agents: retry GitHub Actions jobs and handle setup-bazel network flakes

.agents/skills/analyze-ci-failure/scripts/analyze_ci_failure.py

@@ -39,19 +39,45 @@ def fetch_log(job_name, build_id, job_id, output_path):
         log_url = f"https://buildkite.com/organizations/bazel/pipelines/rules-python-python/builds/{build_id}/jobs/{job_id}/download.txt"
 
     # Check if this is a GitHub Actions job
-    gh_match = re.search(r"github\.com/.*/job/(\d+)", log_url)
+    gh_repo_match = re.search(
+        r"github\.com/([^/]+/[^/]+)/actions/runs/\d+/job/(\d+)", log_url
+    )
+    gh_match = gh_repo_match or re.search(r"github\.com/.*/job/(\d+)", log_url)
     if not gh_match and "github" in job_name.lower() and re.match(r"^\d+$", job_id):
         gh_match = re.match(r"^(\d+)$", job_id)
 
     if gh_match:
-        gh_job_id = gh_match.group(1)
+        if gh_repo_match:
+            gh_repo = gh_repo_match.group(1)
+            gh_job_id = gh_repo_match.group(2)
+        else:
+            gh_repo = "bazel-contrib/rules_python"
+            gh_job_id = gh_match.group(1)
         print(f"📥 Fetching GitHub Action log for job {gh_job_id} using gh CLI...")
-        cmd = ["gh", "run", "view", "--job", gh_job_id, "--log"]
+        api_cmd = ["gh", "api", f"repos/{gh_repo}/actions/jobs/{gh_job_id}/logs"]
+        try:
+            res = subprocess.run(api_cmd, capture_output=True, text=True, check=True)
+            if res.stdout.strip():
+                with open(output_path, "w") as f:
+                    f.write(res.stdout)
+                return True
+        except Exception as e:
+            print(
+                f"⚠️ Failed to fetch GitHub log via gh api for job {gh_job_id}: {e}",
+                file=sys.stderr,
+            )
+
+        cmd = ["gh", "run", "view", "--job", gh_job_id, "--repo", gh_repo, "--log"]
         try:
             res = subprocess.run(cmd, capture_output=True, text=True, check=True)
-            with open(output_path, "w") as f:
-                f.write(res.stdout)
-            return True
+            if (
+                res.stdout.strip()
+                and "is still in progress; logs will be available when it is complete"
+                not in res.stdout
+            ):
+                with open(output_path, "w") as f:
+                    f.write(res.stdout)
+                return True
         except Exception as e:
             print(
                 f"⚠️ Failed to fetch GitHub log via gh CLI for job {gh_job_id}: {e}",
@@ -118,6 +144,7 @@ def parse_log(log_path):
                 "connection reset by peer",
                 "input/output error",
                 "errno 5",
+                "unable to find bazelisk version",
             ]
         ):
             if clean_line:
@@ -168,6 +195,13 @@ def create_plan(job_name, log_path, errors):
     ):
         is_flake = True
         flake_reason = "Transient runner host disk / Darwin sandbox I/O error (OSError: [Errno 5] Input/output error). This is an infrastructure flake, not a codebase defect."
+    elif any("unable to find bazelisk version" in e.lower() for e in errors):
+        is_flake = True
+        flake_reason = (
+            "Transient network error fetching Bazelisk release metadata during "
+            "GitHub Actions `setup-bazel` step (`Unable to find Bazelisk version`). "
+            "This is an infrastructure network flake, not a codebase bug."
+        )
 
     classification = (
         "⚡ **Classification**: **Infrastructure / Flake Issue** (Not a codebase bug)"
@@ -215,7 +249,8 @@ def main():
     os.makedirs(scratch_dir, exist_ok=True)
 
     safe_jname = re.sub(r"[^a-zA-Z0-9]", "_", args.job_name)
-    log_path = os.path.join(scratch_dir, f"ci_{safe_jname}_{args.job_id}.log")
+    safe_jid = re.sub(r"[^a-zA-Z0-9_-]", "_", args.job_id)
+    log_path = os.path.join(scratch_dir, f"ci_{safe_jname}_{safe_jid}.log")
 
     fetch_log(args.job_name, args.build_id, args.job_id, log_path)
 

.agents/skills/buildkite-retry-job/SKILL.md

@@ -1,21 +1,32 @@
 ---
 name: buildkite-retry-job
-description: Retry failed Buildkite jobs, unblock manual approval steps, or
-  rebuild full builds
+description: Retry failed Buildkite or GitHub Actions jobs, unblock manual
+  approval steps, or rebuild full builds
 ---
 
-Use `scripts/retry_buildkite_jobs.py` to retry individual jobs, unblock manual
-approval steps (such as `.bazelci/presubmit.yml` changes), or rebuild full
-Buildkite builds.
+Use `scripts/retry_buildkite_jobs.py` to retry individual Buildkite or GitHub
+Actions jobs, unblock manual approval steps (such as `.bazelci/presubmit.yml`
+changes), or rebuild full builds.
+
+When a GitHub Actions job fails early (for example, `Setup Bazel` failing with
+`Error: Unable to find Bazelisk version ...` due to a transient network error)
+while sibling matrix jobs in the same workflow run are still running, GitHub
+rejects rerun requests until the parent workflow run completes. The script
+automatically waits for an in-progress workflow run to finish before triggering
+`gh run rerun`.
 
 ### Retrying & Unblocking Modes
 
-1. **Default Retry & Unblock**:
-   Unblocks any blocked manual approval steps (`bk job unblock <job_id>
-   --no-input`) and retries specific failing jobs (`bk job retry <job_id>`):
+1. **Default Retry & Unblock (by PR, Buildkite URL, or GitHub Actions URL)**:
+   Unblocks any blocked Buildkite manual approval steps (`bk job unblock
+   <job_id> --no-input`), retries specific failing Buildkite jobs (`bk job
+   retry <job_id>`), and reruns failing GitHub Actions jobs/runs (`gh run
+   rerun`):
    ```bash
    ./.agents/skills/buildkite-retry-job/scripts/retry_buildkite_jobs.py \
      <pr_number_or_url>
+   ./.agents/skills/buildkite-retry-job/scripts/retry_buildkite_jobs.py \
+     "https://github.com/<owner>/<repo>/actions/runs/<run_id>/job/<job_id>"
    ```
 
 2. **Unblock Manual Approval Steps (e.g. `.bazelci/presubmit.yml` changes)**:
@@ -27,16 +38,21 @@ Buildkite builds.
    ```
 
 3. **Retry Specific Jobs by Name or ID**:
+   Retries matching Buildkite or GitHub Actions jobs by name/regex pattern,
+   Buildkite job UUID, or numeric GitHub Actions job ID:
    ```bash
    ./.agents/skills/buildkite-retry-job/scripts/retry_buildkite_jobs.py \
      <pr_number> --jobs "compile_pip_requirements"
    ./.agents/skills/buildkite-retry-job/scripts/retry_buildkite_jobs.py \
-     --job-id <job_uuid>
+     <pr_number> --jobs "Build linux"
+   ./.agents/skills/buildkite-retry-job/scripts/retry_buildkite_jobs.py \
+     --job-id <job_uuid_or_github_job_id>
    ```
 
 4. **Rebuild Full Build**:
-   Re-runs the entire pipeline build from scratch via `bk build rebuild`:
+   Re-runs the entire pipeline/workflow build from scratch via `bk build
+   rebuild` or `gh run rerun`:
    ```bash
    ./.agents/skills/buildkite-retry-job/scripts/retry_buildkite_jobs.py \
-     <pr_number> --rebuild
+     <pr_number_or_url> --rebuild
    ```