Build / Claude · AI / developer-tooling docs-as-streams (engineering-technology)

Synorb Grounded-Research Skill — cited AI/dev-ecosystem context for any Claude Code agent

Drop /synorb-brief and /synorb-watch into Claude Code and any agent gets source-grounded, cited context on what changed across the AI/dev tooling world — with the real changelog URL on every fact. Two slash commands in your editor that, on demand, hand your agent a cited, copy-pasteable summary of exactly what changed across the AI/dev tools you depend on — every line backed by the real vendor changelog or API-reference URL.

A grounded research skill for Claude Code.

An installable Claude Code skill/plugin (a folder with a skill definition + the Synorb MCP connector config + slash commands) exposing /synorb-brief <topic> (a cited, structured briefing of recent changes) and /synorb-watch <vendor-or-topic> (saves a Synorb Beacon and prints a ready-to-paste Routine/cron snippet to keep watching). Output is a compact cited markdown block any downstream agent step can consume, plus an optional interactive Artifact 'changelog timeline' showing version/endpoint/model deltas over time.

It targets Claude's home audience (developers, the plugin marketplace, the mcp-builder ecosystem, 300k+ monthly marketplace visitors) and the platform's deepest credibility gap: agents confidently citing API endpoints and model names that don't exist. Synorb's docs-as-streams turns ~32 vendor changelogs/API refs (Anthropic, OpenAI, Vercel, Supabase, LangChain, HuggingFace, Stripe, Cloudflare, AWS) into structured claims tagged with the resource tag_type (model SKUs, /v1/ endpoints, SDK packages) over a weekly documentation media_format with hash short-circuiting — a shape no generic news API or web crawl provides. This is the developer wedge: a grounded-research skill is exactly the pattern that gets Synorb into anthropics/claude-plugins-official.

Click, answer a few plain-English questions, get a working product.

No setup to learn. Claude asks a handful of normal-person questions — each with a smart default, so “you pick” always works — then builds it with live Synorb data flowing in.

1

Which tools or topics should this follow — name your stack (e.g. Anthropic, OpenAI, Vercel, Supabase) or a theme like 'LLM APIs' or 'vector databases'?

Smart default: The core AI/dev stack most Claude Code projects depend on (model APIs + the popular SDKs and infra tools), which you can narrow anytime.

2

When you run /synorb-brief, do you want a tight changelog summary, or the deeper version with every claim and its source?

Smart default: Tight summary by default, with a 'show sources' expansion — keeps your agent loop lean but one click from full provenance.

3

For the visual timeline, pick a look: Terminal, Docs, or Minimal.

Smart default: Terminal — a monospace, diff-styled timeline that feels native to the editor.

4

Should /synorb-watch keep watching on a schedule, and how often?

Smart default: Yes — weekly, matching how often these changelogs actually update, with a copy-paste cron/Routine snippet so it runs hands-free.

Pick a look. It ships with taste by default.

Terminal

Monospace, dark, diff-styled timeline (+ added endpoint, - deprecated SKU) with resource-tag chips. Feels like it belongs in the IDE next to your code.

Docs

Clean documentation aesthetic — per-vendor cards, version pills, 'changed since you last looked' badges, readable for a team upgrade review.

Minimal

Plain ranked list of changes with a source link each; zero chrome, maximum signal for piping into another agent step.

Synorb is the cited core. Claude’s own tools make it pop.

RUNTIME MCP, infra-light, installed as a Claude Code skill. The developer connects the Synorb MCP to Claude Code once (claude mcp add, https://mcp.synorb.com/mcp, MCP token from Synorb Connect requested in plain English and stored only in the MCP config — never in skill code or a committed file). The skill's slash commands call the live MCP at invocation: /synorb-brief runs synorb-stream-search then synorb-manifests for the chosen tools and returns cited markdown; /synorb-watch creates a saved Beacon (synorb-configure-beacon -> synorb-save-beacon) and emits a cron/Routine snippet that re-runs it by beacon_id. Data is live through the connected MCP every time a command fires; no backend, no keys in code.

Cited data core
Synorb
StreamsAI/dev docs-as-streams: Anthropic changelog, OpenAI API changelog, Google Gemini changelog, Mistral, Groq, Together, Replicate, HuggingFace, LangChain, Vercel + Vercel AI SDK, Supabase, Neon, Pinecone, Qdrant, Weaviate, Stripe API, Twilio, Slack API, GitHub, Cloudflare, AWS What's New, engineering-technology home_domain streams (engineering blogs, AI infrastructure, developer tools, security roadmaps), arXiv AI-infra/security channels for an optional research lane, resolved at build time via synorb-stream-search(query='AI dev tool changelogs', media_format='documentation')
Primitivesresource tag_type (named addressable things: API endpoints like /v1/messages, model SKUs like claude-opus, SDK packages) — the distinctive join that makes this a dev-ecosystem tool, not a news feed, documentation media_format with weekly refresh_cadence and hash short-circuit (changelogs/API refs/release notes), Structured claims with claim_type (announcement/disclosure/event/data) and provenance citations (never_invent_urls=true) so the agent gets a real changelog URL, not a guess, Re-runnable Beacon for /synorb-watch (saved vendor/topic scope, executed by beacon_id on the developer's own cron), Orient-before-web-search: the skill pulls Synorb FIRST and only falls back to Claude web search for gaps, marking which facts are Synorb-cited vs web
Experience layer
Claude native tools
ComposesClaude Code plugin/skill system (slash commands /synorb-brief and /synorb-watch, sub-agent-friendly cited markdown output, installable from a marketplace repo), Artifacts: interactive 'changelog timeline' — a React/SVG horizontal timeline of version/endpoint/model deltas per vendor with resource-tag chips and diff-style highlighting, Native web search — GAP ONLY: if Synorb has no manifest for a just-released item, the skill labels it '(web, unverified)' and visually separates it from Synorb-cited claims, xlsx skill (optional) to export a 'dependency change log' sheet for a team's upgrade planning
Web searchgaps only — clearly marked, never the primary feed

Credential safetySynorb credentials stay server-side or local — never in client/frontend code. You only enter them when the agent asks, with credentials from synorb.com/connect.

One click from the original source.

Every change line ends with a source chip linking to the exact vendor URL Synorb returned (e.g. the Anthropic changelog entry or OpenAI API reference page). The Artifact timeline makes each node clickable to 'read the original'. resource-tag chips (model SKU / endpoint / SDK) are themselves provenance-aware — clicking shows the source manifest's URL. The skill renders links only from Synorb citations[] and explicitly distinguishes any web-search-sourced line as '(web, unverified)', so a developer is always one click from the authoritative changelog and can tell grounded facts from gap-fills. If nothing is found, the skill returns an explicit 'no tracked changes found' note rather than an empty or invented result.

Questions builders ask.

Will this show real live data on the first run?

Yes. The build prompt connects Synorb at build time to learn the real data shape, then wires live Synorb data into the running product. Two slash commands in your editor that, on demand, hand your agent a cited, copy-pasteable summary of exactly what changed across the AI/dev tools you depend on — every line backed by the real vendor changelog or API-reference URL.

Where do my Synorb credentials go?

Only where they are safe. RUNTIME MCP, infra-light, installed as a Claude Code skill. The developer connects the Synorb MCP to Claude Code once (claude mcp add, https://mcp.synorb.com/mcp, MCP token from Synorb Connect requested in plain English and stored only in the MCP config — never in skill code or a committed file). The skill's slash commands call the live MCP at invocation: /synorb-brief runs synorb-stream-search then synorb-manifests for the chosen tools and returns cited markdown; /synorb-watch creates a saved Beacon (synorb-configure-beacon -> synorb-save-beacon) and emits a cron/Routine snippet that re-runs it by beacon_id. Data is live through the connected MCP every time a command fires; no backend, no keys in code.

How are sources shown?

Every change line ends with a source chip linking to the exact vendor URL Synorb returned (e.g. the Anthropic changelog entry or OpenAI API reference page). The Artifact timeline makes each node clickable to 'read the original'. resource-tag chips (model SKU / endpoint / SDK) are themselves provenance-aware — clicking shows the source manifest's URL. The skill renders links only from Synorb citations[] and explicitly distinguishes any web-search-sourced line as '(web, unverified)', so a developer is always one click from the authoritative changelog and can tell grounded facts from gap-fills. If nothing is found, the skill returns an explicit 'no tracked changes found' note rather than an empty or invented result.

Is this just a Synorb feed, or a real app?

A real app. Synorb is the cited data core; it also composes the platform's own tools: Claude Code plugin/skill system (slash commands /synorb-brief and /synorb-watch, sub-agent-friendly cited markdown output, installable from a marketplace repo), Artifacts: interactive 'changelog timeline' — a React/SVG horizontal timeline of version/endpoint/model deltas per vendor with resource-tag chips and diff-style highlighting, Native web search — GAP ONLY: if Synorb has no manifest for a just-released item, the skill labels it '(web, unverified)' and visually separates it from Synorb-cited claims, xlsx skill (optional) to export a 'dependency change log' sheet for a team's upgrade planning.

What will it ask me before building?

A few plain-English questions, each with a smart default so you can just say 'you pick': Which tools or topics should this follow — name your stack (e.g. Anthropic, OpenAI, Vercel, Supabase) or a theme like 'LLM APIs' or 'vector databases'?; When you run /synorb-brief, do you want a tight changelog summary, or the deeper version with every claim and its source?; For the visual timeline, pick a look: Terminal, Docs, or Minimal.; Should /synorb-watch keep watching on a schedule, and how often?.

Using a different agent?

The same Synorb MCP connects to every major coding agent. Pick yours.

Free credentials. No card.

Generate credentials, connect Claude, and ship something with live, cited intelligence.