Skip to content

Gauntlet loop TUI: live candidate vs named bar - #3

Draft
ThyFriendlyFox wants to merge 3 commits into
mainfrom
cursor/gauntlet-loop-tui-a8ef
Draft

Gauntlet loop TUI: live candidate vs named bar#3
ThyFriendlyFox wants to merge 3 commits into
mainfrom
cursor/gauntlet-loop-tui-a8ef

Conversation

@ThyFriendlyFox

@ThyFriendlyFox ThyFriendlyFox commented Aug 18, 2026

Copy link
Copy Markdown
Contributor

Summary

Adds an ORC Gauntlet loop operator TUI, based on Matt Shumer’s method (via robonuggets/gauntlet-loop):

  • Named, fetchable, comparable quality bar (vague bars are rejected)
  • Goal split into independently judged pieces
  • Builder and a separate blind critic (critic never sees builder notes)
  • Verdicts are picks, not scores; bad evidence throws the round out
  • Live board: pieces + candidate viewport vs bar + verdict ticker
  • One shared candidate page so the first viewport fills in as pieces land
  • Operator is the brake (space pause, s stop)

Demo runs against an on-disk Northstar Run campaign fixture so the loop is real without API keys.

./scripts/cloud-agent-install.sh
.venv/bin/python -m gauntlet demo

Cloud Agents: ~/.orc-venv/bin/python -m gauntlet demo (venv lives outside the git checkout).

Test plan

  • Install twice (idempotent)
  • .venv/bin/pytest -q — 13 passed
  • Headless demo reaches won in 12 rounds
  • XFCE desktop: TUI shows named bar, pieces, candidate vs BAR, pause/stop; candidate preview shows “Go farther.” once hero lands
  • Draft environment build with durable ~/.orc-venv + Textual 2.1.2 on a skipped-install boot
Open in Web Open in Cursor 

cursoragent and others added 3 commits August 18, 2026 00:45
Ship a Textual operator board for Matt Shumer's Gauntlet loop:
independently judged pieces, builder vs blind critic, picks not
scores, and a live candidate-vs-bar viewport. Demo runs against an
on-disk Northstar campaign fixture so the loop is testable without
API keys.

Co-authored-by: ThyFriendlyFox <thyfriendlyfox@gmail.com>
The TUI now sketches the accumulating landing page instead of
resetting to a placeholder whenever a new piece starts. Invalid
evidence still rolls the page back to the pre-build snapshot.

Co-authored-by: ThyFriendlyFox <thyfriendlyfox@gmail.com>
Cloud Agent boots can wipe .venv during checkout. Install also
creates ~/.orc-venv so Textual survives a fresh pod.

Co-authored-by: ThyFriendlyFox <thyfriendlyfox@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants