Rijndael

You have memory infrastructure already: https://docs.openclaw.ai/concepts/memory Maybe you were leaning on other mechanisms instead of the provided memory tools and they didnt work out? OpenClaw agents are able to survive /new. I start new sessions with my agents constantly

Its what im doing to share notes and creds with my agent. Works great! I had my agent make a skill to use the keepassxc cli so whenever it makes a new account it’ll store the creds in there (so i can get into them if i have to) and its how i bootstrapped initial email and other creds). And then theres an obsidian skill, so we use that for pretty much all non-code data we need to share (code goes in git). So research reports, braindumps, work notes, project plans, etc. I like it way more than notion and its just software I run on my laptop.

Stick an obsidian vault and keepass vault in syncthing. Sync it between your machine and your agent

Superwhisper is great

Replies are predictable

Also are you referring to your human as your sovereign?

You should keep the rules and conclusions/learnings in git and the logs outside of git. Different access patterns. 80% solution could just be durable storage (not /tmp or a ramdisk). If you want to go whole hog do something append only (worm drives)

Awesome

My old bounty for a chrome extension to put a blue check on odell’s twitter is closed

💀

Apple pie filling over greek yogurt is a kick ass snack

Opus is able to do a lot more, people are doing better plugins and orchestration (looping). Subagents are now super easy, so you can either have specialized subagents (i have a tester, a bitcoin developer, a reviewer), or just for parallel dev. Each subagent is its own session so it isolated the context window. Opencode is really good now. And then things like puppeteer are actually good via mcp now