Skillsync Wants to Be GitHub for Your Agent Sessions

Every engineer has one session they wish they could bottle. The bug that ate three days, cracked in twenty minutes. The migration that finally clicked at 1 a.m. Then the terminal closes, the tab count drops by one, and all of it - the dead ends, the working fix, the reason it worked - is gone. Skillsync, a Y Combinator Winter 2026 company, thinks that vanishing act is the most expensive thing happening in software right now.

The pitch fits on a bumper sticker: GitHub for your agent sessions. As teams pour work into AI coding agents - Claude Code, Codex, Cursor and a growing zoo of others - each of those agents runs, produces something, and forgets. The knowledge lives in scattered logs on individual machines. Skillsync's founders, Narayana Aaditya Ganeshkumar and Nishant Joshi, want to catch that knowledge before it evaporates, put it somewhere everyone can search, and let the best sessions become reusable skills any agent on the team can load.

01 / The problemKnowledge that dies at end of session

Software already solved a version of this problem once. Before Git, code lived on individual laptops and in emailed zip files. Version control turned private work into shared, inspectable history. Skillsync's argument is that agent work is now in the pre-Git era: valuable, constant, and completely undocumented. What worked, what didn't, and why is spread across a half-dozen tools and lost the moment a session ends.

The cost is quiet but real. A teammate's agent solves a gnarly problem on Monday; on Thursday, someone three desks over has their agent grind through the same wall from scratch. Multiply that across a company and you get a lot of expensive re-derivation - the same reasoning, paid for again and again.

02 / The productWhat Skillsync actually does

In practice, Skillsync sits underneath the agents a team already runs. You install it, keep working, and your sessions land in one searchable place instead of dying on a laptop. Share any session with a URL. Promote the good ones into skills - packaged, reusable know-how that any agent can load next time it faces a similar task. The company frames the payoff plainly: so every agent builds like your best engineer, and the more your team ships, the smarter your agents get.

Getting started is deliberately unremarkable, which is the point for infrastructure that wants to disappear into a workflow:

03 / The pivotFrom finding engineers to remembering them

Skillsync did not start here. Its first product answered a different frustration the founders knew personally. At Juspay, they helped grow an open-source Rust payments project to more than 20,000 GitHub stars and hundreds of contributors - and still found it tedious to identify and hire their best people. So the original Skillsync turned public GitHub activity into structured skill profiles, letting recruiters search for real capabilities ("WASM compiler experience") instead of resume keywords. No candidate signup required; a browser extension even surfaced inferred skills straight on a GitHub profile.

The through-line is consistent: both products are about making hidden expertise legible. The first found the people. The second captures how the work gets done - and, increasingly, that work is done with an agent in the loop. Hundreds of recruiters and founders used the discovery tool before the team turned its attention to the sessions themselves.

04 / The marketWhere it fits

Skillsync is placing a bet on a direction, not just a feature. As agent-written code becomes a larger share of what ships, a layer that remembers and redistributes agent work looks less like a nice-to-have and more like plumbing. Today that gap gets filled by internal wikis, copy-pasted prompts, and homegrown scripts. Skillsync wants to be the default instead - the searchable home that spans every agent a team touches.

The competitive picture cuts two ways. On the memory-and-skills side, Skillsync bumps against emerging agent-memory tools and every team's instinct to just build something internal. On the older discovery side, the reference points were GitHub-based sourcing tools and traditional recruiting platforms. What Skillsync is trying to own is the connective tissue - the boring, cross-tool layer that no single agent vendor is incentivized to build.

05 / The peopleTwo Rust engineers, one flywheel

Narayana Aaditya Ganeshkumar (Nars) is CEO; Nishant Joshi is CTO. Both come out of Juspay's open-source Rust work, which shows in the company's instincts: deep systems background, a comfort with infrastructure that lives beneath the surface, and a habit of iterating on positioning in public. The team is small - two to three people - which makes the ambition-to-headcount ratio steep. Building a foundational layer for AI coding with a handful of people is either early or reckless, and YC's Winter 2026 batch, plus a reported $500K from Y Combinator and Character Capital, suggests investors are willing to call it early.

Whether the flywheel spins is the whole question. If capturing sessions is frictionless and the reused skills are genuinely good, each shipped feature makes the next one easier, and Skillsync becomes stickier the longer a team uses it. If the captured knowledge is noise, it's another dashboard nobody opens. The idea is clean. The execution is everything.

06 / Will it matterThe ten-year question

Strip away the specifics and Skillsync is a wager on a simple observation: as more work is done by agents, someone has to remember how. Models will keep improving on their own. Memory - shared, searchable, reusable memory across a team - is the part nobody automatically gets. If agent-written software keeps growing, a company that owns that layer has a real seat. If it doesn't, Skillsync is a well-built answer to a question the industry decided not to ask. For now, it's a two-person team with a single curl command, betting on the first.