Skip to content

feat: support configurable model names and concurrent Science runtimes - #66

Open
RerrentLinden wants to merge 3 commits into
SuperJJ007:mainfrom
RerrentLinden:agent/configurable-model-names-and-science-runtime
Open

feat: support configurable model names and concurrent Science runtimes#66
RerrentLinden wants to merge 3 commits into
SuperJJ007:mainfrom
RerrentLinden:agent/configurable-model-names-and-science-runtime

Conversation

@RerrentLinden

Copy link
Copy Markdown

Summary

  • add optional display-name inputs for the default, quality, fast, and Fable model routes while keeping exact upstream model IDs unchanged
  • restore Science Skill binding/readback through the current browser nonce flow, with the legacy nonce endpoint retained as an explicit compatibility path
  • allow the official and CSSwitch-isolated Science instances to remain available concurrently by treating the mode switch as an operation view and isolating browser cookie origins

Root cause

Provider model IDs must remain exact for routing, but slug-like IDs can be filtered when reused as Science-facing display names. The configuration UI previously could not edit display_name independently.

Current Science consumes its control-plane nonce from GET /?nonce=..., while CSSwitch only used the older POST /api/auth/nonce flow. In addition, browser cookies are scoped by host rather than port, so opening both Science instances on localhost could replace the other session. The old mode switch also coupled UI selection with teardown of the managed third-party runtime.

Branch history

This draft supersedes #64. The three functional commits were replayed onto the latest upstream main, so the PR is 0 commits behind and 3 commits ahead. The two fork-only upstream synchronization merge commits are intentionally excluded.

Validation

  • git diff --check upstream/main...HEAD
  • node --test test/model_catalog_state.test.mjs test/runtime_status_state.test.mjs: 22 passed
  • python3 -m unittest test.test_skill_runtime_boundary: 11 passed
  • Rust tests were not rerun in the current shell because cargo is unavailable; the PR remains draft rather than claiming a complete release gate

@RerrentLinden
RerrentLinden marked this pull request as ready for review July 20, 2026 17:57
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