Friday, 28 August, 6 PM: a blank terminal, LatentCode connected, and a shelf of already-reviewed skills on SkillPatch that nobody was required to touch. No one had any idea what they’d end up with by Sunday.
245 teams shipped something in that window. Four of them stood out, and the interesting part isn’t just what they built. It’s how much of it they didn’t have to build alone.
Canopy fixes a problem almost every team building with AI coding agents runs into eventually: the instruction files that steer them, CLAUDE.md, AGENTS.md, .cursorrules, quietly start contradicting each other and rotting as a codebase grows, and nothing catches it until the agent gets it wrong. Canopy catches it first, automatically, running ten deterministic checks for conflicts, duplication, and drift before any of it reaches an agent, even leaning on a frontend-design-review skill from SkillPatch to shape its own dashboard.
In India’s North Eastern Region, a landslide or a flood can cut a district off from medicine, food, and supplies, with no way for anyone to know which roads are even passable. Ranjit Bhardwaj and Nikita Mishra used LatentCode to build a live command center that maps the disruptions, flags safe routes, and tracks relief vehicles in real time, pulling in fastapi-router-py, agents-md, and frontend-ui-dark-ts skills from SkillPatch along the way. For a dispatcher deciding whether a relief truck can actually reach a cut-off village, that’s visibility that doesn’t currently exist anywhere else.
For a healthcare worker at a rural Primary Health Centre, the gap between a patient standing in front of them and the right clinical guidance can be the difference between a good outcome and a bad one. ClinAssist India closes that gap: describe a case in Hinglish and it generates a structured, ICMR-grounded treatment card in Hindi with clear next steps, using a prompt-optimizer skill from SkillPatch to keep its prompts sharp. In a district with one clinic and no specialist for fifty kilometers, that card might be the only second opinion a worker gets.
Every hackathon submission needs a demo video, and making one usually eats more time than the actual build. ShipShow, which Ramkumar R put together with LatentCode, skips all of that: give it a URL and a one-line prompt, and it plans, records, narrates, and edits its own demo video, no screen recorder, no mic, no editing software. He then turned the whole thing into a skill, demo-video-pipeline, and published it to SkillPatch, where it’s live now, Verified with a 91/100 security score, ready for anyone to use themselves.
Stay tuned for the next hackathon. Join the Discord to be the first to hear about it, and our latest releases.