Skillsync launches to make AI coding agent sessions portable across Claude Code, Cursor and Codex

Skillsync launches to move AI agent sessions between coding tools

Skillsync, a startup from Y Combinator's W26 batch, has launched on Hacker News with software that makes AI chat sessions portable across coding agents. According to founders Nars and Nishant, the tool converts an entire session — messages, reasoning traces and tool-call history — from one agent's on-disk format into another's, so developers can continue the same working thread in a different tool instead of starting over.

The problem it targets

The launch post argues that most developer work with coding agents now exists as conversations, stored locally but in formats specific to each tool. Because Claude Code, Codex, Cursor and others each write sessions differently, switching agents means abandoning accumulated context, skills and memory. The founders say this lock-in is invisible until it becomes expensive: people often realise they are trapped only after months of work has piled up in a format tied to a single tool.

The pair previously built an open-source payment orchestrator that let merchants route across many processors instead of depending on one, a project they say has more than 30,000 GitHub stars, and they frame Skillsync as the same interoperability playbook applied to agent context. They came into YC with a different idea, they write, before the problem became unavoidable in their own agent-heavy workflow.

An open-source conversion engine at the core

The core is txcript, an open-source Rust engine on GitHub that translates a session between agents' native storage formats, mapping the conversation, the model's reasoning and its tool history. The founders compare it to general-purpose converters like ffmpeg or pandoc, but for agent sessions rather than media or documents.

Around that engine sits a local-first desktop app (currently Mac-only), plus a CLI and an MCP server. The app gathers sessions from the scattered folders where agents keep them into one searchable, readable view, and breaks down what a session is carrying — including which loaded skills the agent actually used — so outdated context stands out. Skills and memory are stored as portable, human-readable markdown that the user owns, exposed to any agent over MCP for search and selective retrieval. Everything runs on the local machine, according to the founders, except content explicitly shared to team workspaces, which sync sessions across a group.

How early users are applying it

The launch post describes three patterns among early users: moving a session between Claude Code, Codex and Cursor mid-task, including when a usage limit on one tool forces a switch; handing a browser-based Claude chat that turns into something worth building over to Claude Code or Codex; and pooling research sessions in shared workspaces so teammates and their agents can build on each other's work instead of repeating it.

The company's website carries testimonials that should be read as marketing but do indicate demand: one YC W26 founder claims to have moved about 7 GB of Cursor sessions to Claude Code and continued them within minutes, while another says heavier new models force switching between tools several times a day.

Why it matters

As coding agents become primary development surfaces, the session — not just the repository — is turning into the unit of work worth owning. Locked, tool-specific formats reduce that work to a disposable log; portability turns it into a durable asset that can be searched, audited, handed to a colleague or fed to a different model. An open-source translation layer is also a neutral substrate: if txcript gains adoption, it lowers switching costs and could pressure agent vendors toward interoperable session formats. The risks are real too — translation depends on storage formats that vendors can change at any time, and fidelity across agents with different capabilities is not guaranteed. Even so, the launch highlights a genuine gap in a fast-growing tool category, and a bet that context, like payments and media files before it, eventually wants to be portable.