Composio powers 1000+ toolkits, tool search, context management, authentication, and a sandboxed workbench to help you build AI agents that turn intent into action.
Mirrored from https://github.com/Camo67/composio.git
Synced Nov 21, 03:37 UTC
Custom component Markdown audit
Audit for DEVREL-35, 2026-09-10. Scope: components registered in mdx-components.tsx and used in content/. The page endpoint and full corpus call getLLMText; local and Algolia search records call mdxToCleanMarkdown on authored MDX.
| Content | Components | Result |
|---|---|---|
| Package installation | PackageInstall |
Preserve all supported package-manager commands and display comments. Node and Python manager definitions are shared with the UI. Test every authored instance. |
| Coding-agent setup | AgentSetupActions, AgentFirstPrompt, AgentSetupGrid |
Preserve the same prompts and client destinations as the UI through shared data modules. |
| Versioned API details | ApiBaseUrl, ApiEndpointsTable |
Existing converters preserve version-specific URLs and endpoint tables. Existing tests cover raw and processed attributes. |
| Decisions and warnings | Callout, Tabs, Tab, TabsContent, FrameworkOption, IntegrationTabs, IntegrationContent, ToolTypeOption, ConnectClientOption, Accordion |
Existing converters retain text, labels, and warning content. |
| Step instructions and definitions | Steps, Step, StepTitle, Glossary, GlossaryTerm |
Existing converters retain headings and content. |
| Linked content | Card, TemplateCard, HomeSurfaces, AIToolsBanner |
Existing converters retain destinations or route readers to canonical setup instructions. |
| Provider inventory | ProviderList |
lib/providers.ts supplies semantic server-rendered lists and Markdown links, including language support and experimental labels. The concise discovery index selects OpenAI and Anthropic from the same inventory; the complete index lists the underlying guides. |
| Media | Figure, YouTube, Video |
Preserve captions and source links. Video links now survive conversion. Transcription of media is outside this audit. |
| Example source | FileBuildup, RepoBrowser |
Existing converters expose staged source or an explicit repository-availability notice. |
| Structural wrappers and icons | Cards, TemplateGrid, QuickstartFlow, FrameworkSelector, ToolTypeFlow, ConnectFlow, Accordions, CapabilityList, MediaSplit, AppLogo, registered Lucide icons |
Wrappers have no task instructions; their child content survives. Icons are decorative. |
| Visual explanations | SessionFlow, TriggersFlow, SlackBotFlow, LocalWorkbenchFlow, LocalSandboxBoundary, ImessageFlow, WorkbenchFlow, AuthConfigFlow, WhiteLabelFlow, ImportConnectionFlow, ManageConnectionsVisual, ConnectionRefreshVisual, InChatAuthTerminal, ClaudeMockUI, ZdrLogVisual |
Diagrams illustrate the surrounding guide. The authored instructions remain the executable path. Full visual-to-text parity is a follow-up, not claimed by this fix. |
| Mermaid | Mermaid |
getLLMText preserves processed diagram code. Raw component diagrams in search require separate parity coverage. |
| Homepage and catalogs | DocsHero, HomeFeatures, HomeResources, ToolkitsLanding, ManagedAuthList |
Dynamic content is not fully serialized by the generic converter. Product selection survives through HomeSurfaces; toolkit routes have dedicated Markdown renderers, and managed-auth prose includes the API lookup. Full catalog and homepage parity remains a follow-up. |
CapabilityCard is registered but has no authored instances at this revision. Add a Markdown representation before introducing it into task-critical content.
Remaining checks
The semantic parity suite for DEVREL-32 checks actual HTTP page output, the full corpus, and search ingestion against the same expected facts. Its export checker applies those expectations to external retrieval captures. Static tests cannot prove an external index refreshed. DEVREL-38 can use the remaining visual and catalog gaps to prioritize task-complete pages. Do not treat this audit as evidence of benchmark improvement.