interest-radar

Interest Radar is a small skill module that judges whether a piece of outside content is relevant to you. The README states what it does not do as clearly as what it does: it does not collect data, it does not push notifications, and it does not store anything. Its single job is to look at a link or a passage you share and tell you how related it is to your interests, why it reached that conclusion, and whether you should bother reading further.

The project is described as platform-agnostic. It is not tied to one specific agent system, so the same files can run inside Hermes, Claude Code, Codex, Cursor, or ChatGPT, according to the README. It also has no code dependency. The whole behavior is defined in a SKILL.md file rather than a Python script or external runtime, which the author calls a zero code dependency design. At runtime the skill is meant to detect whatever memory or search tools happen to be available and use those, without assuming any particular tool exists.

The README lists three small capabilities the author calls MVP features. The first is forwarding a link and asking about it, which triggers a relevance check and returns a judgment. The second is a batch_judge interface that other recommendation skills can call to filter a list down to only the high-relevance items. The third is a feedback loop: when the user replies with confirm or dismiss, those responses are supposed to adjust later scoring so the judgments improve with use.

Installation is described as copying the SKILL.md file and a references folder into the agent's skills directory. The README also points to a set of reference documents in that folder covering the batch interface, the feedback loop, the output format, and a snapshot cache scheme. The README itself is short and does not go into deeper implementation detail beyond that.