Repository navigation
fix: allow model fallback on Claude session limits - #11
LaurensBosscher wants to merge 6 commits into
Conversation
Deep @oh-my-pi/* imports fail under OMP 18.2.5 extension loading because host onLoad only rewrites legacy scopes onto in-process shims. Switch to @earendil-works root imports (proven on this host) and register claude-opus-5 so existing modelRoles/fallbackChains keep resolving.
…, Fable 5.1, Mythos) + max effort Derive the picker from the host's anthropic catalog instead of a hardcoded id list, keeping the 1M/200K variant entries. Measured [1m] entitlements stay in tables; unmeasured 1M-capable models default to 200K with a -1m alternate. Map thinking levels onto each model's catalog effort ladder (adds max). Fix unsuffixed entries throwing when their preferred window has no runtime.
OMP 18 ends some turns with developer messages (subagent reminders, synthetic continues). The bridge only read user messages, so it replaced them with "[continue]" and dropped them from rebuilt history. Treat them as user turns, matching OMP's anthropic provider; empty ones are skipped.
…session per OMP session
OMP 18 clears each extension source's providers when it creates a session on
the shared ModelRegistry and replays only that instance's registrations. The
re-registration guard made parallel subagents lose the provider entirely
("No API key found for claude-bridge"). Always register; streamSimple still
routes to the first active instance.
Subagents share this module instance, so the single global sharedSession let
parallel sessions resume each other's Claude Code session (cross-talk) and let
a subagent's session_start/shutdown reset the parent's. Key session state by
OMP session id.
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars.
 |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5ed40d8e09
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| const allResults = activeQueryContexts.size > 0 ? extractAllToolResults(context) : []; | ||
| const resultCtx = allResults.length > 0 ? contextForToolResults(allResults) : undefined; | ||
| const isReentrantUserQuery = activeQuery && lastMsgRole === "user" && allResults.length === 0; | ||
| const isReentrantUserQuery = activeQuery && lastMsgRole != null && isPromptRole(lastMsgRole) && allResults.length === 0; |
There was a problem hiding this comment.
When OMP appends a developer reminder or synthetic continue in the same provider call as a matching tool result, allResults.length > 0 sends the call through the result-delivery path instead of the newly broadened prompt path here. That path's deferral guard at line 1177 still accepts only lastMsgRole === "user", even though extractUserPrompt now accepts developer, so the reminder is neither sent immediately nor added to deferredUserMessages and Claude silently skips it; apply isPromptRole to that guard as well.
Useful? React with 👍 / 👎.
Changes
Stacked after #10: this branch includes earlier commits until their PRs merge.
Verification
openai-codex/gpt-6.1-terra, and completed.@earendil-works/*declarations.