awkweb

Mount your API onto your CLI. Now supported in incur. → Compatible with any framework that supports the Web Fetch API (most of them – Hono, Elysia, Bun, Deno, and more). → OpenAPI support for extra context & validation → Supports streaming out-of-the-box

Serve your CLI as an API. Now supported in incur. → Exposes a Web Fetch API handler on your CLI to pass through to a compatible framework → Automatically exposes command skills under /.well-known/:command/skill.md routes → Automatically exposes an /mcp endpoint for your CLI

Mount your API onto your CLI. Now supported in incur. → Compatible with any framework that supports the Web Fetch API (most of them – Hono, Elysia, Bun, Deno, and more). → OpenAPI support for extra context & validation → Supports streaming out-of-the-box

Serve your CLI as an API. Now supported in incur. → Exposes a Web Fetch API handler on your CLI to pass through to a compatible framework → Automatically exposes command skills under /.well-known/:command/skill.md routes → Automatically exposes an /mcp endpoint for your CLI

skills are really tricky because everyone tries to have their AI help them write the skill -- but if the AI already knows it, it shouldn't go in a skill also i think that the AI might have a private key to use as a hot wallet as models get better my advice is to: 1) have a skill file clear published for humans and bots to find right on your homepage /skill.md 2) have the skill file be only the stuff LLMs don't already know and if you have a lot of content have it be an index of everything they can go learn (keep the context lean) 3) assume agents will have more and more agency and intelligence going forward, including their own wallet 4) over and over you should be looping with a blank agent: give it the skill, tell it to use it/build with it, see where it fails, improve the skill

still waiting for modern chat app a la linear, raycast, craft, and others. how are we still all using slack and discord 😥

Without call-to-actions, agents guess what to do next, or worse, ask the user. incur CLIs tell the agent exactly which commands to run after each step, with the `cta` property.

the best in nyc so far 10/10 at gage & tollner

wrong, this has been a pain point for 10+ years. it’s just cope the default ux requires reinventing so much shit to be at least slightly usable. if ethereum is serious about self-sovereignty, it should be trivial to get account-centric data, and fast. at least some projects like otterscan and trueblocks can see that.