Hacker Newsnew | past | comments | ask | show | jobs | submit | hatappo's commentslogin

I’ve been building `agent-skill-porter`, a CLI for managing AI agent Skills across ecosystems:

https://github.com/skill-mill/agent-skill-porter

One thing that bothered me is that Skills usually sit outside normal package-manager protections.

Package ecosystems are increasingly using ideas like pnpm’s `minimumReleaseAge` and uv’s `--exclude-newer` to avoid pulling in artifacts that are too new. But Skills often aren’t package-managed at all, even though they may include helper scripts, downloaded artifacts, and even binaries.

Given recent supply-chain incidents like the axios compromise, and research like Snyk’s ToxicSkills report, that felt like a real gap.

So I added a cooldown mechanism based on the Git tree hash of the skill directory.

The CLI now resolves a Skill from Git history, treats the skill directory tree as the release unit, and only installs/updates to a tree that is older than a configured minimum age. If the latest one is too new, it falls back to the newest eligible older tree.

This is implemented as `--min-age` for add/download/update. ex. `sk add https://github.com/remotion-dev/skills --min-age 30`

I’d be especially interested in feedback on whether “directory tree hash” seems like the right trust/version boundary for non-package-managed AI artifacts.


I think TanStack Intent is quite close to that direction.

Packaging skills with libraries/CLIs and letting agents discover them from installed packages makes a lot of sense. I see Harbor as addressing a different layer on top of that: organizational collection, cataloging, provenance, governance, and safety.


  Yes, I agree that MCP-based prompt/skill delivery would be a very interesting direction.

  If tooling vendors broadly supported MCP prompts, an MCP server could become a dynamic distribution layer for team-managed skills, which would remove a lot of sync-oriented workflow.

  My current assumption is that we still need something Git-native today because:
  - skills are mostly authored and reviewed in Git
  - teams need provenance and governance around them
  - tool support for MCP prompt delivery is still incomplete

  So I see Harbor more as a practical system for the current ecosystem, not necessarily the final shape.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: