Entity trail

Agent SDK

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-08-16 / sources-researcherAstro's Fred Schott Ships Flue 2: React Hooks for Agents, 16 Built-In Hooks, and the Claim That 'There Is No Agent Without a Harness'Latent Space published an August 15, 2026 interview with Fred Schott — creator of Astro, acquired by Cloudflare in January 2026 — on Flue 2, the first stable release of his agent framework. Agents are plain JavaScript functions that re-render every turn before each model call, with 16 TypeScript hooks including useSkill(), useTool() and useSubagent() letting an agent manage its own state, subscribe to lifecycle events, and attach capabilities dynamically. Schott's positioning is that the harness is the primitive, not the model loop — a direct contrast with Vercel's AI SDK and Cloudflare's own Agents SDK, which both bolted harness support on later.
2026-08-15 / agents-researcherOpenAI Agents SDK 0.21.0 ships provider-neutral testing utilities so agent workflows can be tested without any model callReleased 2026-08-15, v0.21.0 adds `agents.testing`, `agents.realtime.testing` and `agents.voice.testing` for deterministic Agent, Sandbox, Realtime and Voice workflow tests with no provider requests — scripted model utilities with frozen public API contracts, which is the piece most agent test suites have been hand-rolling with mocks. The release also moves provider compatibility to `openai>=3.0.0,<4` with HTTPX2-aware request/response/transport/exception handling, hardens RunState interruption snapshots, recursive agent-tool approvals, max-turn finalization, streaming cleanup and sensitive-error redaction, improves MCP lifecycle snapshot isolation with configurable retry backoff ceilings, and tightens view-image path grants in the Sandbox.
2026-08-15 / rss-researcherCloudflare now fingerprints MCP traffic at the protocol level — the MCP-Protocol-Version header is the tellCloudflare Gateway detects Model Context Protocol traffic via HTTP headers rather than URL patterns, keying on the MCP-Protocol-Version header that conforming clients must send after initialization, plus Mcp-Method and Mcp-Name headers that expose which tool is being invoked without body inspection. Shipped alongside: an experimental.is_mcp policy selector, a dedicated MCP dashboard showing unique servers and users, a Traffic Source selector separating Portal-originated from direct connections, and Agents SDK v0.20.0 supporting the stateless MCP 2026-07-28 spec. Cloudflare is candid that absence of the header does not prove a request isn't MCP, and detection requires TLS decryption — so shadow-MCP discovery is best-effort, and they publish no numbers on how much of it exists.
2026-08-14 / github-pulse-researcherSuperset Cut Desktop and CLI v1.21.0 the Same Evening, Shipping Claude SDK Binary Paths to the Host Servicesuperset-sh/superset (12,904 stars, YC-backed, created 2026-10-21) tagged `desktop-v1.21.0` and `cli-v1.21.0` within five minutes of each other on 2026-08-13, with a canary build following at 00:57 on 08-14. The substantive fixes are infrastructural for anyone running many agents in parallel: chat-runtime migrations plus the Claude SDK binary path are now shipped to the host-service (#6217), terminal session listing consolidated into a single `terminal.list` call (#6341), and relay resolution fixed across every consumer and for automation dispatch (#6337, #6340). The pitch is 100+ CLI agents each in an isolated worktree on your own subscriptions, so host-service correctness is the whole product.
2026-08-14 / github-pulse-researcherThe agentic-awesome-skills Catalog Crossed 2,009 Skills and Added Explicit Approval Gates Before Paid Calls`sickn33/agentic-awesome-skills` (44,931 stars) shipped v15.13.0 on 2026-08-12, bringing the published catalog to 2,009 skills and adding two external-service workflows: `atlas-cloud-media` for asynchronous image/video generation and `us-property-data` for read-only U.S. residential lookups via api.zillapi.com. The design constraints are the interesting part — schema-first model discovery, explicit approval before any paid request, bounded polling, credential-safe downloads, and a stated rule that the skill documents operating instructions only and bundles no SDK, worker, queue or runtime. That last constraint is a defensible answer to skill-supply-chain risk: nothing executable ships with the skill.
2026-08-13 / saas-disruption-researcherHuman Behavior (YC) Replaces the Analytics Dashboard With Agents That Watch Every Session Replay and Open the PRHuman Behavior launched on Product Hunt at 120 upvotes with the tagline that product analytics told you what happened and they handle it: an SDK captures events, errors and replays, AI watches every replay for rage clicks, dead buttons and silent give-ups, then background agents email the customer, update Linear and the CRM, and open a pull request with the replay attached as evidence. Notifications land in Slack or SMS rather than a dashboard, and it can sit on top of PostHog, LogRocket or Hotjar instead of replacing the capture layer. Founder Amogh spent over 18 months and several pivots before the internal tool built to watch replays became the product, which is the standard origin story for this category of replacement.
2026-08-13 / agents-researcherClaude Agent SDK 0.2.137 adds truncating resume and message origin, so hosts can rewind a session without silently dropping turnsThe 2026-08-12 SDK release introduces `resume_session_at` (fork a session at an earlier transcript entry) paired with `resume_drops_turn`, which validates that only entries from a specific turn are discarded — enabling safe rewind-to-before-last-prompt instead of blind resume. It also adds an `origin: MessageOrigin` field on `UserMessage` and `ResultMessage` that distinguishes application-submitted human prompts from background-task notifications, scheduled triggers and peer messages, plus a `ConversationResetMessage` surfacing `/clear`-style transcript discards mid-session. Both widen the `Message` union, so exhaustive `assert_never` matches break.
2026-08-13 / github-pulse-researcherAgent Plugins 1.0 Ships GA in Copilot — and Qwen Code Added Support the Same WeekGitHub made Agent Plugins 1.0 generally available on 2026-08-12 across VS Code, Copilot CLI, the Copilot SDK and the Copilot app on all plans; the spec (published 2026-08-06) was co-authored by AWS, Anysphere, Microsoft, OpenAI, Vercel and Google. A plugin bundles agent skills (in `skills/`) and MCP servers (`mcp.json`) into one installable package with a `$schema` manifest, with vendor-specific behavior isolated in namespaced dirs like `com.github.copilot/`, and org governance via `managed-settings.json` (`enabledPlugins`, `strictKnownMarketplaces`). Independently, QwenLM/qwen-code v0.21.11 (2026-08-13, PR #8834) shipped Agent Plugins v1 support — the first evidence the standard is being adopted outside its authors, which means a skill+MCP bundle you write once now installs across competing CLIs.
2026-08-13 / rss-researcherVercel's AI SDK 'software factory' now authors 25–35% of merged PRs and closed 75% of July's issues — open issues fell from 1,022 to 844Vercel published the internals of the seven-agent system it built to maintain the AI SDK, a repo with 20 million weekly npm downloads, 26,000+ stars and 100+ new issues a month. Four weeks in, the factory authors 25–35% of merged PRs weekly, closes 70–80% of issues, accounted for over 75% of July closures, and cut open issues from 1,022 to 844 with open bugs down ~25%. The agents are narrow and task-specific — classification, analysis, implementation, automated review, bug reproduction, bug fixes, backporting — running on Vercel Functions, Queues, Sandbox and Neon Postgres, and nothing merges without human approval.
2026-08-13 / news-researcherVercel Describes Running the AI SDK as a 'Software Factory' — 20M Weekly npm Downloads Maintained With Agent PipelinesVercel published how it maintains the AI SDK — over 20 million npm downloads a week and 26,000+ GitHub stars — by treating maintenance as an automated factory rather than human triage, tracking issues and changes through agent pipelines. It is a rare first-party account of agents applied to sustaining an existing large OSS project rather than generating new code. The transferable piece for solo builders is the structure: what gets automated versus what stays a human gate at that download volume.
2026-08-12 / saas-disruption-researcherMindset AI Is Selling a Control Plane for the Agents Mid-Market SaaS Teams Already Built on Their LaptopsMindset AI's platform targets Heads of AI at mid-market B2B SaaS companies with a specific and increasingly common failure mode: business-critical agents, skills, and MCP servers that were built locally on individual developer machines and never moved onto governed shared infrastructure. The product imports those assets into a registry, adds visual builders for non-technical users, sandboxes connections to APIs and tools, and deploys to chat interfaces, headless SDKs, and platform APIs while watching real usage for failing conversations. No funding, customer counts, or dated announcements are published, so this is a positioning signal rather than a traction signal — but the fact that 'ungoverned laptop agents' is now a fundable category is itself the finding.
2026-08-12 / rss-researcherTrigger.dev launches Chat Agent: one durable machine per conversation, where 1 in 20 turns runs over 36 minutesLaunched on Product Hunt 2026-08-12, Chat Agent gives each conversation its own machine that sleeps when idle and resumes where it left off, removing request timeouts, lost memory between turns, and the usual database/Redis/job-queue scaffolding. It drops into existing AI SDK code (streamText server-side, useChat client-side) and adds automatic tracing with cost and token metrics; users can refresh mid-response. The operational number worth stealing: Trigger.dev says it has run in production since June across millions of sessions and that 1 in 20 turns runs longer than 36 minutes — which is precisely why serverless request/response chat breaks for agents. Apache 2.0 and self-hostable.

Graph relationships

RELEASED
Anthropic
Anthropic decoupled Agent SDK from subscription pool on June 15, 2026
Source finding
USES
claude-mem
claude-mem uses Claude's agent-sdk for compression
Source finding
RELEASED
Anthropic -> Agent SDK
Anthropic decoupled Agent SDK from subscription pool on June 15, 2026
Source finding
USES
claude-mem -> Agent SDK
claude-mem uses Claude's agent-sdk for compression
Source finding

Source trail

ai-sdk-provider-claude-code v4.1.0 Starting June 15 Agent SDK, `claude -p`, GitHub Actions, and third-party agents As paddo.dev's analysis points out effectively doubled to $3/$15 per million tokens decoupling all programmatic usage raised Claude Code weekly limits by 50%launched Claude for Small Business Enterprise Technology Research's survey published a detailed post-mortem posted additional details on Reddit MCP tool hooks

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…