Session Portability Becomes a Market Layer: Skillsync (YC W26) Unifies Agent Transcripts Across Harnesses

Session Portability Becomes a Market Layer: Skillsync (YC W26) Unifies Agent Transcripts Across Harnesses

A new interoperability layer is forming between agent harnesses. Skillsync (Launch HN, Sept 17, 2026, YC W26, 65 points / 58 comments) builds a unified schema for AI agent sessions — Claude Code, Codex, Antigravity, claude.ai chats — so that a session started in one agent can be continued in another "as if the entire conversation history was there the whole time." The core library (txcript) is open-source on npm; past sessions are also exposed via MCP, and the desktop app syncs sessions across machines over SSH.

The portability thesis, in the founders' words

Co-founder cat-whisperer, answering the obvious objection ("just tell the agent to keep a work journal in markdown"):

"A journal comes out of a prompt, so it records one way of looking at the session. Whatever that prompt didn't ask about is gone… Being able to unify the schema lets you do things notes can't. You can continue a teammate's session on your own machine, or pick up a claude.ai conversation in Claude Code or Codex. The new agent gets the transcript: all the decisions in the trajectory, not a summary of it." — comment

Co-founder Narsagna states the anti-lock-in value prop directly:

"You should not get locked into one agent simply because you built up a conversation history + memory with that agent.¹" — comment

The debate that matters for research tools: transcript vs. distilled note

The thread's sharpest exchange is a proxy fight over what the durable artifact of agent work is. pjm331 argued syncing "is trivially solved by just telling the agent to keep a work journal in markdown" (comment); agentdev001 shot back:

"Sure, you can completely destroy the session by distilling it into a markdown — but thats like explaining what you did yesterday, vs having an immutable log of what you did yesterday." — comment

This is exactly the design tension in long-horizon research tools: the immutable trajectory preserves evidence and re-interpretability; the distilled note is what a human (or a later cycle) can actually reuse. Hey, Lefty's own architecture — findings as living notes plus a grep-searchable raw history log — is a bet on keeping both layers. Skillsync's founders land on the same side: keep the raw transcript as the source of truth, distill on demand.

Evidence a standard is waiting to happen

The comments are a census of people who already built private versions of this: btown's team runs a homegrown /resume skill that forces Claude Code to ingest a colleague's full session JSONL ("mind melds… including their temporary scripts and unmerged changes," comment); yoavfr built usecontinuo with "my own translation layer"; bhkdotdev's cross-harness test tool is pained by transcript formats that "are all so different" (comment); sdesol wants it for session discovery tooling; swyx: "kudos making the core open source. if you didnt do it, i wouldve" (comment). btown also flags the unsolved governance layer: team-shared "skills" carry guardrails and local state, and "who watches the watchmen?" (comment).

What it means for the market map

Three takeaways for Market Map & Positioning: Where Does Hey, Lefty Fit?:

1. Workspace lock-in is now explicitly under attack from below. Destination vs. Backend: The Research-Tool Market Bifurcates — and Both Layers Converge on MCP framed the destination layer's moat as the persistent workspace. Skillsync's whole pitch is that session/history lock-in is a bug users will pay to escape. A destination layer that can't interoperate will be argued out of the stack.
2. The portability instinct rhymes with the memory layer. OKF Agent Memory: Git-Native Persistent Project Memory for AI Research and Coding Agents showed git-native agent memory emerging bottom-up; Skillsync shows session-transcript portability emerging the same way. Persistent research memory should be portable and inspectable by default — that's becoming table stakes, not differentiation.
3. The differentiation moves up a level. If transcripts are portable and models are commodities, what's left is what the destination layer adds: multi-source orchestration over research MCP backends (Sacra: Premium Private Market Research as Agent-Ready Infrastructure, Institutional MCP Financial Data Servers: FactSet, S&P Global, and PitchBook Unbundle), evidence standards, and the distillation pipeline (Panel and the Agentic Research Workspace: When the Agent Builds Its Own UI shows the workspace mechanic itself commoditizing). Hey, Lefty should treat session portability as an input format to embrace, not a feature to fight.

Open risk: harness vendors have a precedent for killing exactly this — cat-whisperer notes "isn't that what they did to opencode? when they were using the subs by reverse engineering their APIs" (comment). Registered as a watch.