# Cursor introduces the Agent SDK for building and deploying agents in code [Canonical reading](https://escapemeridian.com/readings/cursor-introduces-the-agent-sdk-for-building-and-deploying-27d4197c) Cursor introduces the Agent SDK, a framework for defining agents in code and deploying them to Cursor infrastructure. Cursor says agents can respond to team events, follow policies teams can inspect and change, and hand work to specialized agents or subagents. Cursor describes kb-reviewer as a team-built example, created in about a day to check every pull request in Cursor's main codebase against lessons in a knowledge base. It flags changes that contradict a recorded lesson, cites the note behind each finding, and encourages engineers to reconcile reported discrepancies. As an Escape Meridian interpretation, this example illustrates Cursor's argument that teams can tailor agents to their workflows and carry prior decisions into code review. The post does not report an independent evaluation of the agent's accuracy or effects. Cursor says the SDK includes evals and event replay. The post describes agents' instructions, tools, connections to outside services, and evals as code that can go through the same version control and review process as other software. It says agents can run on Cursor infrastructure or a team's own infrastructure. How well do agents built with the SDK perform on real team workflows, including whether their findings and policy decisions are reliable? Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source Cursor · Article Original source date: Oct 20, 2026 (2026-10-20T12:00:00.000Z). Date basis: Date stated by the publisher. Added to Escape Meridian: 2026-11-05T03:54:58.277Z. First observed: 2026-10-26T19:21:22.331Z. Retrieved: 2026-11-05T03:54:03.940Z. - [Introducing the Agent SDK · Cursor](https://cursor.com/blog/agent-sdk) ## Limits This is a Cursor product announcement; it offers no independent evaluation of agent accuracy, reliability, or effects on team workflows. The kb-reviewer account describes a Cursor team's own use and does not report measured outcomes. The PGA Tour software development factory example includes a testimonial from PGA America Principal Engineer Michael Gyarmathy, not independent performance evidence. ## Open question How well do agents built with the SDK perform on real team workflows, including whether their findings and policy decisions are reliable? ## Source-stated Cursor announces the Agent SDK as a framework for defining agents in code and deploying them to Cursor infrastructure. > Today, we're releasing the Agent SDK, the framework we use to build our own managed agents like Bugbot, Security Reviewer, and Rollouts. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Source-stated Cursor says agents can respond to team events and follow inspectable, changeable policies. > They respond to the events your team already handles and follow policies your team can inspect and change. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Source-stated Cursor presents kb-reviewer as a team-built agent that checks pull requests against the team's recorded lessons and cites supporting notes. > It now runs on every PR in their main codebase, flags changes that contradict a recorded lesson, cites the note behind each finding, and encourages engineers to reconcile the discrepancies it reports. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Source-stated Cursor says the SDK supports evals and event replay, and describes deployment both on Cursor infrastructure and on a team's own infrastructure. > The Agent SDK also supports evals and event replay out of the box, allowing for rapid development with assurances that critical workflows execute the way you expect. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Escape Meridian interpretation The post's examples illustrate Cursor's claim that agents can be tailored to team workflows, but do not independently establish their accuracy or effects. > With the Agent SDK, they built kb-reviewer in about a day. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Source and contributor context Rustam Lalkaka · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media Media declared by the original publisher; availability checked separately from the editorial reading. - [image · Cursor](https://ptht05hbb1ssoooe.public.blob.vercel-storage.com/OG%20Images/agent-sdk-blog-og-WlJMIsDgzkUUfvyf1R9PJSIfytr7Sx.png) ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Evals Structured tests of what an AI system can do or how it behaves, run before or after release. - [Definition reference](https://developers.google.com/machine-learning/glossary#evals) ## Cursor Cursor makes an AI coding environment and publishes evaluations of agents working on software tasks. - [Explore this context](/entities/cursor) - [Reference](https://cursor.com/) - [Reference](https://cursor.com/evals) - [Reference](https://cursor.com/blog/real-time-rl-for-composer) ## Editorial provenance Version 1. Publication ID: ed06e261-2ba2-43f2-8ec8-3b23447c51b8. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-05T03:54:58.277Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 383d49bbd0541bb5f9951f9ea040031d5b76541e0cc5645557371e780c212316. Source text hash: b39117e61867f46e7e10198ffa583c7c8f89b2c239cee2b3af6715f8ed6ba606. ## Revision and source-check history ### Version 1 — 2026-11-05T03:54:58.277Z Cursor introduces the Agent SDK, a framework for defining agents in code and deploying them to Cursor infrastructure. Cursor says agents can respond to team events, follow policies teams can inspect and change, and hand work to specialized agents or subagents. Cursor describes kb-reviewer as a team-built example, created in about a day to check every pull request in Cursor's main codebase against lessons in a knowledge base. It flags changes that contradict a recorded lesson, cites the note behind each finding, and encourages engineers to reconcile reported discrepancies. As an Escape Meridian interpretation, this example illustrates Cursor's argument that teams can tailor agents to their workflows and carry prior decisions into code review. The post does not report an independent evaluation of the agent's accuracy or effects. Cursor says the SDK includes evals and event replay. The post describes agents' instructions, tools, connections to outside services, and evals as code that can go through the same version control and review process as other software. It says agents can run on Cursor infrastructure or a team's own infrastructure. How well do agents built with the SDK perform on real team workflows, including whether their findings and policy decisions are reliable? ## Original source Cursor · Article Original source date: Oct 20, 2026 (2026-10-20T12:00:00.000Z). Date basis: Date stated by the publisher. Added to Escape Meridian: 2026-11-05T03:54:58.277Z. First observed: 2026-10-26T19:21:22.331Z. Retrieved: 2026-11-05T03:54:03.940Z. - [Introducing the Agent SDK · Cursor](https://cursor.com/blog/agent-sdk) ## Limits This is a Cursor product announcement; it offers no independent evaluation of agent accuracy, reliability, or effects on team workflows. The kb-reviewer account describes a Cursor team's own use and does not report measured outcomes. The PGA Tour software development factory example includes a testimonial from PGA America Principal Engineer Michael Gyarmathy, not independent performance evidence. ## Open question How well do agents built with the SDK perform on real team workflows, including whether their findings and policy decisions are reliable? ## Source-stated Cursor announces the Agent SDK as a framework for defining agents in code and deploying them to Cursor infrastructure. > Today, we're releasing the Agent SDK, the framework we use to build our own managed agents like Bugbot, Security Reviewer, and Rollouts. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Source-stated Cursor says agents can respond to team events and follow inspectable, changeable policies. > They respond to the events your team already handles and follow policies your team can inspect and change. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Source-stated Cursor presents kb-reviewer as a team-built agent that checks pull requests against the team's recorded lessons and cites supporting notes. > It now runs on every PR in their main codebase, flags changes that contradict a recorded lesson, cites the note behind each finding, and encourages engineers to reconcile the discrepancies it reports. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Source-stated Cursor says the SDK supports evals and event replay, and describes deployment both on Cursor infrastructure and on a team's own infrastructure. > The Agent SDK also supports evals and event replay out of the box, allowing for rapid development with assurances that critical workflows execute the way you expect. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Escape Meridian interpretation The post's examples illustrate Cursor's claim that agents can be tailored to team workflows, but do not independently establish their accuracy or effects. > With the Agent SDK, they built kb-reviewer in about a day. - [Supporting source](https://cursor.com/blog/agent-sdk) ## Source and contributor context Rustam Lalkaka · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media Media declared by the original publisher; availability checked separately from the editorial reading. - [image · Cursor](https://ptht05hbb1ssoooe.public.blob.vercel-storage.com/OG%20Images/agent-sdk-blog-og-WlJMIsDgzkUUfvyf1R9PJSIfytr7Sx.png) ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Evals Structured tests of what an AI system can do or how it behaves, run before or after release. - [Definition reference](https://developers.google.com/machine-learning/glossary#evals) ## Cursor Cursor makes an AI coding environment and publishes evaluations of agents working on software tasks. - [Explore this context](/entities/cursor) - [Reference](https://cursor.com/) - [Reference](https://cursor.com/evals) - [Reference](https://cursor.com/blog/real-time-rl-for-composer) ## Editorial provenance Version 1. Publication ID: ed06e261-2ba2-43f2-8ec8-3b23447c51b8. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-05T03:54:58.277Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 383d49bbd0541bb5f9951f9ea040031d5b76541e0cc5645557371e780c212316. Source text hash: b39117e61867f46e7e10198ffa583c7c8f89b2c239cee2b3af6715f8ed6ba606. ### uncertain — 2026-11-05T03:54:58.277Z — anthropic/claude-opus-5.5 Publication: ed06e261-2ba2-43f2-8ec8-3b23447c51b8 attribution — pass: Claims are attributed to Cursor's announcement. The testimonial is correctly credited to Michael Gyarmathy, Principal Engineer at PGA America. The kb-reviewer example is credited to a Cursor team, and the interpretive sentence is explicitly labeled as Escape Meridian interpretation. > Michael Gyarmathy Principal Engineer, PGA America summary — pass: The reading leads with the actor (Cursor) and the development (the Agent SDK, described plainly as a framework for defining agents in code). It conveys the tailoring argument, the kb-reviewer example, evals, version control and deployment options, and notes the missing independent evaluation. Terms such as evals and event replay could use tooltips, but they do not block understanding. The question follows from the evidence without leading. It does not mention that Cursor builds its own Bugbot and similar agents with this framework, though a claim covers that. > Until now, building one has generally meant adopting a product designed around someone else's workflow or starting from scratch. date — pass: The originalPublished date of 2026-10-20, based on the publisher page, matches the feed date. Nothing indicates an earlier first publication. The draft makes no conflicting date claim. support — uncertain: The summary says kb-reviewer checks every pull request in Cursor's main codebase. The source says a team within Cursor runs it on every PR in their main codebase, which may mean that team's codebase rather than Cursor's overall main codebase. The smallest repair is to say 'the team's main codebase.' The other claims and limitations are supported and appropriately hedged. > It now runs on every PR in their main codebase, flags changes that contradict a recorded lesson media — pass: The selected media is the verified direct link to the original Cursor blog post. It is the most direct way to access the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/27d4197c23d465e16094baaf79de1ba3c2a04d66eca773d55729f56f6c036195/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/cursor-introduces-the-agent-sdk-for-building-and-deploying-27d4197c.json)