zsun79/conferencewatch

ConferenceWatch is an Agent Skill, a set of instructions meant to be loaded into coding assistants like Claude Code or Codex, that tracks deadlines and details for upcoming AI and machine learning conferences. Instead of being a standalone app, it is a bundle of Markdown and JSON files that an AI agent reads and follows as its own workflow, so it works with any agent that can search the web and write files. When it runs, the skill produces two outputs. The first is a JSON file listing each tracked conference by its short name, including every submission deadline stage, the location, any special sessions or calls for papers, official links, and a five year trend of acceptance rates. The second is a plain language Markdown report built from that data, with a summary, a table of the nearest deadlines, and a per-conference breakdown, plus a section that clearly separates confirmed facts from estimates. The agent also answers directly in chat with whatever deadlines are most time sensitive. The workflow starts by anchoring on today's date so it does not confuse past and future dates. It then narrows down which conferences to track, either from what the user names directly or by asking a few follow up questions about research area, venue tier, or acceptance difficulty, defaulting to a curated list of top AI venues if the user has no preference. For each conference, it searches official call for papers and important dates pages first, and only estimates a future deadline from historical patterns when no official date has been announced yet, always labeling that estimate as approximate. The project states clearly that it should never invent a date that is not backed by a source. To install it, users clone the repository into their agent's skills folder, for example Codex's or Claude Code's local skills directory, or symlink an existing checkout into that folder. No API keys, build steps, or special dependencies are required beyond an agent that can search the web and save files. Generated reports are marked private by default since they may contain personal notes about planned submissions.