Skip to content

v1.8.9

Choose a tag to compare

View all tags
@Alan-TheGentleman Alan-TheGentleman released this 25 Mar 23:49
· 783 commits to main since this release
v1.8.9
f74e039

What's new in v1.8.9

🎯 Model assignment fixed — your choices are respected

The installer no longer overwrites models you already configured in opencode.json. We introduced a 3-way priority system:

  1. TUI model picker choice → always wins
  2. Agent already exists in your config → preserved as-is, never touched
  3. New agent + root model set → inherits your root model (breaks OpenCode runtime inheritance)

This means reinstalling gentle-ai is now safe — your model choices survive.

🎼 sdd-orchestrator model is now selectable

The orchestrator appears as the first row in the multi-mode model picker, separate from sub-agents and excluded from "Set all phases". Previously it silently inherited whatever model happened to be running — now you choose it explicitly.

The model you pick for sdd-orchestrator is automatically mirrored to the gentleman agent too, since both are primary conductors in OpenCode.

⚡ ~37% fewer tokens per session

All embedded LLM prompt assets were compacted for token efficiency — same behavior, less overhead on every message. Tables with self-evident columns became bullet lists, verbose sentences became terse, redundant emphasis was consolidated.

Validated with a dual adversarial review (two independent agents, blind to each other) before shipping. 6 critical enforcement phrases that were accidentally stripped were caught and restored before release.

⚖️ New skill: judgment-day

A parallel adversarial review protocol installed with every preset. Two independent judge agents review the same target simultaneously, their findings are compared, confirmed issues are fixed, and re-judged until both pass clean.

Triggers: "judgment day", "juzgar", "dual review", "que lo juzguen"

🔄 Self-update no longer loops

brew update now runs automatically before brew upgrade during self-update. Previously, a stale Homebrew formula cache could cause the installer to report "update available" in a loop even after upgrading.

What changed since v1.8.8

  • internal/assets/ — all LLM prompts compacted + enforcement language validated
  • internal/assets/skills/judgment-day/SKILL.md — new skill, installed in all presets
  • internal/tui/screens/model_picker.go — orchestrator row added to picker
  • internal/components/sdd/inject.go — 3-way model injection + orchestrator→gentleman mirror
  • internal/update/upgrade/strategy.go — brew update before brew upgrade
Assets 9
Loading