Kill the Default Route — listen to this quick discussion on why your network's most permissive policy is hiding in the routing table https://t.co/d5ZJydTXaw
That's somebody else's problem. That's not Skillseal’s goal. If someone's attesting to a skill, it means they (may have done) did the dirty work of checking the skill and its safety. Skillseal only says that -> person said this -> skill is good or bad. They could be wrong. You trust them or you don’t.
Proposing a trust layer for AI agent skills: SkillSeal v0.2.0.
Multi-key signing (GPG + SSH), content-addressed integrity, third-party attestations, and a local trust store that lets you define who your agents trust.
Compiled binaries on GitHub Releases.
https://t.co/zUCZ7VcE3m
@MircoBalda33@AtlasForgeAI It's definitely a seed mechanism. Anthropic, open AI, Google: anyone can publish a bundle. If you trust them, use it. Or if you're a research team that wants credit for the work you do. Publish a bundle.
v0.2.5 just shipped. Key additions:
- Destatements — trusted reviewers can publish negative attestations that block dangerous skills, even if the author is trusted
- Trust bundles — orgs can publish signed lists of trusted authors/reviewers that users subscribe to
- GPG revocation detection — compromised keys are caught before verification
130 tests. Whitepaper updated.
https://t.co/lmxbrQBXV0
@AtlasForgeAI It's not built yet, but there will be a file for revocation that you can maintain. It will take priority over the trusted store. I'm not for certain yet; haven't built it. But it's on the road map.
What if your API's rate limit was a paywall instead of a dead end?
Built it. Shipped it. $0.17 on Base gets you through.
(wow what a long weekend can accomplish)
https://t.co/T4jm9X57eY