Skip to content

Workshop fixes, Copilot catalog refresh, preset update - #310

Merged
lukas-grigis merged 5 commits into
mainfrom
feature/misc
Aug 18, 2026
Merged

Workshop fixes, Copilot catalog refresh, preset update#310
lukas-grigis merged 5 commits into
mainfrom
feature/misc

Conversation

@lukas-grigis

Copy link
Copy Markdown
Owner

Three batches from tonight's pre-workshop session, each verified green (typecheck/lint/tests) before commit:

Workshop fixes (from the release-readiness audit)

  • ralphctl demo now detects installed provider CLIs and seeds a single-provider preset instead of the cross-provider defaults — the demo's implement sprint no longer requires both claude and codex. --script keeps its claude-code pin.
  • The v evaluation chord now works in the wide (>=140 col) Execute layout; regression test covers both width regimes.

Copilot model catalog — reconciled to the 2026-08-18 supported-models doc

  • Added gemini-3.7-flash, mai-code-1.1-flash, kimi-k3, grok-4.5, grok-4.6 (first xAI group).
  • Preview graduations: gemini-3.1-pro, raptor-mini. Delisted: gemini-2.5-pro, gemini-3-flash.
  • RETIRED_MODEL_REMAPS migrates persisted settings pinned to the four retired slugs.

Preset refresh (multi-agent evaluation, must+should items)

  • gpt-5.4-mini (retires 2026-08-31) replaced by gpt-5.6-luna across all codex rows incl. every row of codex-fast; Copilot cheap tier moves off gpt-5-mini to gpt-5.6-luna.
  • mixed / mixed-frontier implement evaluator moves to openai-codex/gpt-5.6-sol — an independent cross-provider gate matching the shipped DEFAULT_SETTINGS split.
  • mixed refine sol → terra; claude-only readiness haiku → sonnet-5.
  • The apply-preset missing-CLI warning now also covers the implement evaluator's provider.

Full suite: 617 files / 6052 tests green.

https://claude.ai/code/session_01PHikH4LdejozJdjzTzhPes

… layout

- ralphctl demo now probes PATH and applies a single-provider preset (claude-only fallback) instead of the cross-provider DEFAULT_SETTINGS; --script keeps its claude-code pin
- thread onOpenEvaluation through ImplementLayout wide path so v opens the evaluation overlay at >=140 cols; regression test covers both width regimes

Claude-Session: https://claude.ai/code/session_01PHikH4LdejozJdjzTzhPes
- add gemini-3.7-flash, mai-code-1.1-flash, kimi-k3, grok-4.5/4.6 (first xAI group)
- graduate gemini-3.1-pro and raptor-mini out of -preview slugs
- delist gemini-2.5-pro and gemini-3-flash; RETIRED_MODEL_REMAPS migrates pinned settings
- slugs convention-derived, not CLI-validated (copilot-cli #700)

Claude-Session: https://claude.ai/code/session_01PHikH4LdejozJdjzTzhPes
…gates

- gpt-5.4-mini (retires 2026-08-31) -> gpt-5.6-luna on all codex rows; copilot cheap tier gpt-5-mini -> gpt-5.6-luna
- mixed/mixed-frontier implement.evaluator -> openai-codex/gpt-5.6-sol, matching DEFAULT_SETTINGS' independent-gate split
- mixed refine sol -> terra; claude-only readiness haiku -> sonnet-5
- apply-preset missing-CLI warning now covers the implement evaluator's provider

Claude-Session: https://claude.ai/code/session_01PHikH4LdejozJdjzTzhPes
@lukas-grigis
lukas-grigis merged commit 0d5b54e into main Aug 18, 2026
3 checks passed
@lukas-grigis
lukas-grigis deleted the feature/misc branch August 18, 2026 21:14
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.

1 participant