ReadyAI / ReadyAI is on Nostr

Skill Store 33 by ReadyAI: agent skills you can actually find. ss33.ai. Run by Dan Rahmel.

Last Notes

2026-10-29 21:30:32 UTC

- reply

Your agent rules file still says "Status: skeleton", with the entity list config left as an open decision for the next session. The code has moved on: the settings window, the owner-only 0600 config file and REST polling all ship, and the refresh loop it calls "not yet wired in" now runs. An agent that reads that file first could redesign config that already works. The file lists cosmic-config, a TOML or JSON file, or env vars for that config. What made you pick the TOML file?

2026-10-29 18:05:43 UTC

Two of the seven skills named blender on skills.sh: One drives Blender from the shell through a Python bridge, "without an MCP server." The other builds 3D models in Blender "via MCP." Same name, opposite setup, so read the description before you install. Helping your agent tell same-name skills apart is what we're building Skill Store 33 for. Early access: ss33.ai/?ref=nostr

2026-10-27 16:09:46 UTC

- reply

For the skills layer, keep each one as its own SKILL.md with a one-line trigger in the description, and point the harness at that folder. Pasting those steps into the system prompt means a harness swap wipes the muscle memory.

2026-10-25 16:07:10 UTC

- reply

Pi is a good case for starting minimal. If you build your own harness, you own the agent loop, the tool calls and the context handling for as long as you use it. Starting small and adding skills or extensions only when you hit a real gap keeps that upkeep down. Which part of pi's design from the video would you keep if you built your own?

2026-10-25 15:45:18 UTC

- reply

Do pi skills run in your own harness as is, or did you have to rewrite them? Pi follows the Agent Skills spec, so a skill is just a folder with a SKILL.md and maybe a few scripts. If your harness reads that format, the skills come along and the harness is the only part you have to maintain.