Repository navigation
[pull] main from can1357:main - #108
Merged
Merged
Conversation
…ize the GPT-5.6 ladder Two follow-ups to the #12740 repoint, both grounded in the vendor's published contract (docs.singularityapi.dev). GPT-5.6 never sent `reasoning_effort`. `GET /v1/models` publishes no reasoning metadata — the documented `capabilities` shape is endpoint, context/max/default output tokens and pricing — so `gpt-5.6-sol`, `gpt-5.6-terra` and `gpt-5.6-luna` were discovered as non-reasoning while the KDL asserted a ladder that never materialized. Every reasoning branch in `resolveOpenAICompatPolicy` is gated on `Boolean(model.reasoning)`, so omp omitted `reasoning_effort` entirely, and the gateway answers 400 from upstream when chat completions carry `tools` without one. `thinking-upgrade-neutral #true` — the axis the DeepSeek rules already use — materializes the five-tier ladder, so the session's thinking level reaches the wire on every request and "off" still sends the literal `"none"` the contract asks for. Image rows had no transport. `kind "image"` was asserted from the KDL while discovery left those rows on `api: "openai-completions"`, and nothing maps kind to transport for runtime-discovered rows (`KIND_API_KINDS` / `kind-apis` is consumed only by the models.dev catalog layer). `generateImage` switches on `model.api` and throws for `openai-completions`, the image role accepts rows by kind alone, so `generate_image` listed them as skipped and could end in an exhausted image chain. The mapper now reads each row's own `capabilities` list and routes a row whose only surface is `/v1/images/generations` to `openai-images` with `kind: "image"` and no tool support — the base URL already ends in `/v1`, which is the documented `POST /v1/images/generations` path. Tests: a bare GPT-5.6 row must materialize the ladder, and the image assertion covers the transport (`api`), not just `kind`.
fix(catalog): singularityapi image transport and GPT-5.6 reasoning ladder
… docs searches - Move the omp:// scope grammar (docs-scope/root detection, host+path, and the traversal-safe canonical doc path) into internal-urls/omp-scope.ts. The protocol handler, grep's virtual-resource expansion, and find's temp corpus now read scope semantics from one place instead of three near-copies; the find copy had also dropped the handler's absolute-path guard. - `omp find` resolves settings and extensions against the caller's cwd when the scope is omp://. Passing the temp corpus as the extension-discovery base silently dropped project extensions and their providers, so judge resolution could change or fail for exactly the docs scope. - Materialization honors the caller's abort signal and reports a "materializing omp:// docs" progress line; the CLI now passes the same ResolveContext the tool does. - Apply oxfmt to the touched files and to the two tui files whose long import/call lines failed `bun check`, and re-indent the find output block the original try/finally left visually outside the try. CHANGELOG: record the omp:// docs scope under Unreleased.
- one doc: canonical scope path, the single materialized file, and cascade hits remapped to the omp:// URL - root scope: the corpus matches the completion set, a doc's bytes match a direct router resolve, and cleanup removes the temp dir
…inally, and clarify the selector rejection
- printReport received displayRoot unconditionally, so its cwd-relative
fallback was dead and plain directory scopes printed absolute roots;
pass ompScope?.scopePath.
- process.exit(1) skipped the finally blocks, orphaning the materialized
omp corpus and skipping authStorage.close(); set process.exitCode
instead, which the CLI runner leaves in place on success.
- restore the dropped "resolving judge" progress line.
- selector-on-path rejection now states the reason ("find searches whole
files; line-range selectors are not supported"), documented in
docs/tools/find.md and the tool prompt; assertion updated.
- FindHit.rel doc comment covers omp:// doc URLs; the root-scope test
cleans up in finally.
feat(find): search omp:// harness docs via materialized corpus
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.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )