Dev

.github/workflows/marketplace-publish.yml

@@ -57,14 +57,6 @@ jobs:
           git tag -a "v${current_package_version}" -m "Release v${current_package_version}"
           git push origin "v${current_package_version}" --no-verify
           echo "Successfully created and pushed git tag v${current_package_version}"
-      - name: Publish Extension
-        env:
-          VSCE_PAT: ${{ secrets.VSCE_PAT }}
-          OVSX_PAT: ${{ secrets.OVSX_PAT }}
-        run: |
-          current_package_version=$(node -p "require('./src/package.json').version")
-          pnpm --filter roo-cline publish:marketplace
-          echo "Successfully published version $current_package_version to VS Code Marketplace"
       - name: Create GitHub Release
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/nightly-publish.yml

@@ -42,11 +42,3 @@ jobs:
           EOF
       - name: Build VSIX
         run: pnpm vsix:nightly # Produces bin/roo-code-nightly-0.0.[count].vsix
-      - name: Publish to VS Code Marketplace
-        env:
-          VSCE_PAT: ${{ secrets.VSCE_PAT }}
-        run: npx vsce publish --packagePath "bin/$(/bin/ls bin | head -n1)"
-      - name: Publish to Open VSX Registry
-        env:
-          OVSX_PAT: ${{ secrets.OVSX_PAT }}
-        run: npx ovsx publish "bin/$(ls bin | head -n1)"

build.sh

@@ -0,0 +1,4 @@
+corepack prepare pnpm@10.8.1 --activate
+pnpm -v
+cd src
+pnpm vsix

docs/superpowers/specs/2026-05-09-openai-compatible-reasoning-design.md

@@ -0,0 +1,129 @@
+# OpenAI-compatible reasoning payload shim
+
+Date: 2026-05-09
+
+## Problem
+
+OpenAI-compatible request path currently sends `reasoningEffort` into AI SDK, but it does not add Responses-style `reasoning` payload.
+
+User expectation for OpenAI-compatible requests with reasoning enabled:
+
+- keep `reasoning_effort`
+- also send `reasoning: { effort, summary: "auto" }`
+
+This should match OpenAI Codex / OpenAI Native behavior shape, while staying inside OpenAI-compatible chat completions path.
+
+## Goals
+
+- When resolved model params include `reasoningEffort`, send both:
+    - `reasoning_effort`
+    - `reasoning: { effort, summary: "auto" }`
+- Apply same behavior for:
+    - `createMessage`
+    - `completePrompt`
+- Keep `reasoning_effort` value unchanged, including extended values like `xhigh` or `none` when current model/settings resolve them.
+- Leave request unchanged when reasoning is off.
+
+## Non-goals
+
+- No change to `openai.ts`.
+- No change to `openai-native.ts` or `openai-codex.ts`.
+- No UI/settings changes.
+- No new provider capability detection.
+
+## Design
+
+### 1) Pass reasoning to AI SDK provider
+
+`OpenAICompatibleHandler` will build `providerOptions` for `streamText` and `generateText`.
+
+Use generic AI SDK key:
+
+```ts
+providerOptions: {
+  openaiCompatible: {
+    reasoningEffort: model.reasoningEffort,
+  },
+}
+```
+
+Only include this block when `model.reasoningEffort` is defined.
+
+### 2) Add request-body transform hook
+
+Extend `OpenAICompatibleConfig` with AI SDK `transformRequestBody` hook and pass it into `createOpenAICompatible(...)`.
+
+Hook behavior:
+
+- if body has no `reasoning_effort`, return body unchanged
+- if body has `reasoning_effort`, add/update:
+    ```ts
+    reasoning: {
+      effort: body.reasoning_effort,
+      summary: "auto",
+    }
+    ```
+- keep `reasoning_effort` in body
+- preserve other fields
+
+This hook must apply to both streaming and non-streaming calls, because AI SDK uses same provider transform for both.
+
+### 3) Keep logic localized
+
+Implement inside `src/api/providers/openai-compatible.ts` only.
+
+Reason:
+
+- current `OpenAICompatibleHandler` is only consumer path
+- current OpenAI-compatible provider family is Moonshot
+- local change keeps scope tight and avoids touching shared reasoning transforms for unrelated providers
+
+## Data flow
+
+1. `MoonshotHandler.getModel()` resolves final `reasoningEffort` from settings/model defaults.
+2. `OpenAICompatibleHandler.createMessage()` / `completePrompt()` pass `providerOptions.openaiCompatible.reasoningEffort` into AI SDK.
+3. AI SDK serializes top-level `reasoning_effort`.
+4. `transformRequestBody` adds `reasoning: { effort, summary: "auto" }` when `reasoning_effort` exists.
+5. Provider receives both fields.
+
+## Fallback behavior
+
+- No reasoning selected: no `providerOptions`, no `reasoning`, no `reasoning_effort`.
+- Disabled reasoning: same as above.
+- Extended values: pass through unchanged.
+- If body already contains `reasoning`, overwrite `effort` and `summary` only; do not drop unrelated keys.
+
+## Tests
+
+Add tests around current Moonshot/OpenAI-compatible path:
+
+1. **Constructor wiring**
+
+    - `createOpenAICompatible` receives `transformRequestBody`.
+
+2. **Streaming request**
+
+    - `createMessage()` passes `providerOptions.openaiCompatible.reasoningEffort` when enabled.
+    - transform adds both `reasoning_effort` and `reasoning`.
+
+3. **Non-streaming request**
+
+    - `completePrompt()` uses same providerOptions path.
+    - transform adds both fields.
+
+4. **Disabled path**
+
+    - when reasoning is disabled, request has neither `reasoning_effort` nor `reasoning`.
+
+5. **Extended effort path**
+    - `xhigh` stays `xhigh` in both fields.
+
+Prefer a small pure helper or captured transform callback in test so request-body mapping is asserted directly, not through SDK internals.
+
+## Acceptance criteria
+
+- OpenAI-compatible requests with reasoning enabled send both fields.
+- `reasoning_effort` still exists.
+- `reasoning.summary` is always `auto`.
+- Behavior matches in stream and completion paths.
+- No changes to unrelated provider paths.

docs/superpowers/specs/2026-05-10-disable-commit-push-checks-design.md

@@ -0,0 +1,89 @@
+# Disable local commit/push checks design
+
+Date: 2026-05-10
+
+## Problem
+Repo currently enforces local Git checks through Husky hooks:
+- `.husky/pre-commit` blocks commits to `main`
+- `.husky/pre-commit` runs `lint-staged`
+- `.husky/pre-commit` runs `pnpm lint`
+- `.husky/pre-push` blocks pushes to `main`
+- `.husky/pre-push` runs `check-types`
+- `.husky/pre-push` may run tests
+- `.husky/pre-push` prints changeset reminder
+
+User wants all local commit/push enforcement removed.
+User also wants `lint-staged` config removed from `package.json`.
+
+## Goals
+- Remove all local Husky checks from commit and push workflow.
+- Remove branch protection logic from local Husky hooks.
+- Remove `lint-staged` configuration from root `package.json`.
+- Keep normal repo scripts (`lint`, `check-types`, `test`, etc.) available for manual use.
+
+## Non-goals
+- No CI workflow changes.
+- No package script removals for `lint`, `check-types`, `test`, or `build`.
+- No Husky uninstall.
+- No change to hooks other than commit/push hooks.
+
+## Design
+### 1) Delete commit/push hook entrypoints
+Delete:
+- `.husky/pre-commit`
+- `.husky/pre-push`
+
+Result:
+- no local commit blocking on `main`
+- no local push blocking on `main`
+- no automatic `lint-staged`, `lint`, `check-types`, test, or changeset reminder during commit/push
+
+### 2) Keep Husky base wiring intact
+Keep:
+- `.husky/_/**`
+- root `package.json` script: `"prepare": "husky"`
+
+Reason:
+- smallest change for requested scope
+- avoids changing global Git hook bootstrap behavior
+- commit/push checks disappear because entrypoint hook files are gone
+
+### 3) Remove lint-staged config from package.json
+Delete root-level `lint-staged` block from `package.json`.
+
+Reason:
+- user explicitly asked to remove it
+- after deleting `.husky/pre-commit`, this config is unused
+- removes stale config from repo root
+
+## Data flow after change
+Before:
+1. `git commit` -> Husky runs `.husky/pre-commit`
+2. hook blocks `main`, runs `lint-staged`, runs `pnpm lint`
+3. `git push` -> Husky runs `.husky/pre-push`
+4. hook blocks `main`, runs `check-types`, optional tests, changeset reminder
+
+After:
+1. `git commit` -> no repo-defined pre-commit entrypoint
+2. `git push` -> no repo-defined pre-push entrypoint
+3. manual checks still possible via package scripts
+
+## Risks
+- Developers can commit/push broken code locally without warning.
+- Developers can commit/push directly to `main` locally unless remote branch protections exist.
+- Formatting on staged files will no longer run automatically.
+
+These risks are accepted by request.
+
+## Testing
+- Verify `.husky/pre-commit` no longer exists.
+- Verify `.husky/pre-push` no longer exists.
+- Verify `package.json` no longer contains `lint-staged` block.
+- Optional smoke check: `git status` and `git commit` path should no longer invoke repo checks.
+
+## Acceptance criteria
+- `.husky/pre-commit` deleted.
+- `.husky/pre-push` deleted.
+- root `package.json` no longer contains `lint-staged` config.
+- `prepare: husky` remains unchanged.
+- local commit/push no longer run repo-defined checks.