fix: remove report-bundle CodeBuild secondary artifact and add --local-run-dir support (#162)

* fix: remove report-bundle CodeBuild secondary artifact and add --local-run-dir support

* fix: address PR review feedback for codebuild workflow

- Replace report artifact fallback name with static 'report-head' to
  avoid invalid characters from branch names
- Narrow evaluation secondary artifact from '**/*' to specific YAML
  metric and report files only
- Bump upload-artifact from v6 to v7
- Add archive: false to all upload-artifact steps to prevent double-zip

Author: Kalindi-Dev

.github/workflows/codebuild.yml

@@ -238,7 +238,7 @@ jobs:
                     echo "========================================"
                     uv run python run.py full --rules-ref "$RULES_REF"
 
-                    # Package evaluation results into a zip for the trend report
+                    # Locate the evaluation run directory for trend report input
                     EVAL_RUN_DIR=$(ls -dt "$EVALUATOR_DIR/runs/"*/*/ 2>/dev/null | head -1)
                     LOCAL_BUNDLE_ARG=""
                     if [[ -n "$EVAL_RUN_DIR" ]]; then
@@ -249,13 +249,7 @@ jobs:
                         sed -i "s|rules_ref:.*|rules_ref: pr-$PR_NUMBER|" "$EVAL_RUN_DIR/run-meta.yaml"
                         echo "Patched run-meta.yaml: rules_ref -> pr-$PR_NUMBER"
                       fi
-                      BUNDLE_ZIP="/tmp/current-pr-bundle.zip"
-                      (cd "$EVAL_RUN_DIR" && zip -j "$BUNDLE_ZIP" \
-                        run-meta.yaml run-metrics.yaml test-results.yaml \
-                        contract-test-results.yaml quality-report.yaml \
-                        qualitative-comparison.yaml 2>/dev/null) && \
-                      LOCAL_BUNDLE_ARG="--local-bundle $BUNDLE_ZIP"
-                      cp "$BUNDLE_ZIP" "$CODEBUILD_SRC_DIR/.codebuild/report-bundle.zip"
+                      LOCAL_BUNDLE_ARG="--local-run-dir $EVAL_RUN_DIR"
                     else
                       echo "WARNING: No evaluation run folder found -- trend report will not include current PR"
                     fi
@@ -287,7 +281,16 @@ jobs:
               secondary-artifacts:
                 evaluation:
                   files:
-                    - '**/*'
+                    - '**/contract-test-results.yaml'
+                    - '**/evaluation-config.yaml'
+                    - '**/qualitative-comparison.yaml'
+                    - '**/quality-report.yaml'
+                    - '**/report.yaml'
+                    - '**/report.md'
+                    - '**/report.html'
+                    - '**/run-meta.yaml'
+                    - '**/run-metrics.yaml'
+                    - '**/test-results.yaml'
                   name: evaluation
                   discard-paths: no
                   base-directory: .codebuild/regression-runs
@@ -297,12 +300,6 @@ jobs:
                   name: trend
                   discard-paths: no
                   base-directory: .codebuild/trend-runs
-                report-bundle:
-                  files:
-                    - 'report-bundle.zip'
-                  name: report-bundle
-                  discard-paths: yes
-                  base-directory: .codebuild
 
       - name: Build ID
         if: always() && steps.cache-check.outputs.cache-hit != 'true'
@@ -356,44 +353,71 @@ jobs:
           key: ${{ env.CODEBUILD_PROJECT_NAME }}-${{ github.ref_name }}-${{ github.sha }}
 
       - name: Upload CodeBuild primary artifact
-        if: ${{ !env.ACT }} # incompatible with act (upload-artifact v6)
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6.0.0
+        if: ${{ !env.ACT }}
+        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
         with:
           name: ${{ env.CODEBUILD_PROJECT_NAME }}.zip
           path: ${{ github.workspace }}/.codebuild/downloads/${{ env.CODEBUILD_PROJECT_NAME }}.zip
           if-no-files-found: error
-      
+          archive: false
 
       - name: Upload Evaluation Report
-        if: ${{ !env.ACT }} # incompatible with act (upload-artifact v6)
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6.0.0
+        if: ${{ !env.ACT }}
+        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
         with:
           name: evaluation.zip
           path: ${{ github.workspace }}/.codebuild/downloads/evaluation.zip
           if-no-files-found: error
-          
+          archive: false
 
       - name: Upload Trend Report
-        if: ${{ !env.ACT }} # incompatible with act (upload-artifact v6)
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6.0.0
+        if: ${{ !env.ACT }}
+        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
         with:
           name: trend.zip
           path: ${{ github.workspace }}/.codebuild/downloads/trend.zip
           if-no-files-found: error
+          archive: false
 
 
+      - name: Extract Report Bundle from Evaluation
+        if: steps.cache-check.outputs.cache-hit != 'true'
+        run: |
+          DOWNLOADS="${ACT_CODEBUILD_DIR:-${GITHUB_WORKSPACE}/.codebuild/downloads}"
+          BUNDLE_DIR="$DOWNLOADS/report-bundle-staging"
+          mkdir -p "$BUNDLE_DIR"
+
+          YAML_FILES=(
+            run-meta.yaml
+            run-metrics.yaml
+            test-results.yaml
+            contract-test-results.yaml
+            quality-report.yaml
+            qualitative-comparison.yaml
+          )
+          for f in "${YAML_FILES[@]}"; do
+            unzip -j -o "$DOWNLOADS/evaluation.zip" "*/$f" -d "$BUNDLE_DIR" 2>/dev/null || true
+          done
+
+          if [[ -f "$BUNDLE_DIR/run-meta.yaml" ]]; then
+            (cd "$BUNDLE_DIR" && zip -j "$DOWNLOADS/report-bundle.zip" "${YAML_FILES[@]}" 2>/dev/null) || true
+            echo "Created report-bundle.zip from evaluation.zip contents"
+          else
+            echo "WARNING: run-meta.yaml not found in evaluation.zip — report bundle will be empty"
+          fi
+
       - name: Upload Report Bundle
         if: ${{ !env.ACT }}
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6.0.0
+        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
         with:
           name: >-
             ${{ github.event_name == 'pull_request'
                 && format('report-pr-{0}', github.event.pull_request.number)
                 || github.ref == 'refs/heads/main' && 'report-main'
-                || format('report-{0}', github.ref_name) }}
+                || format('report-head') }}
           path: ${{ github.workspace }}/.codebuild/downloads/report-bundle.zip
           if-no-files-found: warn
-    
+          archive: false
 
       - name: Upload artifacts to release
         if: startsWith(github.ref, 'refs/tags/v')

scripts/aidlc-evaluator/packages/trend-reports/src/trend_reports/__init__.py

@@ -2,14 +2,19 @@
 
 Usage:
     from trend_reports import collect_trend_data, render_trend_markdown
-    trend = collect_trend_data(zip_paths, baseline_path, repo, work_dir)
+    trend = collect_trend_data(bundle_paths, baseline_path, repo, work_dir)
     markdown = render_trend_markdown(trend)
 
 CLI:
     python -m trend_reports trend --baseline golden.yaml --format all
 """
 
-from trend_reports.collector import collect_trend_data, compute_deltas, sort_runs
+from trend_reports.collector import (
+    collect_from_directory,
+    collect_trend_data,
+    compute_deltas,
+    sort_runs,
+)
 from trend_reports.gate import check_regressions
 from trend_reports.models import (
     BaselineMetrics,
@@ -35,6 +40,7 @@
     "TrendReportError",
     "VersionDelta",
     "check_regressions",
+    "collect_from_directory",
     "collect_trend_data",
     "compute_deltas",
     "render_trend_html",

scripts/aidlc-evaluator/packages/trend-reports/src/trend_reports/__main__.py

@@ -76,6 +76,12 @@ def main() -> None:
         dest="local_bundles",
         help="Local zip bundle path(s) to include as additional data points",
     )
+    trend_parser.add_argument(
+        "--local-run-dir",
+        nargs="*",
+        dest="local_run_dirs",
+        help="Local run directory path(s) to include as additional data points",
+    )
 
     args = parser.parse_args()
 
@@ -104,6 +110,7 @@ def main() -> None:
             gate=args.gate,
             tags=args.tags,
             local_bundles=args.local_bundles,
+            local_run_dirs=args.local_run_dirs,
         )
         sys.exit(exit_code)
     except TrendReportError as exc:
@@ -122,6 +129,7 @@ def cmd_trend(
     gate: bool,
     tags: list[str] | None,
     local_bundles: list[str] | None = None,
+    local_run_dirs: list[str] | None = None,
 ) -> int:
     """Main orchestration.  Returns 0 on success, 1 on gate failure."""
     from .collector import collect_trend_data
@@ -145,30 +153,39 @@ def cmd_trend(
 
         # 2a. Release bundles (required)
         logger.info("Fetching release bundles from %s …", repo)
-        zip_paths = fetch_release_bundles(repo, tags, work_dir)
-        logger.info("Fetched %d release bundle(s)", len(zip_paths))
+        bundle_paths = fetch_release_bundles(repo, tags, work_dir)
+        logger.info("Fetched %d release bundle(s)", len(bundle_paths))
 
         # 2b. Local bundles (from --local-bundle flag)
         if local_bundles:
             for bundle_str in local_bundles:
                 bundle_path = Path(bundle_str)
                 if not bundle_path.exists():
                     raise TrendReportError(f"Local bundle not found: {bundle_path}")
-                zip_paths.append(bundle_path)
+                bundle_paths.append(bundle_path)
                 logger.info("Added local bundle: %s", bundle_path)
 
+        # 2b-2. Local run directories (from --local-run-dir flag)
+        if local_run_dirs:
+            for dir_str in local_run_dirs:
+                dir_path = Path(dir_str)
+                if not dir_path.is_dir():
+                    raise TrendReportError(f"Local run directory not found: {dir_path}")
+                bundle_paths.append(dir_path)
+                logger.info("Added local run directory: %s", dir_path)
+
         # 2c. Remote pre-release bundles (from GitHub Actions Artifacts)
         logger.info("Fetching pre-release bundles …")
         prerelease_paths = fetch_prerelease_bundles(repo, cache_prefix, work_dir)
         if prerelease_paths:
             logger.info("Fetched %d pre-release bundle(s)", len(prerelease_paths))
-            zip_paths.extend(prerelease_paths)
+            bundle_paths.extend(prerelease_paths)
         else:
             logger.info("No pre-release bundles found — continuing with releases only")
 
         # 3. Collect and assemble
         logger.info("Parsing bundles …")
-        trend = collect_trend_data(zip_paths, Path(baseline), repo, work_dir)
+        trend = collect_trend_data(bundle_paths, Path(baseline), repo, work_dir)
         logger.info("Assembled trend data for %d runs", len(trend.runs))
 
         # 4. Render into a timestamped subdirectory

scripts/aidlc-evaluator/packages/trend-reports/src/trend_reports/collector.py

@@ -1,4 +1,4 @@
-"""Zip extraction, YAML parsing, run classification, and trend assembly."""
+"""Zip/directory extraction, YAML parsing, run classification, and trend assembly."""
 
 from __future__ import annotations
 
@@ -32,7 +32,7 @@
 
 logger = logging.getLogger(__name__)
 
-# The YAML files we expect inside every report zip.
+# The YAML files we expect inside every report bundle (zip or directory).
 REQUIRED_YAML = {
     "run-meta": "run-meta.yaml",
     "run-metrics": "run-metrics.yaml",
@@ -306,13 +306,15 @@ def classify_run(rules_ref: str) -> tuple[RunType, str, SemVer | None, int | Non
 # ---------------------------------------------------------------------------
 
 
-def collect_from_zip(zip_path: Path, work_dir: Path) -> RunData:
-    """Extract a zip bundle and parse all YAML files into a RunData."""
-    run_dir = extract_zip(zip_path, work_dir)
+def _collect_from_run_dir(run_dir: Path, source_label: str) -> RunData:
+    """Parse YAML files in *run_dir* into a RunData.
+
+    *source_label* is used in error messages (e.g. the zip path or directory).
+    """
     yaml_files = find_yaml_files(run_dir)
 
     if "run-meta" not in yaml_files:
-        raise CollectorError(f"run-meta.yaml missing from {zip_path} — cannot classify run")
+        raise CollectorError(f"run-meta.yaml missing from {source_label} — cannot classify run")
 
     meta = parse_run_meta(yaml_files["run-meta"])
     run_type, label, semver, pr_number = classify_run(meta.config.rules_ref)
@@ -365,6 +367,23 @@ def collect_from_zip(zip_path: Path, work_dir: Path) -> RunData:
     )
 
 
+def collect_from_zip(zip_path: Path, work_dir: Path) -> RunData:
+    """Extract a zip bundle and parse all YAML files into a RunData."""
+    run_dir = extract_zip(zip_path, work_dir)
+    return _collect_from_run_dir(run_dir, source_label=str(zip_path))
+
+
+def collect_from_directory(dir_path: Path) -> RunData:
+    """Parse all YAML files from a plain directory into a RunData.
+
+    Unlike :func:`collect_from_zip`, no extraction step is needed.
+    The directory must contain the expected YAML files directly.
+    """
+    if not dir_path.is_dir():
+        raise CollectorError(f"Not a directory: {dir_path}")
+    return _collect_from_run_dir(dir_path, source_label=str(dir_path))
+
+
 def load_baseline(golden_path: Path) -> BaselineMetrics:
     """Parse a golden.yaml baseline file into BaselineMetrics."""
     if not golden_path.exists():
@@ -443,12 +462,12 @@ def compute_deltas(runs: list[RunData]) -> list[VersionDelta]:
 
 
 def collect_trend_data(
-    zip_paths: list[Path],
+    bundle_paths: list[Path],
     baseline_path: Path,
     repo: str,
     work_dir: Path | None = None,
 ) -> TrendData:
-    """Parse all zip bundles and assemble a TrendData."""
+    """Parse all bundles (zip files or directories) and assemble a TrendData."""
     import tempfile
 
     if work_dir is None:
@@ -457,13 +476,16 @@ def collect_trend_data(
     baseline = load_baseline(baseline_path)
 
     runs: list[RunData] = []
-    for zp in zip_paths:
-        logger.info("Collecting data from %s …", zp.name)
+    for bp in bundle_paths:
+        logger.info("Collecting data from %s …", bp.name)
         try:
-            run = collect_from_zip(zp, work_dir)
+            if bp.is_dir():
+                run = collect_from_directory(bp)
+            else:
+                run = collect_from_zip(bp, work_dir)
             runs.append(run)
         except CollectorError as exc:
-            logger.warning("Skipping %s: %s", zp.name, exc)
+            logger.warning("Skipping %s: %s", bp.name, exc)
 
     if not runs:
         raise CollectorError("No runs could be parsed from the provided bundles.")