Repository navigation
Discover model capabilities and allow manual reasoning profiles - #402
Merged
Merged
Conversation
fylorn
marked this pull request as ready for review
November 12, 2026 13:01
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this changes
Model capabilities are discovered automatically and can be overridden per provider and exact model ID.
model_specs.<id>.reasoning_effortsupplies an atomic effort profile, including explicit non-adjustable models and an optional default. Removing it restores upstream discovery.Why
Relays that omit capability metadata leave Codex without effort choices, even when a model supports reasoning. Background inventory also discarded capabilities that the upstream did report. Related to ThinkWatchProject/ThinkWatch-Lite#438.
How it was verified
Only declared capabilities are retained from ordinary upstream metadata; instructions and tools retain the existing native-profile boundary. Credential-scoped native caching, request authorization, and model permission checks remain covered by the existing regressions.
Notes for review
Control API version 60 adds reported, manual, effective, and protocol-filtered effort fields. Effort edits explicitly distinguish keep, set, and automatic; omitted fields preserve existing profiles. Nonempty manual profiles imply reasoning support without manufacturing a boolean override. Unknown candidates never gain guessed effort choices.
Prepares Core 0.95.0, including release notes, the message-code manifest, and control API contract checks. Companion Lite PR ThinkWatchProject/ThinkWatch-Lite#439 will pin the published v0.95.0 tag before production packaging.