Show HN: Skillzero – save tokens by omitting skills from agent context

Show HN: Skillzero – save tokens by omitting skills from agent context(github.com)

2 points by kurtextrem 25 days ago | 2 comments

Hi HN, this is a first for me. I use agents for multiple things, from daily life, fitness, video editing, to (performance) engineering. For some of those I have skills globally installed, because I may need them across different projects. But this means in any unrelated task, all skill names + descriptions will still make it into the agent context [1]. This is not the end of the world, but also not great as it could lead to your agent no longer invoking skills reliably (see Addy Osmani's https://addyo.substack.com/p/audit-your-agent-files).

With skillzero, you can manage the skills via CLI. It splits them in 3 categories: - hidden skills (entirely removed from context) → they now behave like /-commands - collection skills (bundled into one skill, so the agent sees only one name + description) → agent can still invoke these when it thinks it should use them - regular skills (untouched by skillzero)

You can install it via `npx skillzero` (or `pnpx` or whatever you prefer).

[1]: Actually Claude Code & Codex have an optimization, if the name+desc's exceed 1%/2% of the context window, both will omit some skills entirely.

Happy to hear any feedback!

keks0r 24 days ago |

The bundling is pretty interesting, does this also work on repo level? e.g. I have added some sales skills into our company brain, but for sessions that others would do, if they are not doing sales related tasks, they would not require any of it. But I still want to be able to directly invoke one of my sales related skills directly.

kurtextrem 24 days ago | |

yes :) you can pick the path it should run on or invoke in a path that has .claude or .agents as a subfolder, it will then auto-detect and allow you to write collections (bundles).