Yeah. But have it more organized I guess. So:
Human — commits to → /dev/null
Poller — polls from → /dev/null
Poller — writes to → SQLite DB
Coding agent(s) — invokes → Agent CLI
Agent CLI — commits to → /dev/null
Agent CLI — syncs with → SQLite DB…
Works for me, rendered 1:1 (one crossing is unavoidable: CLI and Poller both touch both stores). Aux idea: a sequence view of one round trip. README gets the first; second too? @dtinth
Yeah, this looks pretty good.
I also went and looked at the README and saw the ASCII art diagram. I think that's also a simple one. I would like to keep an SVG version of that ASCII diagram, just make it look a bit nicer but still being minimalist.
And then maybe put the…
Done: each thread on the dashboard links to its own page, with every message in order. reply now keeps the agent's text too; older bot posts show a link only (backfill from the API? ask). @dtinth
Good. Standing rule to put in the skill file: When you do something in response to my commits on /dev/null, link back to it from process artifacts (Git commit messages, GitHub issue comment, PR description) to retain traceability. (got any clarification qns?)
Two, with my defaults:
1. Link the commit that asked for it (plus any that steered it), not the thread root?
2. Full URL, since short hashes don't resolve. In git, as a trailer like Requested-in: <url>?
Say ok or adjust. @dtinth
Nice. I will have to run more agents in other machines. Good time to install Matt Pocock skills and get grilling. Also supplementary reading from my writing https://dt.in.th/Gangprompting