feat: auto-merge Speakeasy SDK generation PRs (#332)

Pipe gh pr list output through jq instead of passing --arg to gh --jq,
and add a repo-level concurrency group to prevent cross-PR merge races.

.github/workflows/auto-merge-speakeasy-pr.yaml

@@ -0,0 +1,86 @@
+name: Auto-merge Speakeasy PR
+
+# Speakeasy mode:pr opens PRs (speakeasy-sdk-regen-*) and labels them
+# patch/minor/major. Merge automatically so sdk_publish.yaml can run.
+
+on:
+  pull_request:
+    types: [labeled]
+
+permissions:
+  contents: write
+  pull-requests: write
+
+concurrency:
+  group: auto-merge-speakeasy
+  cancel-in-progress: false
+
+jobs:
+  auto-merge:
+    if: |
+      github.event.sender.login == 'github-actions[bot]' &&
+      github.event.pull_request.user.login == 'github-actions[bot]' &&
+      startsWith(github.event.pull_request.head.ref, 'speakeasy-sdk-regen-') &&
+      contains(github.event.pull_request.title, '🐝 Update SDK') &&
+      contains(fromJSON('["patch", "minor", "major"]'), github.event.label.name)
+    runs-on: ubuntu-latest
+    steps:
+      - name: Close superseded Speakeasy PRs
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          set -euo pipefail
+          CURRENT_PR="${{ github.event.pull_request.number }}"
+
+          PRIOR_JSON=$(gh pr list \
+            --repo "${{ github.repository }}" \
+            --state open \
+            --search "head:speakeasy-sdk-regen-" \
+            --limit 500 \
+            --json number \
+            | jq -r --arg cur "$CURRENT_PR" \
+              '[.[].number | select(tostring != $cur)] | .[]')
+
+          if [ -z "$PRIOR_JSON" ]; then
+            echo "No superseded Speakeasy PRs to close"
+            exit 0
+          fi
+
+          mapfile -t PRIOR <<< "$PRIOR_JSON"
+
+          echo "Closing ${#PRIOR[@]} superseded Speakeasy PR(s)"
+          for N in "${PRIOR[@]}"; do
+            gh pr close "$N" --repo "${{ github.repository }}" --delete-branch \
+              --comment "Superseded by newer Speakeasy SDK generation PR #${CURRENT_PR}" \
+              || echo "::warning::Failed to close PR #$N (continuing)"
+            sleep 1
+          done
+
+      - name: Auto-merge Speakeasy PR
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          set -euo pipefail
+          PR_NUM="${{ github.event.pull_request.number }}"
+          REPO="${{ github.repository }}"
+
+          STATE=$(gh pr view "$PR_NUM" --repo "$REPO" --json state --jq .state)
+          if [ "$STATE" != "OPEN" ]; then
+            echo "PR #$PR_NUM is $STATE — skipping merge"
+            exit 0
+          fi
+
+          # Prefer GitHub auto-merge when required checks exist; otherwise
+          # squash-merge directly (same pattern as sdk-release-prs.yaml).
+          AUTO_MERGE_NOOP_PATTERN='is in clean status|protected branch rules|Branch does not have required protected branch rules'
+          if gh pr merge "$PR_NUM" --repo "$REPO" --squash --auto --delete-branch 2> /tmp/gh-merge.err; then
+            echo "Auto-merge enabled for Speakeasy PR #$PR_NUM"
+          elif grep -qiE "$AUTO_MERGE_NOOP_PATTERN" /tmp/gh-merge.err; then
+            echo "Auto-merge not applicable — merging PR #$PR_NUM directly"
+            cat /tmp/gh-merge.err >&2
+            gh pr merge "$PR_NUM" --repo "$REPO" --squash --delete-branch
+          else
+            echo "::error::Failed to enable auto-merge on Speakeasy PR #$PR_NUM"
+            cat /tmp/gh-merge.err >&2
+            exit 1
+          fi

.github/workflows/sdk_generation_for_spec_change.yaml

@@ -28,7 +28,7 @@ jobs:
     uses: speakeasy-api/sdk-generation-action/.github/workflows/workflow-executor.yaml@v15
     with:
       force: ${{ github.event.inputs.force }}
-      mode: direct
+      mode: pr
       set_version: ${{ github.event.inputs.set_version }}
     secrets:
       github_access_token: ${{ secrets.GITHUB_TOKEN }}

.speakeasy/in.openapi.yaml

@@ -12202,6 +12202,15 @@ components:
           image: '0'
           prompt: '0.00003'
           request: '0'
+        reasoning:
+          default_effort: medium
+          default_enabled: true
+          mandatory: false
+          supported_efforts:
+            - high
+            - medium
+            - low
+            - minimal
         supported_parameters:
           - temperature
           - top_p
@@ -12264,6 +12273,8 @@ components:
           $ref: '#/components/schemas/PerRequestLimits'
         pricing:
           $ref: '#/components/schemas/PublicPricing'
+        reasoning:
+          $ref: '#/components/schemas/ModelReasoning'
         supported_parameters:
           description: List of supported parameters for this model
           items:
@@ -12426,6 +12437,47 @@ components:
       description: Model to use for completion
       example: openai/gpt-4
       type: string
+    ModelReasoning:
+      description: Reasoning effort configuration. Omitted for non-reasoning models and dynamic router models.
+      example:
+        default_effort: medium
+        default_enabled: true
+        mandatory: false
+        supported_efforts:
+          - high
+          - medium
+          - low
+          - minimal
+      properties:
+        default_effort:
+          allOf:
+            - $ref: '#/components/schemas/ReasoningEffort'
+            - description: >-
+                Default reasoning effort when the client enables reasoning without specifying effort. Maps to
+                `reasoning.effort` in chat requests. When `"none"`, prefer omitting effort unless the user explicitly
+                disables reasoning.
+        default_enabled:
+          description: Default reasoning enabled state when the client does not set `reasoning.enabled`.
+          type: boolean
+        mandatory:
+          description: When true, reasoning cannot be disabled and effort "none" is rejected.
+          type: boolean
+        supported_efforts:
+          description: >-
+            Allowed reasoning effort values for this model, in descending effort order (highest first). Null means no
+            allowlist — all gateway effort values are accepted.
+          items:
+            $ref: '#/components/schemas/ReasoningEffort'
+          nullable: true
+          type: array
+        supports_max_tokens:
+          description: >-
+            Present and `true` when the model accepts `reasoning.max_tokens` in requests (Anthropic-style) instead of or
+            in addition to `reasoning.effort`. Omitted otherwise.
+          type: boolean
+      required:
+        - mandatory
+      type: object
     ModelResponse:
       description: Single model response
       example:
@@ -28837,7 +28889,7 @@ paths:
                     id: openai/gpt-4
                     knowledge_cutoff: null
                     links:
-                      details: /api/v1/models/openai/gpt-5.4/endpoints
+                      details: /api/v1/models/openai/gpt-4/endpoints
                     name: GPT-4
                     per_request_limits: null
                     pricing: