You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
fix(agents): propagate runtime context windows, maxTokens, and image modality across harnesses - #443
Apply configured context limits before merge_runtime_context_windows in fetch_codex_runtime_models.
Extend CodexRuntimeModel and merge_runtime_context_windows to propagate input_modalities and max_tokens (harness_metadata.maxTokens) alongside context_window.
Enable runtime metadata merging for Hermes, OpenClaw, OpenCode, ClaudeCode, and ClaudeDesktop, and write limit.context/limit.output in OpenCode, max_output_size/image_in in Kimi Code, and maxOutputTokens in VSCode Copilot.
Add gemini-4-argon-{low,medium,high} (1,048,576) to claude_models/model-catalog.json and support macOS bundle/CLI path discovery for WorkBuddy, OpenClaw, Kimi Code, Grok Build, and OpenCode.
Handle macOS app bundle paths in version detection
src-tauri/src/agents/deepseek_desktop.rs:80
Manual macOS .app overrides are accepted by manual_desktop_path and passed unchanged here by inspect_agent_config_with_executables. For a bundle path, parent()?.parent()? resolves outside the bundle, so version detection misses an existing Contents/Resources/app.asar. Handle bundle paths separately from discovered Contents/MacOS executables.
Addressed the "Handle macOS app bundle paths in version detection" finding (src-tauri/src/agents/deepseek_desktop.rs:80) in 598f5c5: when a manual desktop override points at a .app bundle directory, the version is now read from <bundle>/Contents/Resources/app.asar; executables under Contents/MacOS keep the existing lookup. Added deepseek_desktop_macos_version_reads_bundle_and_executable_paths covering both path forms and a bundle without app.asar. cargo test: 858 passed, 0 failed.
The agent model picker read /v1/models without `client_version`, so the kernel
answered with the bare `{id,object,owned_by}` list. Nothing carried
`input_modalities` or `context_window`, so WorkBuddy's config writer had no value
for `supportsImages` - and the client gates image input on
`supportsImages === true`, meaning every model looked text-only.
- also fetch the catalog variant (`client_version=cpa`) and graft its capability
fields onto the bare listing, so config.yaml aliases and custom model IDs stay
in the picker
- parse `slug`, the key the catalog uses for a model name
- default WorkBuddy's `supportsImages` to true when the catalog has no entry, so
hand-typed custom IDs are not silently text-only
Upstream reports the same gap as router-for-me#442; its fix
(router-for-me#443) is still open.
Rebased onto current dev. I dropped two parts that dev now covers: the client_version=cpa request (b80ee99 pins the Codex version) and the macOS DeepSeek Harness bundle fix (#485). The rest is unchanged, and I updated the description to match.
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
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.
Summary
merge_runtime_context_windowsinfetch_codex_runtime_models.CodexRuntimeModelandmerge_runtime_context_windowsto propagateinput_modalitiesandmax_tokens(harness_metadata.maxTokens) alongsidecontext_window.Hermes,OpenClaw,OpenCode,ClaudeCode, andClaudeDesktop, and writelimit.context/limit.outputin OpenCode,max_output_size/image_inin Kimi Code, andmaxOutputTokensin VSCode Copilot.gemini-4-argon-{low,medium,high}(1,048,576) toclaude_models/model-catalog.jsonand support macOS bundle/CLI path discovery for WorkBuddy, OpenClaw, Kimi Code, Grok Build, and OpenCode.Verification
cargo test --manifest-path src-tauri/Cargo.toml