perf(ci): cut E2E smoke wall-clock from ~8min to ~3min

.github/workflows/opencode-review.yml

@@ -23,6 +23,7 @@ permissions:
   issues: write
 
 concurrency:
+  # One group per PR across all three trigger paths; new run cancels stale review run
   group: opencode-review-${{ github.event.pull_request.number || github.event.issue.number || github.event.inputs.pr_number }}
   cancel-in-progress: true
 
@@ -49,9 +50,11 @@ jobs:
         env:
           GH_TOKEN: ${{ github.token }}
         run: |
+          # The runner workspace has no git repo on issue_comment (main-branch)
+          # runs, so gh cannot infer the repo — pass -R explicitly.
           if [ "${{ github.event_name }}" = "issue_comment" ]; then
             PR_NUMBER="${{ github.event.issue.number }}"
-            REF=$(gh pr view "$PR_NUMBER" --json headRefOid --jq '.headRefOid')
+            REF=$(gh pr view -R "${{ github.repository }}" "$PR_NUMBER" --json headRefOid --jq '.headRefOid')
             echo "ref=$REF" >> "$GITHUB_OUTPUT"
             echo "Checked out PR #$PR_NUMBER at $REF"
           else
@@ -77,6 +80,22 @@ jobs:
             conventions, then analyze the diff and delegate to your specialist
             subagents as appropriate. Post your synthesized review as a comment.
 
+            The PR thread is included as context (in <pull_request_comments> and
+            <pull_request_reviews>). It contains ALL previous comments and reviews
+            on this PR, including reviews from earlier runs of this bot.
+
+            DEDUPLICATION REQUIREMENT:
+            - Before reporting any finding, check whether that exact issue was
+              already reported in <pull_request_comments> or <pull_request_reviews>.
+            - Do NOT repeat findings that were already reported. Never restate an
+              existing comment.
+            - For each previously reported issue, only mention it again if its
+              status changed (e.g. was FIXED or REGRESSED since the last review).
+              Otherwise omit it entirely.
+            - Only raise NEW issues or issues whose status has changed.
+            - If nothing new to report, post exactly: "No new issues — see the
+              previous review comment."
+
             Focus areas:
             - Code: type safety, testing, telemetry, architecture consistency
             - Docs: AGENTS.md context loading table, subsystem AGENTS.md,

.github/workflows/pytest.yml

@@ -31,6 +31,13 @@ env:
   OPENAI_BASE_URL: ${{ secrets.OPENAI_BASE_URL }}
   TEST_DEFAULT_MODEL: ${{ secrets.TEST_DEFAULT_MODEL }}
 
+# Free up runners sooner: every push to a PR supersedes its in-flight PR run,
+# so the old run (and its queued e2e jobs) are cancelled instead of consuming
+# runner capacity that delays new commits.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
 jobs:
 
   lint:
@@ -271,9 +278,64 @@ jobs:
         run: python tests/check_no_mock_in_integration.py
         continue-on-error: true
 
-  e2e-smoke:
-    # L4a smoke: ~30s, PR-blocking. Runs in parallel with other test stages.
-    name: E2E smoke (L4a)
+  e2e-acp:
+    # L4a smoke: ACP protocol servers (stdio / streamable-http / WS).
+    # Runs in parallel with the opencode/misc e2e jobs; each job owns its own
+    # subprocess server cache instance, so wall-clock ≈ the slowest bucket
+    # instead of the serial sum.
+    name: E2E smoke (L4a · ACP)
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    needs: smoke
+    steps:
+      - uses: actions/checkout@v6
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+          cache-dependency-glob: pyproject.toml
+          cache-suffix: py3.13
+      - uses: actions/setup-python@v6
+        with:
+          python-version: "3.13"
+      - name: Install dependencies
+        run: uv sync --all-extras --no-group lint
+      - name: L4a smoke tests (ACP)
+        run: uv run pytest tests/e2e/test_acp*.py -m "e2e and not slow" --timeout=60 --tb=short --no-cov -q
+      - uses: actions/upload-artifact@v4
+        if: failure()
+        with:
+          name: e2e-acp-results
+          path: /tmp/pytest-*.txt
+          retention-days: 3
+
+  e2e-opencode:
+    name: E2E smoke (L4a · OpenCode)
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    needs: smoke
+    steps:
+      - uses: actions/checkout@v6
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+          cache-dependency-glob: pyproject.toml
+          cache-suffix: py3.13
+      - uses: actions/setup-python@v6
+        with:
+          python-version: "3.13"
+      - name: Install dependencies
+        run: uv sync --all-extras --no-group lint
+      - name: L4a smoke tests (OpenCode)
+        run: uv run pytest tests/e2e/test_opencode*.py -m "e2e and not slow" --timeout=60 --tb=short --no-cov -q
+      - uses: actions/upload-artifact@v4
+        if: failure()
+        with:
+          name: e2e-opencode-results
+          path: /tmp/pytest-*.txt
+          retention-days: 3
+
+  e2e-misc:
+    name: E2E smoke (L4a · AG-UI/APIs)
     runs-on: ubuntu-latest
     timeout-minutes: 20
     needs: smoke
@@ -289,19 +351,19 @@ jobs:
           python-version: "3.13"
       - name: Install dependencies
         run: uv sync --all-extras --no-group lint
-      - name: L4a smoke tests
-        run: uv run pytest -m "e2e and not slow" --timeout=60 --tb=short --no-cov -q
+      - name: L4a smoke tests (AG-UI / APIs)
+        run: uv run pytest tests/e2e/test_agui*.py tests/e2e/test_openai*.py tests/e2e/test_user*.py tests/e2e/test_turn_split_e2e.py tests/e2e/test_team_mode_smoke.py -m "e2e and not slow" --timeout=60 --tb=short --no-cov -q
       - uses: actions/upload-artifact@v4
         if: failure()
         with:
-          name: e2e-smoke-results
+          name: e2e-misc-results
           path: /tmp/pytest-*.txt
           retention-days: 3
 
   report:
     name: CI Report
     runs-on: ubuntu-latest
-    needs: [lint, format, typecheck, import-linter, smoke, unit, integration, core, vcr, cassette-hygiene, e2e-smoke]
+    needs: [lint, format, typecheck, import-linter, smoke, unit, integration, core, vcr, cassette-hygiene, e2e-acp, e2e-opencode, e2e-misc]
     if: always()
     steps:
       - name: Build summary table
@@ -320,7 +382,9 @@ jobs:
           echo '| Core tests | ${{ needs.core.result }} |' >> $GITHUB_STEP_SUMMARY
           echo '| VCR tests (L3 replay) | ${{ needs.vcr.result }} |' >> $GITHUB_STEP_SUMMARY
           echo '| Cassette hygiene | ${{ needs.cassette-hygiene.result }} |' >> $GITHUB_STEP_SUMMARY
-          echo '| E2E smoke (L4a) | ${{ needs.e2e-smoke.result }} |' >> $GITHUB_STEP_SUMMARY
+          echo '| E2E smoke (ACP) | ${{ needs.e2e-acp.result }} |' >> $GITHUB_STEP_SUMMARY
+          echo '| E2E smoke (OpenCode) | ${{ needs.e2e-opencode.result }} |' >> $GITHUB_STEP_SUMMARY
+          echo '| E2E smoke (AG-UI/APIs) | ${{ needs.e2e-misc.result }} |' >> $GITHUB_STEP_SUMMARY
           echo '' >> $GITHUB_STEP_SUMMARY
           if [ '${{ needs.lint.result }}' = 'failure' ] || \
              [ '${{ needs.format.result }}' = 'failure' ] || \
@@ -332,7 +396,9 @@ jobs:
              [ '${{ needs.core.result }}' = 'failure' ] || \
              [ '${{ needs.vcr.result }}' = 'failure' ] || \
              [ '${{ needs.cassette-hygiene.result }}' = 'failure' ] || \
-             [ '${{ needs.e2e-smoke.result }}' = 'failure' ]; then
+             [ '${{ needs.e2e-acp.result }}' = 'failure' ] || \
+             [ '${{ needs.e2e-opencode.result }}' = 'failure' ] || \
+             [ '${{ needs.e2e-misc.result }}' = 'failure' ]; then
             echo '### Some checks failed' >> $GITHUB_STEP_SUMMARY
             echo 'Download the artifact for test details.' >> $GITHUB_STEP_SUMMARY
           else

.opencode/agents/review-lead.md

@@ -28,19 +28,41 @@ single review comment. You do NOT edit files or run commands.
 ## Your Workflow
 
 1. **Analyze the diff** — Review all changed files in the pull request.
-2. **Categorize changes**:
+2. **Read the PR thread context** — `<pull_request_comments>` and
+   `<pull_request_reviews>` contain ALL prior comments and reviews on this PR,
+   including reviews from earlier runs of this bot. Read them before delegating
+   so your specialists know what was already reported.
+3. **Categorize changes**:
    - **Code changes**: anything under `src/`, `tests/`, `scripts/`
    - **Documentation changes**: `AGENTS.md`, `docs/`, `openspec/`
    - **Config changes**: `pyproject.toml`, `.github/`, `mkdocs.yml`
-3. **Delegate to specialists**:
+4. **Delegate to specialists**:
    - If there are code changes → invoke `@review-code` with the list of changed
      code files.
    - If there are documentation-relevant changes (new modules, new subsystems,
      changed conventions, or actual doc file edits) → invoke `@review-docs` with
      the list of changed files and a summary of what changed.
    - If the PR is docs-only with no code changes, skip `@review-code`.
    - If the PR is code-only with no documentation implications, skip `@review-docs`.
-4. **Synthesize findings** — Combine specialist reports into a unified review.
+5. **Synthesize findings** — Combine specialist reports into a unified review.
+
+## De-duplication (MANDATORY)
+
+This bot may review the same PR multiple times (reopened, ready_for_review,
+manual triggers) and will produce a new comment on every run. Follow these rules
+so findings are not repeated across runs:
+
+- Before reporting a finding, check whether the same issue was already reported in
+  `<pull_request_comments>` or `<pull_request_reviews>`.
+- Do NOT repeat a finding that already exists in the thread. Never restate an
+  existing comment.
+- Re-mention a previously reported issue ONLY if its status changed (e.g. it was
+  FIXED or REGRESSED since the last review), and say what changed.
+- Report only NEW issues, or previously-reported issues whose status changed.
+- When passing a changed-files list to a specialist, also pass the already-reported
+  findings so the specialist avoids re-reporting them.
+- If there is nothing new to report, post exactly: "No new issues — see the
+  previous review comment."
 
 ## Delegation Rules
 

opencode.json

@@ -0,0 +1,4 @@
+{
+  "$schema": "https://opencode.ai/config.json",
+  "default_agent": "review-lead"
+}

tests/e2e/test_opencode_misc.py

@@ -168,6 +168,9 @@ async def test_get_health(subprocess_server: SubprocessServer) -> None:
     strict=False,
 )
 @pytest.mark.known_bug
+# Known to hang until the per-test timeout fires; cap it so CI doesn't burn the
+# full 60s (the job-level timeout) just to reach the xfail.
+@pytest.mark.timeout(10)
 @pytest.mark.parametrize("subprocess_server", [_OPENCODE_PARAMS], indirect=True)
 async def test_post_init(subprocess_server: SubprocessServer) -> None:
     """C10.2: POST /session/{session_id}/init, verify 200."""