.d.ts +4 -0
- package/dist/platform/runtime-lease.js +58 -0
- package/dist/platform/runtime-root.d.ts +9 -0
- package/dist/platform/runtime-root.js +61 -0
- package/dist/platform/state-backup.d.ts +92 -0
- package/dist/platform/state-backup.js +399 -0
- package/dist/plugins/reference/plugin.d.ts +3 -0
- package/dist/plugins/reference/plugin.js +13 -0
- package/dist/server.js +5 -1
- package/dist/terminal-manager.d.ts +7 -0
- package/dist/terminal-manager.js +10 -0
- package/dist/tool-dispatcher.js +5 -1
- package/dist/tools/improved-process-tools.js +4 -0
- package/dist/types.d.ts +1 -0
- package/dist/version.d.ts +1 -1
- package/dist/version.js +1 -1
- package/package.json +29 -6
- package/skills/mcp-device-context/SKILL.md +40 -0
|
@@ -0,0 +1,146 @@
|
|
|
1
|
+
<!DOCTYPE html>
|
|
2
|
+
<html lang="en">
|
|
3
|
+
<head>
|
|
4
|
+
<meta charset="UTF-8">
|
|
5
|
+
<meta name="viewport" content="width=device-width, initial-scale=1.0">
|
|
6
|
+
<title>Eval Set Review - __SKILL_NAME_PLACEHOLDER__</title>
|
|
7
|
+
<link rel="preconnect" href="https://fonts.googleapis.com">
|
|
8
|
+
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
|
|
9
|
+
<link href="https://fonts.googleapis.com/css2?family=Poppins:wght@500;600&family=Lora:wght@400;500&display=swap" rel="stylesheet">
|
|
10
|
+
<style>
|
|
11
|
+
* { box-sizing: border-box; margin: 0; padding: 0; }
|
|
12
|
+
body { font-family: 'Lora', Georgia, serif; background: #faf9f5; padding: 2rem; color: #141413; }
|
|
13
|
+
h1 { font-family: 'Poppins', sans-serif; margin-bottom: 0.5rem; font-size: 1.5rem; }
|
|
14
|
+
.description { color: #b0aea5; margin-bottom: 1.5rem; font-style: italic; max-width: 900px; }
|
|
15
|
+
.controls { margin-bottom: 1rem; display: flex; gap: 0.5rem; }
|
|
16
|
+
.btn { font-family: 'Poppins', sans-serif; padding: 0.5rem 1rem; border: none; border-radius: 6px; cursor: pointer; font-size: 0.875rem; font-weight: 500; }
|
|
17
|
+
.btn-add { background: #6a9bcc; color: white; }
|
|
18
|
+
.btn-add:hover { background: #5889b8; }
|
|
19
|
+
.btn-export { background: #d97757; color: white; }
|
|
20
|
+
.btn-export:hover { background: #c4613f; }
|
|
21
|
+
table { width: 100%; max-width: 1100px; border-collapse: collapse; background: white; border-radius: 6px; overflow: hidden; box-shadow: 0 1px 3px rgba(0,0,0,0.08); }
|
|
22
|
+
th { font-family: 'Poppins', sans-serif; background: #141413; color: #faf9f5; padding: 0.75rem 1rem; text-align: left; font-size: 0.875rem; }
|
|
23
|
+
td { padding: 0.75rem 1rem; border-bottom: 1px solid #e8e6dc; vertical-align: top; }
|
|
24
|
+
tr:nth-child(even) td { background: #faf9f5; }
|
|
25
|
+
tr:hover td { background: #f3f1ea; }
|
|
26
|
+
.section-header td { background: #e8e6dc; font-family: 'Poppins', sans-serif; font-weight: 500; font-size: 0.8rem; color: #141413; text-transform: uppercase; letter-spacing: 0.05em; }
|
|
27
|
+
.query-input { width: 100%; padding: 0.4rem; border: 1px solid #e8e6dc; border-radius: 4px; font-size: 0.875rem; font-family: 'Lora', Georgia, serif; resize: vertical; min-height: 60px; }
|
|
28
|
+
.query-input:focus { outline: none; border-color: #d97757; box-shadow: 0 0 0 2px rgba(217,119,87,0.15); }
|
|
29
|
+
.toggle { position: relative; display: inline-block; width: 44px; height: 24px; }
|
|
30
|
+
.toggle input { opacity: 0; width: 0; height: 0; }
|
|
31
|
+
.toggle .slider { position: absolute; inset: 0; background: #b0aea5; border-radius: 24px; cursor: pointer; transition: 0.2s; }
|
|
32
|
+
.toggle .slider::before { content: ""; position: absolute; width: 18px; height: 18px; left: 3px; bottom: 3px; background: white; border-radius: 50%; transition: 0.2s; }
|
|
33
|
+
.toggle input:checked + .slider { background: #d97757; }
|
|
34
|
+
.toggle input:checked + .slider::before { transform: translateX(20px); }
|
|
35
|
+
.btn-delete { background: #c44; color: white; padding: 0.3rem 0.6rem; border: none; border-radius: 4px; cursor: pointer; font-size: 0.75rem; font-family: 'Poppins', sans-serif; }
|
|
36
|
+
.btn-delete:hover { background: #a33; }
|
|
37
|
+
.summary { margin-top: 1rem; color: #b0aea5; font-size: 0.875rem; }
|
|
38
|
+
</style>
|
|
39
|
+
</head>
|
|
40
|
+
<body>
|
|
41
|
+
<h1>Eval Set Review: <span id="skill-name">__SKILL_NAME_PLACEHOLDER__</span></h1>
|
|
42
|
+
<p class="description">Current description: <span id="skill-desc">__SKILL_DESCRIPTION_PLACEHOLDER__</span></p>
|
|
43
|
+
|
|
44
|
+
<div class="controls">
|
|
45
|
+
<button class="btn btn-add" onclick="addRow()">+ Add Query</button>
|
|
46
|
+
<button class="btn btn-export" onclick="exportEvalSet()">Export Eval Set</button>
|
|
47
|
+
</div>
|
|
48
|
+
|
|
49
|
+
<table>
|
|
50
|
+
<thead>
|
|
51
|
+
<tr>
|
|
52
|
+
<th style="width:65%">Query</th>
|
|
53
|
+
<th style="width:18%">Should Trigger</th>
|
|
54
|
+
<th style="width:10%">Actions</th>
|
|
55
|
+
</tr>
|
|
56
|
+
</thead>
|
|
57
|
+
<tbody id="eval-body"></tbody>
|
|
58
|
+
</table>
|
|
59
|
+
|
|
60
|
+
<p class="summary" id="summary"></p>
|
|
61
|
+
|
|
62
|
+
<script>
|
|
63
|
+
const EVAL_DATA = __EVAL_DATA_PLACEHOLDER__;
|
|
64
|
+
|
|
65
|
+
let evalItems = [...EVAL_DATA];
|
|
66
|
+
|
|
67
|
+
function render() {
|
|
68
|
+
const tbody = document.getElementById('eval-body');
|
|
69
|
+
tbody.innerHTML = '';
|
|
70
|
+
|
|
71
|
+
// Sort: should-trigger first, then should-not-trigger
|
|
72
|
+
const sorted = evalItems
|
|
73
|
+
.map((item, origIdx) => ({ ...item, origIdx }))
|
|
74
|
+
.sort((a, b) => (b.should_trigger ? 1 : 0) - (a.should_trigger ? 1 : 0));
|
|
75
|
+
|
|
76
|
+
let lastGroup = null;
|
|
77
|
+
sorted.forEach(item => {
|
|
78
|
+
const group = item.should_trigger ? 'trigger' : 'no-trigger';
|
|
79
|
+
if (group !== lastGroup) {
|
|
80
|
+
const headerRow = document.createElement('tr');
|
|
81
|
+
headerRow.className = 'section-header';
|
|
82
|
+
headerRow.innerHTML = `<td colspan="3">${item.should_trigger ? 'Should Trigger' : 'Should NOT Trigger'}</td>`;
|
|
83
|
+
tbody.appendChild(headerRow);
|
|
84
|
+
lastGroup = group;
|
|
85
|
+
}
|
|
86
|
+
|
|
87
|
+
const idx = item.origIdx;
|
|
88
|
+
const tr = document.createElement('tr');
|
|
89
|
+
tr.innerHTML = `
|
|
90
|
+
<td><textarea class="query-input" onchange="updateQuery(${idx}, this.value)">${escapeHtml(item.query)}</textarea></td>
|
|
91
|
+
<td>
|
|
92
|
+
<label class="toggle">
|
|
93
|
+
<input type="checkbox" ${item.should_trigger ? 'checked' : ''} onchange="updateTrigger(${idx}, this.checked)">
|
|
94
|
+
<span class="slider"></span>
|
|
95
|
+
</label>
|
|
96
|
+
<span style="margin-left:8px;font-size:0.8rem;color:#b0aea5">${item.should_trigger ? 'Yes' : 'No'}</span>
|
|
97
|
+
</td>
|
|
98
|
+
<td><button class="btn-delete" onclick="deleteRow(${idx})">Delete</button></td>
|
|
99
|
+
`;
|
|
100
|
+
tbody.appendChild(tr);
|
|
101
|
+
});
|
|
102
|
+
updateSummary();
|
|
103
|
+
}
|
|
104
|
+
|
|
105
|
+
function escapeHtml(text) {
|
|
106
|
+
const div = document.createElement('div');
|
|
107
|
+
div.textContent = text;
|
|
108
|
+
return div.innerHTML;
|
|
109
|
+
}
|
|
110
|
+
|
|
111
|
+
function updateQuery(idx, value) { evalItems[idx].query = value; updateSummary(); }
|
|
112
|
+
function updateTrigger(idx, value) { evalItems[idx].should_trigger = value; render(); }
|
|
113
|
+
function deleteRow(idx) { evalItems.splice(idx, 1); render(); }
|
|
114
|
+
|
|
115
|
+
function addRow() {
|
|
116
|
+
evalItems.push({ query: '', should_trigger: true });
|
|
117
|
+
render();
|
|
118
|
+
const inputs = document.querySelectorAll('.query-input');
|
|
119
|
+
inputs[inputs.length - 1].focus();
|
|
120
|
+
}
|
|
121
|
+
|
|
122
|
+
function updateSummary() {
|
|
123
|
+
const trigger = evalItems.filter(i => i.should_trigger).length;
|
|
124
|
+
const noTrigger = evalItems.filter(i => !i.should_trigger).length;
|
|
125
|
+
document.getElementById('summary').textContent =
|
|
126
|
+
`${evalItems.length} queries total: ${trigger} should trigger, ${noTrigger} should not trigger`;
|
|
127
|
+
}
|
|
128
|
+
|
|
129
|
+
function exportEvalSet() {
|
|
130
|
+
const valid = evalItems.filter(i => i.query.trim() !== '');
|
|
131
|
+
const data = valid.map(i => ({ query: i.query.trim(), should_trigger: i.should_trigger }));
|
|
132
|
+
const blob = new Blob([JSON.stringify(data, null, 2)], { type: 'application/json' });
|
|
133
|
+
const url = URL.createObjectURL(blob);
|
|
134
|
+
const a = document.createElement('a');
|
|
135
|
+
a.href = url;
|
|
136
|
+
a.download = 'eval_set.json';
|
|
137
|
+
document.body.appendChild(a);
|
|
138
|
+
a.click();
|
|
139
|
+
document.body.removeChild(a);
|
|
140
|
+
URL.revokeObjectURL(url);
|
|
141
|
+
}
|
|
142
|
+
|
|
143
|
+
render();
|
|
144
|
+
</script>
|
|
145
|
+
</body>
|
|
146
|
+
</html>
|
|
@@ -0,0 +1,471 @@
|
|
|
1
|
+
#!/usr/bin/env python3
|
|
2
|
+
"""Generate and serve a review page for eval results.
|
|
3
|
+
|
|
4
|
+
Reads the workspace directory, discovers runs (directories with outputs/),
|
|
5
|
+
embeds all output data into a self-contained HTML page, and serves it via
|
|
6
|
+
a tiny HTTP server. Feedback auto-saves to feedback.json in the workspace.
|
|
7
|
+
|
|
8
|
+
Usage:
|
|
9
|
+
python generate_review.py <workspace-path> [--port PORT] [--skill-name NAME]
|
|
10
|
+
python generate_review.py <workspace-path> --previous-feedback /path/to/old/feedback.json
|
|
11
|
+
|
|
12
|
+
No dependencies beyond the Python stdlib are required.
|
|
13
|
+
"""
|
|
14
|
+
|
|
15
|
+
import argparse
|
|
16
|
+
import base64
|
|
17
|
+
import json
|
|
18
|
+
import mimetypes
|
|
19
|
+
import os
|
|
20
|
+
import re
|
|
21
|
+
import signal
|
|
22
|
+
import subprocess
|
|
23
|
+
import sys
|
|
24
|
+
import time
|
|
25
|
+
import webbrowser
|
|
26
|
+
from functools import partial
|
|
27
|
+
from http.server import HTTPServer, BaseHTTPRequestHandler
|
|
28
|
+
from pathlib import Path
|
|
29
|
+
|
|
30
|
+
# Files to exclude from output listings
|
|
31
|
+
METADATA_FILES = {"transcript.md", "user_notes.md", "metrics.json"}
|
|
32
|
+
|
|
33
|
+
# Extensions we render as inline text
|
|
34
|
+
TEXT_EXTENSIONS = {
|
|
35
|
+
".txt", ".md", ".json", ".csv", ".py", ".js", ".ts", ".tsx", ".jsx",
|
|
36
|
+
".yaml", ".yml", ".xml", ".html", ".css", ".sh", ".rb", ".go", ".rs",
|
|
37
|
+
".java", ".c", ".cpp", ".h", ".hpp", ".sql", ".r", ".toml",
|
|
38
|
+
}
|
|
39
|
+
|
|
40
|
+
# Extensions we render as inline images
|
|
41
|
+
IMAGE_EXTENSIONS = {".png", ".jpg", ".jpeg", ".gif", ".svg", ".webp"}
|
|
42
|
+
|
|
43
|
+
# MIME type overrides for common types
|
|
44
|
+
MIME_OVERRIDES = {
|
|
45
|
+
".svg": "image/svg+xml",
|
|
46
|
+
".xlsx": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
|
|
47
|
+
".docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
|
|
48
|
+
".pptx": "application/vnd.openxmlformats-officedocument.presentationml.presentation",
|
|
49
|
+
}
|
|
50
|
+
|
|
51
|
+
|
|
52
|
+
def get_mime_type(path: Path) -> str:
|
|
53
|
+
ext = path.suffix.lower()
|
|
54
|
+
if ext in MIME_OVERRIDES:
|
|
55
|
+
return MIME_OVERRIDES[ext]
|
|
56
|
+
mime, _ = mimetypes.guess_type(str(path))
|
|
57
|
+
return mime or "application/octet-stream"
|
|
58
|
+
|
|
59
|
+
|
|
60
|
+
def find_runs(workspace: Path) -> list[dict]:
|
|
61
|
+
"""Recursively find directories that contain an outputs/ subdirectory."""
|
|
62
|
+
runs: list[dict] = []
|
|
63
|
+
_find_runs_recursive(workspace, workspace, runs)
|
|
64
|
+
runs.sort(key=lambda r: (r.get("eval_id", float("inf")), r["id"]))
|
|
65
|
+
return runs
|
|
66
|
+
|
|
67
|
+
|
|
68
|
+
def _find_runs_recursive(root: Path, current: Path, runs: list[dict]) -> None:
|
|
69
|
+
if not current.is_dir():
|
|
70
|
+
return
|
|
71
|
+
|
|
72
|
+
outputs_dir = current / "outputs"
|
|
73
|
+
if outputs_dir.is_dir():
|
|
74
|
+
run = build_run(root, current)
|
|
75
|
+
if run:
|
|
76
|
+
runs.append(run)
|
|
77
|
+
return
|
|
78
|
+
|
|
79
|
+
skip = {"node_modules", ".git", "__pycache__", "skill", "inputs"}
|
|
80
|
+
for child in sorted(current.iterdir()):
|
|
81
|
+
if child.is_dir() and child.name not in skip:
|
|
82
|
+
_find_runs_recursive(root, child, runs)
|
|
83
|
+
|
|
84
|
+
|
|
85
|
+
def build_run(root: Path, run_dir: Path) -> dict | None:
|
|
86
|
+
"""Build a run dict with prompt, outputs, and grading data."""
|
|
87
|
+
prompt = ""
|
|
88
|
+
eval_id = None
|
|
89
|
+
|
|
90
|
+
# Try eval_metadata.json
|
|
91
|
+
for candidate in [run_dir / "eval_metadata.json", run_dir.parent / "eval_metadata.json"]:
|
|
92
|
+
if candidate.exists():
|
|
93
|
+
try:
|
|
94
|
+
metadata = json.loads(candidate.read_text())
|
|
95
|
+
prompt = metadata.get("prompt", "")
|
|
96
|
+
eval_id = metadata.get("eval_id")
|
|
97
|
+
except (json.JSONDecodeError, OSError):
|
|
98
|
+
pass
|
|
99
|
+
if prompt:
|
|
100
|
+
break
|
|
101
|
+
|
|
102
|
+
# Fall back to transcript.md
|
|
103
|
+
if not prompt:
|
|
104
|
+
for candidate in [run_dir / "transcript.md", run_dir / "outputs" / "transcript.md"]:
|
|
105
|
+
if candidate.exists():
|
|
106
|
+
try:
|
|
107
|
+
text = candidate.read_text()
|
|
108
|
+
match = re.search(r"## Eval Prompt\n\n([\s\S]*?)(?=\n##|$)", text)
|
|
109
|
+
if match:
|
|
110
|
+
prompt = match.group(1).strip()
|
|
111
|
+
except OSError:
|
|
112
|
+
pass
|
|
113
|
+
if prompt:
|
|
114
|
+
break
|
|
115
|
+
|
|
116
|
+
if not prompt:
|
|
117
|
+
prompt = "(No prompt found)"
|
|
118
|
+
|
|
119
|
+
run_id = str(run_dir.relative_to(root)).replace("/", "-").replace("\\", "-")
|
|
120
|
+
|
|
121
|
+
# Collect output files
|
|
122
|
+
outputs_dir = run_dir / "outputs"
|
|
123
|
+
output_files: list[dict] = []
|
|
124
|
+
if outputs_dir.is_dir():
|
|
125
|
+
for f in sorted(outputs_dir.iterdir()):
|
|
126
|
+
if f.is_file() and f.name not in METADATA_FILES:
|
|
127
|
+
output_files.append(embed_file(f))
|
|
128
|
+
|
|
129
|
+
# Load grading if present
|
|
130
|
+
grading = None
|
|
131
|
+
for candidate in [run_dir / "grading.json", run_dir.parent / "grading.json"]:
|
|
132
|
+
if candidate.exists():
|
|
133
|
+
try:
|
|
134
|
+
grading = json.loads(candidate.read_text())
|
|
135
|
+
except (json.JSONDecodeError, OSError):
|
|
136
|
+
pass
|
|
137
|
+
if grading:
|
|
138
|
+
break
|
|
139
|
+
|
|
140
|
+
return {
|
|
141
|
+
"id": run_id,
|
|
142
|
+
"prompt": prompt,
|
|
143
|
+
"eval_id": eval_id,
|
|
144
|
+
"outputs": output_files,
|
|
145
|
+
"grading": grading,
|
|
146
|
+
}
|
|
147
|
+
|
|
148
|
+
|
|
149
|
+
def embed_file(path: Path) -> dict:
|
|
150
|
+
"""Read a file and return an embedded representation."""
|
|
151
|
+
ext = path.suffix.lower()
|
|
152
|
+
mime = get_mime_type(path)
|
|
153
|
+
|
|
154
|
+
if ext in TEXT_EXTENSIONS:
|
|
155
|
+
try:
|
|
156
|
+
content = path.read_text(errors="replace")
|
|
157
|
+
except OSError:
|
|
158
|
+
content = "(Error reading file)"
|
|
159
|
+
return {
|
|
160
|
+
"name": path.name,
|
|
161
|
+
"type": "text",
|
|
162
|
+
"content": content,
|
|
163
|
+
}
|
|
164
|
+
elif ext in IMAGE_EXTENSIONS:
|
|
165
|
+
try:
|
|
166
|
+
raw = path.read_bytes()
|
|
167
|
+
b64 = base64.b64encode(raw).decode("ascii")
|
|
168
|
+
except OSError:
|
|
169
|
+
return {"name": path.name, "type": "error", "content": "(Error reading file)"}
|
|
170
|
+
return {
|
|
171
|
+
"name": path.name,
|
|
172
|
+
"type": "image",
|
|
173
|
+
"mime": mime,
|
|
174
|
+
"data_uri": f"data:{mime};base64,{b64}",
|
|
175
|
+
}
|
|
176
|
+
elif ext == ".pdf":
|
|
177
|
+
try:
|
|
178
|
+
raw = path.read_bytes()
|
|
179
|
+
b64 = base64.b64encode(raw).decode("ascii")
|
|
180
|
+
except OSError:
|
|
181
|
+
return {"name": path.name, "type": "error", "content": "(Error reading file)"}
|
|
182
|
+
return {
|
|
183
|
+
"name": path.name,
|
|
184
|
+
"type": "pdf",
|
|
185
|
+
"data_uri": f"data:{mime};base64,{b64}",
|
|
186
|
+
}
|
|
187
|
+
elif ext == ".xlsx":
|
|
188
|
+
try:
|
|
189
|
+
raw = path.read_bytes()
|
|
190
|
+
b64 = base64.b64encode(raw).decode("ascii")
|
|
191
|
+
except OSError:
|
|
192
|
+
return {"name": path.name, "type": "error", "content": "(Error reading file)"}
|
|
193
|
+
return {
|
|
194
|
+
"name": path.name,
|
|
195
|
+
"type": "xlsx",
|
|
196
|
+
"data_b64": b64,
|
|
197
|
+
}
|
|
198
|
+
else:
|
|
199
|
+
# Binary / unknown — base64 download link
|
|
200
|
+
try:
|
|
201
|
+
raw = path.read_bytes()
|
|
202
|
+
b64 = base64.b64encode(raw).decode("ascii")
|
|
203
|
+
except OSError:
|
|
204
|
+
return {"name": path.name, "type": "error", "content": "(Error reading file)"}
|
|
205
|
+
return {
|
|
206
|
+
"name": path.name,
|
|
207
|
+
"type": "binary",
|
|
208
|
+
"mime": mime,
|
|
209
|
+
"data_uri": f"data:{mime};base64,{b64}",
|
|
210
|
+
}
|
|
211
|
+
|
|
212
|
+
|
|
213
|
+
def load_previous_iteration(workspace: Path) -> dict[str, dict]:
|
|
214
|
+
"""Load previous iteration's feedback and outputs.
|
|
215
|
+
|
|
216
|
+
Returns a map of run_id -> {"feedback": str, "outputs": list[dict]}.
|
|
217
|
+
"""
|
|
218
|
+
result: dict[str, dict] = {}
|
|
219
|
+
|
|
220
|
+
# Load feedback
|
|
221
|
+
feedback_map: dict[str, str] = {}
|
|
222
|
+
feedback_path = workspace / "feedback.json"
|
|
223
|
+
if feedback_path.exists():
|
|
224
|
+
try:
|
|
225
|
+
data = json.loads(feedback_path.read_text())
|
|
226
|
+
feedback_map = {
|
|
227
|
+
r["run_id"]: r["feedback"]
|
|
228
|
+
for r in data.get("reviews", [])
|
|
229
|
+
if r.get("feedback", "").strip()
|
|
230
|
+
}
|
|
231
|
+
except (json.JSONDecodeError, OSError, KeyError):
|
|
232
|
+
pass
|
|
233
|
+
|
|
234
|
+
# Load runs (to get outputs)
|
|
235
|
+
prev_runs = find_runs(workspace)
|
|
236
|
+
for run in prev_runs:
|
|
237
|
+
result[run["id"]] = {
|
|
238
|
+
"feedback": feedback_map.get(run["id"], ""),
|
|
239
|
+
"outputs": run.get("outputs", []),
|
|
240
|
+
}
|
|
241
|
+
|
|
242
|
+
# Also add feedback for run_ids that had feedback but no matching run
|
|
243
|
+
for run_id, fb in feedback_map.items():
|
|
244
|
+
if run_id not in result:
|
|
245
|
+
result[run_id] = {"feedback": fb, "outputs": []}
|
|
246
|
+
|
|
247
|
+
return result
|
|
248
|
+
|
|
249
|
+
|
|
250
|
+
def generate_html(
|
|
251
|
+
runs: list[dict],
|
|
252
|
+
skill_name: str,
|
|
253
|
+
previous: dict[str, dict] | None = None,
|
|
254
|
+
benchmark: dict | None = None,
|
|
255
|
+
) -> str:
|
|
256
|
+
"""Generate the complete standalone HTML page with embedded data."""
|
|
257
|
+
template_path = Path(__file__).parent / "viewer.html"
|
|
258
|
+
template = template_path.read_text()
|
|
259
|
+
|
|
260
|
+
# Build previous_feedback and previous_outputs maps for the template
|
|
261
|
+
previous_feedback: dict[str, str] = {}
|
|
262
|
+
previous_outputs: dict[str, list[dict]] = {}
|
|
263
|
+
if previous:
|
|
264
|
+
for run_id, data in previous.items():
|
|
265
|
+
if data.get("feedback"):
|
|
266
|
+
previous_feedback[run_id] = data["feedback"]
|
|
267
|
+
if data.get("outputs"):
|
|
268
|
+
previous_outputs[run_id] = data["outputs"]
|
|
269
|
+
|
|
270
|
+
embedded = {
|
|
271
|
+
"skill_name": skill_name,
|
|
272
|
+
"runs": runs,
|
|
273
|
+
"previous_feedback": previous_feedback,
|
|
274
|
+
"previous_outputs": previous_outputs,
|
|
275
|
+
}
|
|
276
|
+
if benchmark:
|
|
277
|
+
embedded["benchmark"] = benchmark
|
|
278
|
+
|
|
279
|
+
data_json = json.dumps(embedded)
|
|
280
|
+
|
|
281
|
+
return template.replace("/*__EMBEDDED_DATA__*/", f"const EMBEDDED_DATA = {data_json};")
|
|
282
|
+
|
|
283
|
+
|
|
284
|
+
# ---------------------------------------------------------------------------
|
|
285
|
+
# HTTP server (stdlib only, zero dependencies)
|
|
286
|
+
# ---------------------------------------------------------------------------
|
|
287
|
+
|
|
288
|
+
def _kill_port(port: int) -> None:
|
|
289
|
+
"""Kill any process listening on the given port."""
|
|
290
|
+
try:
|
|
291
|
+
result = subprocess.run(
|
|
292
|
+
["lsof", "-ti", f":{port}"],
|
|
293
|
+
capture_output=True, text=True, timeout=5,
|
|
294
|
+
)
|
|
295
|
+
for pid_str in result.stdout.strip().split("\n"):
|
|
296
|
+
if pid_str.strip():
|
|
297
|
+
try:
|
|
298
|
+
os.kill(int(pid_str.strip()), signal.SIGTERM)
|
|
299
|
+
except (ProcessLookupError, ValueError):
|
|
300
|
+
pass
|
|
301
|
+
if result.stdout.strip():
|
|
302
|
+
time.sleep(0.5)
|
|
303
|
+
except subprocess.TimeoutExpired:
|
|
304
|
+
pass
|
|
305
|
+
except FileNotFoundError:
|
|
306
|
+
print("Note: lsof not found, cannot check if port is in use", file=sys.stderr)
|
|
307
|
+
|
|
308
|
+
class ReviewHandler(BaseHTTPRequestHandler):
|
|
309
|
+
"""Serves the review HTML and handles feedback saves.
|
|
310
|
+
|
|
311
|
+
Regenerates the HTML on each page load so that refreshing the browser
|
|
312
|
+
picks up new eval outputs without restarting the server.
|
|
313
|
+
"""
|
|
314
|
+
|
|
315
|
+
def __init__(
|
|
316
|
+
self,
|
|
317
|
+
workspace: Path,
|
|
318
|
+
skill_name: str,
|
|
319
|
+
feedback_path: Path,
|
|
320
|
+
previous: dict[str, dict],
|
|
321
|
+
benchmark_path: Path | None,
|
|
322
|
+
*args,
|
|
323
|
+
**kwargs,
|
|
324
|
+
):
|
|
325
|
+
self.workspace = workspace
|
|
326
|
+
self.skill_name = skill_name
|
|
327
|
+
self.feedback_path = feedback_path
|
|
328
|
+
self.previous = previous
|
|
329
|
+
self.benchmark_path = benchmark_path
|
|
330
|
+
super().__init__(*args, **kwargs)
|
|
331
|
+
|
|
332
|
+
def do_GET(self) -> None:
|
|
333
|
+
if self.path == "/" or self.path == "/index.html":
|
|
334
|
+
# Regenerate HTML on each request (re-scans workspace for new outputs)
|
|
335
|
+
runs = find_runs(self.workspace)
|
|
336
|
+
benchmark = None
|
|
337
|
+
if self.benchmark_path and self.benchmark_path.exists():
|
|
338
|
+
try:
|
|
339
|
+
benchmark = json.loads(self.benchmark_path.read_text())
|
|
340
|
+
except (json.JSONDecodeError, OSError):
|
|
341
|
+
pass
|
|
342
|
+
html = generate_html(runs, self.skill_name, self.previous, benchmark)
|
|
343
|
+
content = html.encode("utf-8")
|
|
344
|
+
self.send_response(200)
|
|
345
|
+
self.send_header("Content-Type", "text/html; charset=utf-8")
|
|
346
|
+
self.send_header("Content-Length", str(len(content)))
|
|
347
|
+
self.end_headers()
|
|
348
|
+
self.wfile.write(content)
|
|
349
|
+
elif self.path == "/api/feedback":
|
|
350
|
+
data = b"{}"
|
|
351
|
+
if self.feedback_path.exists():
|
|
352
|
+
data = self.feedback_path.read_bytes()
|
|
353
|
+
self.send_response(200)
|
|
354
|
+
self.send_header("Content-Type", "application/json")
|
|
355
|
+
self.send_header("Content-Length", str(len(data)))
|
|
356
|
+
self.end_headers()
|
|
357
|
+
self.wfile.write(data)
|
|
358
|
+
else:
|
|
359
|
+
self.send_error(404)
|
|
360
|
+
|
|
361
|
+
def do_POST(self) -> None:
|
|
362
|
+
if self.path == "/api/feedback":
|
|
363
|
+
length = int(self.headers.get("Content-Length", 0))
|
|
364
|
+
body = self.rfile.read(length)
|
|
365
|
+
try:
|
|
366
|
+
data = json.loads(body)
|
|
367
|
+
if not isinstance(data, dict) or "reviews" not in data:
|
|
368
|
+
raise ValueError("Expected JSON object with 'reviews' key")
|
|
369
|
+
self.feedback_path.write_text(json.dumps(data, indent=2) + "\n")
|
|
370
|
+
resp = b'{"ok":true}'
|
|
371
|
+
self.send_response(200)
|
|
372
|
+
except (json.JSONDecodeError, OSError, ValueError) as e:
|
|
373
|
+
resp = json.dumps({"error": str(e)}).encode()
|
|
374
|
+
self.send_response(500)
|
|
375
|
+
self.send_header("Content-Type", "application/json")
|
|
376
|
+
self.send_header("Content-Length", str(len(resp)))
|
|
377
|
+
self.end_headers()
|
|
378
|
+
self.wfile.write(resp)
|
|
379
|
+
else:
|
|
380
|
+
self.send_error(404)
|
|
381
|
+
|
|
382
|
+
def log_message(self, format: str, *args: object) -> None:
|
|
383
|
+
# Suppress request logging to keep terminal clean
|
|
384
|
+
pass
|
|
385
|
+
|
|
386
|
+
|
|
387
|
+
def main() -> None:
|
|
388
|
+
parser = argparse.ArgumentParser(description="Generate and serve eval review")
|
|
389
|
+
parser.add_argument("workspace", type=Path, help="Path to workspace directory")
|
|
390
|
+
parser.add_argument("--port", "-p", type=int, default=3117, help="Server port (default: 3117)")
|
|
391
|
+
parser.add_argument("--skill-name", "-n", type=str, default=None, help="Skill name for header")
|
|
392
|
+
parser.add_argument(
|
|
393
|
+
"--previous-workspace", type=Path, default=None,
|
|
394
|
+
help="Path to previous iteration's workspace (shows old outputs and feedback as context)",
|
|
395
|
+
)
|
|
396
|
+
parser.add_argument(
|
|
397
|
+
"--benchmark", type=Path, default=None,
|
|
398
|
+
help="Path to benchmark.json to show in the Benchmark tab",
|
|
399
|
+
)
|
|
400
|
+
parser.add_argument(
|
|
401
|
+
"--static", "-s", type=Path, default=None,
|
|
402
|
+
help="Write standalone HTML to this path instead of starting a server",
|
|
403
|
+
)
|
|
404
|
+
args = parser.parse_args()
|
|
405
|
+
|
|
406
|
+
workspace = args.workspace.resolve()
|
|
407
|
+
if not workspace.is_dir():
|
|
408
|
+
print(f"Error: {workspace} is not a directory", file=sys.stderr)
|
|
409
|
+
sys.exit(1)
|
|
410
|
+
|
|
411
|
+
runs = find_runs(workspace)
|
|
412
|
+
if not runs:
|
|
413
|
+
print(f"No runs found in {workspace}", file=sys.stderr)
|
|
414
|
+
sys.exit(1)
|
|
415
|
+
|
|
416
|
+
skill_name = args.skill_name or workspace.name.replace("-workspace", "")
|
|
417
|
+
feedback_path = workspace / "feedback.json"
|
|
418
|
+
|
|
419
|
+
previous: dict[str, dict] = {}
|
|
420
|
+
if args.previous_workspace:
|
|
421
|
+
previous = load_previous_iteration(args.previous_workspace.resolve())
|
|
422
|
+
|
|
423
|
+
benchmark_path = args.benchmark.resolve() if args.benchmark else None
|
|
424
|
+
benchmark = None
|
|
425
|
+
if benchmark_path and benchmark_path.exists():
|
|
426
|
+
try:
|
|
427
|
+
benchmark = json.loads(benchmark_path.read_text())
|
|
428
|
+
except (json.JSONDecodeError, OSError):
|
|
429
|
+
pass
|
|
430
|
+
|
|
431
|
+
if args.static:
|
|
432
|
+
html = generate_html(runs, skill_name, previous, benchmark)
|
|
433
|
+
args.static.parent.mkdir(parents=True, exist_ok=True)
|
|
434
|
+
args.static.write_text(html)
|
|
435
|
+
print(f"\n Static viewer written to: {args.static}\n")
|
|
436
|
+
sys.exit(0)
|
|
437
|
+
|
|
438
|
+
# Kill any existing process on the target port
|
|
439
|
+
port = args.port
|
|
440
|
+
_kill_port(port)
|
|
441
|
+
handler = partial(ReviewHandler, workspace, skill_name, feedback_path, previous, benchmark_path)
|
|
442
|
+
try:
|
|
443
|
+
server = HTTPServer(("127.0.0.1", port), handler)
|
|
444
|
+
except OSError:
|
|
445
|
+
# Port still in use after kill attempt — find a free one
|
|
446
|
+
server = HTTPServer(("127.0.0.1", 0), handler)
|
|
447
|
+
port = server.server_address[1]
|
|
448
|
+
|
|
449
|
+
url = f"http://localhost:{port}"
|
|
450
|
+
print(f"\n Eval Viewer")
|
|
451
|
+
print(f" ─────────────────────────────────")
|
|
452
|
+
print(f" URL: {url}")
|
|
453
|
+
print(f" Workspace: {workspace}")
|
|
454
|
+
print(f" Feedback: {feedback_path}")
|
|
455
|
+
if previous:
|
|
456
|
+
print(f" Previous: {args.previous_workspace} ({len(previous)} runs)")
|
|
457
|
+
if benchmark_path:
|
|
458
|
+
print(f" Benchmark: {benchmark_path}")
|
|
459
|
+
print(f"\n Press Ctrl+C to stop.\n")
|
|
460
|
+
|
|
461
|
+
webbrowser.open(url)
|
|
462
|
+
|
|
463
|
+
try:
|
|
464
|
+
server.serve_forever()
|
|
465
|
+
except KeyboardInterrupt:
|
|
466
|
+
print("\nStopped.")
|
|
467
|
+
server.server_close()
|
|
468
|
+
|
|
469
|
+
|
|
470
|
+
if __name__ == "__main__":
|
|
471
|
+
main()
|