ci: improve pull request labelling logic

Author: seratch

.github/workflows/pr-labels.yml

@@ -98,6 +98,7 @@ jobs:
           mkdir -p "$output_dir"
           echo "output_file=${output_file}" >> "$GITHUB_OUTPUT"
       - name: Run Codex labeling
+        id: run_codex
         if: ${{ (github.event_name == 'workflow_dispatch' || steps.pr.outputs.is_fork != 'true') && github.actor != 'dependabot[bot]' }}
         uses: openai/codex-action@086169432f1d2ab2f4057540b1754d550f6a1189
         with:
@@ -113,6 +114,7 @@ jobs:
           PR_BASE_SHA: ${{ steps.pr.outputs.base_sha }}
           PR_HEAD_SHA: ${{ steps.pr.outputs.head_sha }}
           CODEX_OUTPUT_PATH: ${{ steps.codex-output.outputs.output_file }}
+          CODEX_CONCLUSION: ${{ steps.run_codex.conclusion }}
         run: |
           python - <<'PY'
           import json
@@ -125,9 +127,12 @@ jobs:
           pr_base_sha = os.environ.get("PR_BASE_SHA")
           pr_head_sha = os.environ.get("PR_HEAD_SHA")
           codex_output_path = pathlib.Path(os.environ["CODEX_OUTPUT_PATH"])
+          codex_conclusion = os.environ.get("CODEX_CONCLUSION", "").strip().lower()
           changed_files_path = pathlib.Path(".tmp/pr-labels/changed-files.txt")
           changes_diff_path = pathlib.Path(".tmp/pr-labels/changes.diff")
 
+          codex_ran = bool(codex_conclusion) and codex_conclusion != "skipped"
+
           def read_file_at(commit, path):
               if not commit:
                   return None
@@ -258,6 +263,68 @@ jobs:
           if dependencies_allowed:
               desired.add("dependencies")
 
+          if not codex_ran:
+              feature_prefixes = {
+                  "feature:realtime": (
+                      "src/agents/realtime/",
+                      "tests/realtime/",
+                      "examples/realtime/",
+                      "docs/realtime/",
+                  ),
+                  "feature:voice": (
+                      "src/agents/voice/",
+                      "tests/voice/",
+                      "examples/voice/",
+                      "docs/voice/",
+                  ),
+                  "feature:mcp": (
+                      "src/agents/mcp/",
+                      "tests/mcp/",
+                      "examples/mcp/",
+                      "examples/hosted_mcp/",
+                      "docs/mcp/",
+                  ),
+                  "feature:tracing": (
+                      "src/agents/tracing/",
+                      "tests/tracing/",
+                      "docs/tracing/",
+                      "examples/tracing/",
+                  ),
+                  "feature:sessions": (
+                      "src/agents/memory/",
+                      "tests/memory/",
+                      "examples/memory/",
+                      "docs/memory/",
+                  ),
+              }
+
+              for label, prefixes in feature_prefixes.items():
+                  if any(path.startswith(prefix) for prefix in prefixes for path in changed_files):
+                      desired.add(label)
+
+              if any(
+                  "chatcmpl" in path or "chatcompletions" in path for path in changed_files
+              ):
+                  desired.add("feature:chat-completions")
+
+              if any("litellm" in path for path in changed_files):
+                  desired.add("feature:lite-llm")
+
+              excluded_core_prefixes = (
+                  "src/agents/realtime/",
+                  "src/agents/voice/",
+                  "src/agents/mcp/",
+                  "src/agents/tracing/",
+                  "src/agents/memory/",
+                  "src/agents/extensions/",
+              )
+              if any(
+                  path.startswith("src/agents/")
+                  and not path.startswith(excluded_core_prefixes)
+                  for path in changed_files
+              ):
+                  desired.add("feature:core")
+
           allowed = {
               "documentation",
               "project",
@@ -299,7 +366,7 @@ jobs:
           ).strip()
           existing = {label for label in result.splitlines() if label}
 
-          managed = set(allowed)
+          managed = set(allowed) if codex_ran else set()
           to_add = sorted(desired - existing)
           to_remove = sorted((existing & managed) - desired)