vance

feat: Add AI agent setup directory (#30115) * first pass * ui improvements * Add agent-setup homepage section with illustration * Replace agents illustration with combined SVG, adjust sizing * agent-setup page overhaul * agent specific pages improvements * fix: add primary_type field to AgentData type and schema * chore: fix prettier formatting in agent-setup.css * claude code improvements * first pass using mdx for agent files * finalize switch to mdx * fix troubleshooting UI * mobile ui * format * codex changes * first content accuracy pass * compare agents table improvements * increase space between sections on agent-setup page * fix * address bonk * feedback * formatting * feat: swap homepage section order — agent setup before Build with Cloudflare * Add agent-setup directory entry for llms.txt and Algolia indexing

feat: extend prettier plugin to preserve more mdx components (#29819)

feat: add Flue spam and off-topic filter agent (#30832) * chore: ignore .worktrees/ directory * Revert "chore: ignore .worktrees/ directory" This reverts commit 4c311f351c244e78bb448e8df9407b289a92ad9f. * feat: add Flue spam-filter agent for GitHub issues and PRs * chore: rename Flue spam and off-topic filter * fix: address Flue filter review feedback * just-bash is needed by flue * add observability * deploy should sync skills * add logging to spam agent * add more loggings * disable invocation logs * enable traces * better log messages * disable traces * fix eslint * ignore flue dist in eslint * add urls to more logs * fix: isolate Flue filter sessions * fix: improve Flue orchestrator verdict logs * chore: migrate Flue to 0.7, add flue.config.ts, fix local dev * [Flue] Handle reopened and ready_for_review events in spam filter * chore: fix prettier formatting in orchestrate.ts * [Flue] Comment out low-value webhook received/ignored logs * [Flue] Add spam/off-topic labels when closing issues and PRs * chore: remove unused agents dependency * chore: restore agents dependency (required by Flue generated entry)

chore: consolidate agent skills and create shared style guide references (#30735) * chore: consolidate agent skills and create shared style guide references - Add .agents/references/style-guide.md — canonical writing and formatting rules - Add .agents/references/components.md — full MDX component catalog with props - Add .agents/references/procedures.md — step-writing rules for how-tos and tutorials - Update changelog, docs-review skills to reference shared style guide - Remove content-rules.md (content merged into shared reference) - Fix changelog skill code examples section (WranglerConfig/TypeScriptExample rules) - Fix docs-review quick-reference table (adds PackageManagers row, removes duplicates) - Rename verify-dependabot-pr → dependabot-review - Remove npm-to-pnpm skill (outdated) - Update AGENTS.md: remove duplicated component docs, add skills/references index, fix description field annotation * chore: add missing components to agent reference (Badge, Card, YouTube, Stream, APIRequest, CURL, WranglerCommand, Markdown) * chore: add prebuild check to keep components.md in sync with style guide pages * chore: document all remaining components in agent reference, remove exclusion list from check script * chore: move component docs check to CI pre-build step, not package.json prebuild * fix: format check-component-docs.ts, add formatting reminder to AGENTS.md * chore: remove circular reference from components.md * chore: add bidirectional sync check for component docs, add missing styleGuide.component to badges.mdx * fix: remove confusing WranglerConfig carve-out from code-review skill

feat: improve copy button UX with checkmark feedback and tippy tooltips (#30441) * feat: replace copy button feedback tooltip with checkmark icon, add tippy tooltips to copy buttons - Swap 'Copied!' feedback tooltip on EC code blocks for a checkmark icon - Add tippy.js tooltips to EC copy buttons and PackageManagers copy button - Standardize copy/check icons to Phosphor Regular across EC and PackageManagers - Extract tippy.css overrides from Head.astro into src/styles/tippy.css - Consolidate agent-setup components to use addTooltip() util - Migrate AgentsToolkit tooltips from custom portal impl to tippy * adjust font size * add important to tippy css properties * copy agent-toolkit list styles from TOC * icon button fixes * chore: clean up AgentsToolkit list item and icon button styles - Match TOC link style (sidebar-text colors, py-1.5, leading-snug) - Remove pointless div wrapper from IconButton - Replace sr-only span with aria-label, add aria-hidden to icon - Fix Prettier formatting * fix: guard against double init of tippy instances on page load * fix: remove copy button click listeners on cleanup to prevent accumulation

perf: reduce Workers AI model page HTML payload size (#30350) * perf: reduce Workers AI model page HTML size - Add ModelCardData type: slim projection of ResolvedModel that excludes schema, apiModes, codeSnippets, examples, and metadata fields - Strip ~1.76 MB of unused JSON Schema data from ModelCatalog island props on /workers-ai/models/ and /ai/models/ index pages (catalog only needs name, task, description, properties, and a handful of scalars for filtering/rendering — never the full schema) - Change client:load to client:visible on SchemaTree/SchemaVariantSelector islands so React hydration is deferred until the schema section scrolls into view (it is below the fold on every detail page) - Deduplicate input schema serialisation on multi-mode model pages: when sync and streaming modes share identical input, render the input schema once and show per-mode output sections, eliminating ~145 KB of duplicate SchemaRowData props per affected detail page * perf: serve model JSON schemas as static files, remove Code bloat Replace the 'API Schemas (Raw)' section's syntax-highlighted <Code> blocks with static .json endpoints and a compact link UI. The <Code> components were inflating 227 KB of raw JSON into ~1.95 MB of syntax-highlighted HTML (8.6x ratio) via Expressive Code span wrapping. On models with multiple API modes (sync/streaming/batch), the same schema was rendered multiple times, pushing detail pages to ~3.1 MB. Changes: - Extract detectApiModes() from model-resolver.ts into model-schema.ts so it can be shared without pulling in Astro's getCollection - Add src/pages/workers-ai/models/[...schema].json.ts: generates static JSON endpoints per model per schema (sync-input.json, sync-output.json, streaming-input.json, etc.) for all legacy models (~220 files) - Add src/pages/ai/models/[...schema].json.ts: same for catalog models, handling multi-segment slugs (e.g. /ai/models/openai/gpt-5.4-mini/schema-input.json) - Replace Code blocks in ModelDetailPage.astro with a 2-column card grid linking to the .json endpoints, each card having square open/download icon buttons Expected result: ~3.1 MB -> ~1.15 MB on gemma detail page * chore: fix prettier formatting in model-properties.ts

[agent-setup] Follow-ups: Manifest-driven Skills/MCP lists, static PlatformAccessSection (#30245) * feat: add Cloudflare tools banner to agent setup page * feat: link MCP Code Mode to GitHub repo in tools banner * [agent-setup] Drive Skills and MCP lists from middlecache manifests - Add cloudflare-skills-manifest and cloudflare-mcp-server-card content collections backed by middlecacheLoader with custom parsers - Add Zod schemas for both collections - Rewrite SkillsList.astro to use getCollection instead of hardcoded array - Rewrite McpServerList.astro to use getEntry and derive server names from remote URLs; descriptions remain locally maintained - Remove Tools for agents section and CloudflareToolsBanner from index - Link Cloudflare Skills and MCP servers in the lede to their GitHub repos * [agent-setup] Fully static PlatformAccessSection, drive MCP list from mcps-manifest - Switch cloudflare-mcp-server-card collection to cloudflare-mcps-manifest, backed by v1/cloudflare-mcps/mcps-manifest.json — names and descriptions now come from the manifest with no local hardcoding - Pin Code Mode server first in McpServerList with a 'code mode' chip - Add MCP server URLs to the list UI - PlatformAccessSection is now a zero-prop, zero-slot static component; all per-agent wrangler/docs/skills slot content removed - All six agent pages reduced to <PlatformAccess /> - Clarify Code Mode vs domain-specific servers in MCP section copy - Remove agentName interpolations throughout * chore: fix CSS formatting

feat: add agent setup prompt for AI agent onboarding (#30281) * feat: add agent setup prompt for AI agent onboarding - Add src/content/agent-setup/prompt.md as single source of truth for agent instructions (install Skills, MCP servers, and find next steps) - Serve prompt as text/markdown at /agent-setup/prompt.md via Astro API endpoint - Override Starlight SkipLink slot to inject hidden data-agent-instructions div as first child of <body> on /agent-setup/ only - Add Accept: text/markdown content negotiation to worker for /agent-setup route, proxying to /agent-setup/prompt.md - Inject HTML stop directive as preamble in the hidden div only, keeping the markdown prompt clean * feat: global skills install and restart prompt for MCP servers * feat: per-agent MCP config paths, global skills, fix OAuth note * feat: improve agent setup prompt UX and fix opencode mcp install * feat: add restart and auth instructions to setup completion block * feat: move restart notice inside ascii block * feat: add per-agent MCP auth commands inline with install steps * fix: remove claude mcp auth, OAuth triggers on first tool call for Claude Code * feat: update agent setup with plugin installs, new MCP servers, and refreshed prompts * feat: include remaining agent setup component and page changes * [agent-setup] revamp homepage copy block and /agent-setup index page - Add PromptCopyBlock component with dark terminal-style UI, copy button, and icon swap on success - Add two-path card layout to /agent-setup/ (Quick setup vs Manual setup) with orange border on quick path card - Remove agent directive hidden div from SkipLink.astro - Remove content negotiation for /agent-setup from worker/index.ts - Update homepage agent section: copy block, secondary link, drop SVG on small screens, rework CTA text * [agent-setup] polish copy block, homepage logos, and path cards - Two-path card layout on /agent-setup/: Quick setup (orange border) vs Manual setup, responsive with flex wrap at 1000px - PromptCopyBlock: remove animation, fix copy button click handler, no-select on text, copy/copied label swap, overflow/truncation fixes - Homepage: smaller logos (60px), gap spacing, remove justify-between, tighten secondary link spacing, pcb-root margin overrides * fix: prettier formatting and homepage agent section margin * [agent-setup] light mode fixes and copy block polish - PromptCopyBlock: theme-aware colors for light mode (no more hardcoded dark) - Remove glow, add subtle orange box-shadow that increases on hover/copied - Fix copy button colors, green copied state, text color via CSS vars - Homepage: even section margin, normal text size/color for secondary link * [agent-setup] increase homepage agent section top/bottom margin * [agent-setup] copy block light/dark mode background and dot polish * chore: remove unnecessary SkipLink override and astro.config registration

Handle index.md redirects to preserve markdown format (#29250) * Handle index.md redirects to preserve markdown format * Filter external redirects from llms.txt and resolve redirect paths * abstract reused code

Use Markdown for Agents for all docs markdown generation (#28921) * first pass refactor * handle redirect in code, not cloudflare rules * use style guide as example for markdown pages * removed application handling on index.md * update llms.txt to not use index.md by default * copy edits * copy edit * llms.txt improvements * dont use MFA abbreviation * fix llms.txt spacing issue * add back removed test * Use correct url for llms.txt and markdown files * remove duplicate llms.txt entries * use index.md links in llms.txt * llms.txt block quote updates * use a better example url in ai tooling page * add index.md to starlightLinksValidator exclude * Update astro.config.ts Co-authored-by: ask-bonk[bot] <249159057+ask-bonk[bot]@users.noreply.github.com> * Update src/pages/llms.txt.ts Co-authored-by: ask-bonk[bot] <249159057+ask-bonk[bot]@users.noreply.github.com> * Fix llms.txt section labels sometimes being incorrect --------- Co-authored-by: ask-bonk[bot] <249159057+ask-bonk[bot]@users.noreply.github.com>

fix: preserve dots in directory collection entry IDs and update 1.1.1.1 products frontmatter (#30050) * fix: preserve dots in directory collection entry IDs The glob loader uses github-slugger to generate entry IDs, which strips dots from filenames. This means 1.1.1.1.yaml gets stored with ID '1111' rather than '1.1.1.1', causing getEntries() to return undefined when frontmatter references products: [1.1.1.1]. Add a custom generateId to the directory collection that strips only the file extension, preserving the filename as-is. 1.1.1.1 is the only directory entry whose filename differs from its slugified form. Also add products: - 1.1.1.1 (and missing descriptions) to all 38 pages under src/content/docs/1.1.1.1/ so the product is correctly associated for search metadata, tutorial cross-listing, and the /resources filter. * add back products field

Use pnpm (#29624) * use pnpm * use v5 pnpm actions * devEngines tweaks * update lock file * add unist-util-visit-parents which was only working before as a transative property * more transitive dependencies * more transitive deps * add builtDependencies * use only node 24 * update pnpm lock * Another transitive dep * skip install of reviewdog/action-eslint * more transitive deps and * workaround for reviewdog action that uses npm internally * add .astro to .prettierignore * fix format checks * pnpm dlx tsm * use pnpm exec in actions where we would otherwise need to use local dependencies * pin tsm dependency * specify pnpm v10 in workflows to match devEngines * test lint annotations * check format check fail * replace remaining pnpx * revert formatting check error * remove only-allow from dependencies * update pnpm lock * fix an npx we missed * remove devEngines.packageManager * update relevant docs/comments with pnpm * update npm-to-pnpm skill to install pnpm and remove package-lock * use --frozen-lockfile in npm-to-pnpm skill

feat: implement WebMCP API to expose site tools to AI agents (#30208) * feat: implement WebMCP API to expose site tools to AI agents * chore: fix prettier formatting on webmcp files * refactor: extract Algolia constants to src/util/algolia.ts * formatting

chore: add CODEOWNERS check action and auto-triage Bonk job (#30743) * chore: add CODEOWNERS check action and auto-triage Bonk job * chore: fix prettier formatting in check-codeowner action * chore: add explicit permissions to check-codeowner job * Update .github/actions/check-codeowner/index.ts Co-authored-by: ask-bonk[bot] <249159057+ask-bonk[bot]@users.noreply.github.com> * fix: prettier-compatible regex and literal newline in check-codeowner action --------- Co-authored-by: ask-bonk[bot] <249159057+ask-bonk[bot]@users.noreply.github.com>

respect robots.txt in llms.txt generation (#29230)

feat: custom prettier plugin to correctly format inline MDX components (#29770)

chore: pin GitHub Actions to commit SHAs (#29583) Pins all third-party and GitHub-owned actions to immutable commit SHAs with inline version comments for readability. ask-bonk intentionally remains on @main as it is actively developed and frequently updated.

[DNS] Fix footnotes inside Tabs panels generating duplicate headings (#30459) * [DNS] Fix footnotes inside Tabs panels generating duplicate headings Convert [^n] footnotes in internal-zones-conditions and internal-dns-view-conditions partials to plain text <sup> footnotes. Adds semgrep rule to catch [^n] syntax inside <Tabs> blocks, and updates the footnotes style guide page with a caution against this pattern. * renumber footnote 20 -> 1

fix: link validation failures for fork PRs (#29823)

Improve llms.txt documentation and sidebar LLM Resources (#29450) * cleanup sidebar * fix headers * Update AgentDirective comment to clarify markdown visibility * Add explicit type annotations and replace Object.groupBy with Map * Update llms.txt documentation with clearer guidance and reorder methods * Remove Developer platform condition from llms resources sidebar