What is skillsense?

SkillSense is a small tool that runs alongside AI coding agents like Claude Code and OpenAI Codex. It does not replace them, and it does not add new abilities to them. Instead, right before each of your prompts is processed, it quietly checks whether you already have a relevant skill, command, agent, MCP server, project rule, or instruction file installed locally, and if so, it reminds the agent that this capability exists and where to find it. The problem it targets is not that these agents cannot use installed skills. Both Claude Code and Codex already support automatically picking up a skill's name and short description when deciding what to use, without you having to type an explicit command. The issue is recall: as a person installs more skills, commands, and rules over time, the model has more and more metadata to silently scan on every turn, and it can simply miss the right one, giving a weaker answer or making you redo work by hand that a skill already covered. SkillSense answers one question automatically before each prompt: are there any installed capabilities that look materially relevant right now. It works entirely on your machine. When a prompt is submitted, SkillSense scores your locally indexed capabilities against it using local, deterministic matching, keyword and tag overlap plus a text-similarity method called BM25, with no large language model call, no embeddings, and no API key involved. If nothing scores highly enough, it stays silent. If exactly one capability matches well, it adds a single short line naming it and where it lives. If more than one matches, it lists them and leaves the choice to you rather than guessing. A SkillSense failure is designed to never block your prompt from going through. The README also explains that Claude Code's own built-in skill recall has limits: descriptions loaded into context are capped at roughly one percent of the context window, and past that budget, least-used skills get truncated or dropped, which is the situation SkillSense's reminder is meant to help with. SkillSense needs Node.js 22.5 or newer and uses Node's built-in local database module for storage, so no native compilation or extra database is required. It integrates with Claude Code and Codex through plugin hooks, and is installed by cloning the repository, running npm install and npm run build, then linking the command line tool onto your path.