Sippy e2e post analysis step

ci-operator/step-registry/sippy-e2e/OWNERS

@@ -1,4 +1,11 @@
+reviewers:
 approvers:
+- deads2k
+- deepsm007
 - dgoodwin
+- neisw
+- petr-muller
 - stbenjam
-- DennisPeriquet
+- xueqzhan
+- sosiouxme
+- smg247

ci-operator/step-registry/sippy-e2e/sippy-e2e-claude-post-analysis/OWNERS

@@ -0,0 +1,11 @@
+reviewers:
+approvers:
+- deads2k
+- deepsm007
+- dgoodwin
+- neisw
+- petr-muller
+- stbenjam
+- xueqzhan
+- sosiouxme
+- smg247

ci-operator/step-registry/sippy-e2e/sippy-e2e-claude-post-analysis/sippy-e2e-sippy-e2e-claude-post-analysis-commands.sh

@@ -0,0 +1,111 @@
+#!/bin/bash
+
+set -o nounset
+set -o errexit
+set -o pipefail
+
+# Only run on presubmit jobs (PRs)
+if [[ "${JOB_TYPE:-}" != "presubmit" ]]; then
+    echo "Not a presubmit job. Skipping."
+    exit 0
+fi
+
+# Load GitHub token
+set +x
+if [[ -f "${GITHUB_TOKEN_PATH}" ]]; then
+    export GITHUB_TOKEN
+    GITHUB_TOKEN=$(cat "${GITHUB_TOKEN_PATH}")
+    echo "GitHub token loaded."
+else
+    echo "No GitHub token found at ${GITHUB_TOKEN_PATH}. Cannot comment on PR."
+    exit 0
+fi
+
+GCS_BUCKET="test-platform-results"
+GCS_PATH="gs://${GCS_BUCKET}/pr-logs/pull/${REPO_OWNER}_${REPO_NAME}/${PULL_NUMBER}/${JOB_NAME}/${BUILD_ID}/artifacts"
+PROW_URL="https://prow.ci.openshift.org/view/gs/${GCS_BUCKET}/pr-logs/pull/${REPO_OWNER}_${REPO_NAME}/${PULL_NUMBER}/${JOB_NAME}/${BUILD_ID}"
+
+# Check finished.json for each step to find failures
+echo "Checking for failed steps..."
+FAILED_STEPS=()
+
+while IFS= read -r finished; do
+    result=$(gsutil cat "${finished}" 2>/dev/null | jq -r '.result')
+    step_name=$(basename "$(dirname "${finished}")")
+    if [[ "${result}" == "FAILURE" ]]; then
+        echo "FAILED: ${step_name}"
+        FAILED_STEPS+=("${step_name}")
+    else
+        echo "PASSED: ${step_name}"
+    fi
+done < <(gsutil ls "${GCS_PATH}/**/finished.json" 2>/dev/null)
+
+if [[ ${#FAILED_STEPS[@]} -eq 0 ]]; then
+    echo "No failed steps found. Skipping analysis."
+    exit 0
+fi
+
+echo "Found ${#FAILED_STEPS[@]} failed step(s): ${FAILED_STEPS[*]}"
+
+WORKDIR=$(mktemp -d /tmp/claude-analysis-XXXXXX)
+cd "${WORKDIR}"
+
+copy_reports() {
+    cp "${WORKDIR}"/*.md "${ARTIFACT_DIR}/" 2>/dev/null || true
+    cp "${WORKDIR}"/*.html "${ARTIFACT_DIR}/" 2>/dev/null || true
+
+    # Archive Claude session for continue-session support
+    CLAUDE_HOME="/home/claude/.claude"
+    if [[ -d "${CLAUDE_HOME}/projects" ]]; then
+        echo "Archiving Claude session logs..."
+        if tar -czf "${ARTIFACT_DIR}/claude-sessions-$(date +%Y%m%d-%H%M%S).tar.gz" -C "${CLAUDE_HOME}" projects/ 2>/dev/null; then
+            touch "${SHARED_DIR}/claude-session-available"
+        fi
+    fi
+}
+trap copy_reports EXIT TERM INT
+
+ALLOWED_TOOLS="Bash Read Grep Glob WebFetch"
+
+PROMPT="You are analyzing a failed CI e2e test for the ${REPO_OWNER}/${REPO_NAME} project.
+
+## Job Information
+- PR: https://github.com/${REPO_OWNER}/${REPO_NAME}/pull/${PULL_NUMBER}
+- Prow Job: ${PROW_URL}
+- Failed steps: ${FAILED_STEPS[*]}
+
+## Artifacts
+The GCS artifacts for this job are at:
+${GCS_PATH}/
+
+Use gsutil to browse and read artifacts (build logs, test artifacts, etc.) from the failed steps.
+
+## Instructions
+1. Browse the artifacts from the failed steps to understand what went wrong.
+2. Fetch the PR diff with: gh pr diff ${PULL_NUMBER} --repo ${REPO_OWNER}/${REPO_NAME}
+3. Determine whether the PR's changes are likely responsible for the failure, or if this looks like a pre-existing/flaky issue.
+4. Leave a comment on the PR using: gh pr comment ${PULL_NUMBER} --repo ${REPO_OWNER}/${REPO_NAME} --body \"\$(cat <<'COMMENT'
+<your analysis>
+COMMENT
+)\"
+
+Your PR comment should be concise and helpful:
+- Start with a one-line summary (e.g. 'The e2e failure appears to be caused by ...' or 'This failure looks unrelated to this PR')
+- Include the specific test(s) that failed and why
+- If the PR is responsible, point to the specific change that likely caused it
+- If the PR is NOT responsible, briefly explain what the actual issue appears to be
+- Link to the Prow job: ${PROW_URL}
+- Keep it short - engineers are busy
+
+Important: Do NOT make any code changes. Only analyze and comment."
+
+echo "Invoking Claude for failure analysis..."
+timeout 600 claude \
+    --model "${CLAUDE_MODEL}" \
+    --allowedTools "${ALLOWED_TOOLS}" \
+    --output-format stream-json \
+    --max-turns 50 \
+    -p "${PROMPT}" \
+    --verbose 2>&1 | tee "${ARTIFACT_DIR}/claude-analysis.log" || true
+
+echo "Analysis complete."

ci-operator/step-registry/sippy-e2e/sippy-e2e-claude-post-analysis/sippy-e2e-sippy-e2e-claude-post-analysis-ref.metadata.json

@@ -0,0 +1,16 @@
+{
+	"path": "sippy-e2e/sippy-e2e-claude-post-analysis/sippy-e2e-sippy-e2e-claude-post-analysis-ref.yaml",
+	"owners": {
+		"approvers": [
+			"deads2k",
+			"deepsm007",
+			"dgoodwin",
+			"neisw",
+			"petr-muller",
+			"stbenjam",
+			"xueqzhan",
+			"sosiouxme",
+			"smg247"
+		]
+	}
+}

ci-operator/step-registry/sippy-e2e/sippy-e2e-claude-post-analysis/sippy-e2e-sippy-e2e-claude-post-analysis-ref.yaml

@@ -0,0 +1,38 @@
+ref:
+  as: sippy-e2e-sippy-e2e-claude-post-analysis
+  from_image:
+    namespace: ci
+    name: claude-ai-helpers
+    tag: latest
+  best_effort: true
+  commands: sippy-e2e-sippy-e2e-claude-post-analysis-commands.sh
+  credentials:
+  - namespace: test-credentials
+    name: hypershift-team-claude-prow
+    mount_path: /var/run/claude-code-service-account
+  - namespace: test-credentials
+    name: claude-payload-agent-github-token
+    mount_path: /var/run/github-token
+  env:
+  - name: CLAUDE_CODE_USE_VERTEX
+    default: "1"
+  - name: CLOUD_ML_REGION
+    default: "us-east5"
+  - name: ANTHROPIC_VERTEX_PROJECT_ID
+    default: "itpc-gcp-hybrid-pe-eng-claude"
+  - name: GOOGLE_APPLICATION_CREDENTIALS
+    default: "/var/run/claude-code-service-account/claude-prow"
+  - name: CLAUDE_MODEL
+    default: "claude-sonnet-4-6"
+  - name: GITHUB_TOKEN_PATH
+    default: "/var/run/github-token/token"
+  resources:
+    requests:
+      cpu: 500m
+      memory: 512Mi
+  timeout: 15m0s
+  grace_period: 1m0s
+  documentation: |-
+    Best-effort post step that uses Claude AI to analyze e2e test failures
+    in presubmit jobs and leaves a comment on the PR explaining the failure
+    and whether the PR's changes are likely responsible.

ci-operator/step-registry/sippy-e2e/sippy-e2e-setup/OWNERS

@@ -1,4 +1,11 @@
+reviewers:
 approvers:
+- deads2k
+- deepsm007
 - dgoodwin
+- neisw
+- petr-muller
 - stbenjam
-- DennisPeriquet
+- xueqzhan
+- sosiouxme
+- smg247

ci-operator/step-registry/sippy-e2e/sippy-e2e-setup/sippy-e2e-sippy-e2e-setup-ref.metadata.json

@@ -2,9 +2,15 @@
 	"path": "sippy-e2e/sippy-e2e-setup/sippy-e2e-sippy-e2e-setup-ref.yaml",
 	"owners": {
 		"approvers": [
+			"deads2k",
+			"deepsm007",
 			"dgoodwin",
+			"neisw",
+			"petr-muller",
 			"stbenjam",
-			"DennisPeriquet"
+			"xueqzhan",
+			"sosiouxme",
+			"smg247"
 		]
 	}
 }

ci-operator/step-registry/sippy-e2e/sippy-e2e-test/OWNERS

@@ -1,4 +1,11 @@
+reviewers:
 approvers:
+- deads2k
+- deepsm007
 - dgoodwin
+- neisw
+- petr-muller
 - stbenjam
-- DennisPeriquet
+- xueqzhan
+- sosiouxme
+- smg247

ci-operator/step-registry/sippy-e2e/sippy-e2e-test/sippy-e2e-sippy-e2e-test-ref.metadata.json

@@ -2,9 +2,15 @@
 	"path": "sippy-e2e/sippy-e2e-test/sippy-e2e-sippy-e2e-test-ref.yaml",
 	"owners": {
 		"approvers": [
+			"deads2k",
+			"deepsm007",
 			"dgoodwin",
+			"neisw",
+			"petr-muller",
 			"stbenjam",
-			"DennisPeriquet"
+			"xueqzhan",
+			"sosiouxme",
+			"smg247"
 		]
 	}
 }

ci-operator/step-registry/sippy-e2e/sippy-e2e-workflow.metadata.json

@@ -2,9 +2,15 @@
 	"path": "sippy-e2e/sippy-e2e-workflow.yaml",
 	"owners": {
 		"approvers": [
+			"deads2k",
+			"deepsm007",
 			"dgoodwin",
+			"neisw",
+			"petr-muller",
 			"stbenjam",
-			"DennisPeriquet"
+			"xueqzhan",
+			"sosiouxme",
+			"smg247"
 		]
 	}
 }

ci-operator/step-registry/sippy-e2e/sippy-e2e-workflow.yaml

@@ -5,5 +5,8 @@ workflow:
     - ref: sippy-e2e-sippy-e2e-setup
     test:
     - ref: sippy-e2e-sippy-e2e-test
+    post:
+    - ref: sippy-e2e-sippy-e2e-claude-post-analysis
+    - ref: openshift-claude-post
   documentation: |-
     Workflow to run the sippy e2e tests.