CI Monitoring

Monitor running CI pipelines and report results. Creates task items for each CI check being monitored.

Context-Safe Execution (MANDATORY)

CI log downloads MUST go to files. Status checks (gh pr checks) are small and OK inline.

export LOG_DIR="${LOG_DIR:-${WORKSPACE_DIR:-/tmp}/rossoctl-ci}"
mkdir -p "$LOG_DIR"

# When downloading logs after completion:
gh run view <run-id> --log-failed > $LOG_DIR/ci-run-<run-id>.log 2>&1; echo "EXIT:$?"
# Analyze in subagent: Task(subagent_type='Explore') with Grep

When to Use

• After pushing to a PR, to track CI completion
• When user says "monitor CI" or "check CI"
• When waiting for HyperShift E2E (long-running, needs manual trigger)
• To track multiple CI checks in parallel

Workflow

1. Create Monitoring Tasks

For each pending CI check, create a task:

TaskCreate: "<worktree> | <PR> | CI | Monitor | <check-name>"
  metadata: plan="ad-hoc", runner="main-session"

Example:

#33 [in_progress] mlflow-ci | PR#569 | CI | Monitor | Kind Deploy & Test
#34 [pending]     mlflow-ci | PR#569 | CI | Monitor | HyperShift E2E

2. Check Status Periodically

gh pr checks <PR-number>

3. On Completion

For each check that completes:

• Mark monitoring task as completed
• If passed: note in task description
• If failed: create RCA task and invoke rca:ci

TaskUpdate: #33 status=completed
TaskCreate: "<worktree> | <PR> | Kind CI | RCA | <failure summary>"

4. Report Summary

Present final status table:

#	Check	Status	Action
#33	Kind Deploy & Test	passed	-
#34	HyperShift E2E	failed	→ #35 RCA created
#35	Kind CI	RCA	Agent empty response

CI Check Types

Check	Trigger	Duration	Manual?
CodeQL	Auto on push	~2 min	No
Deploy & Test (Kind)	Auto on push	~12 min	No
E2E HyperShift	/test comment	~45 min	Yes
Build, Lint, Security	Auto on push	~1 min	No

HyperShift E2E

HyperShift requires manual trigger:

# Trigger via PR comment (done by user in GitHub UI)
# The workflow responds to issue_comment with "/test"

Monitor:

gh run view <run-id> --json status --jq '.status'
gh run view <run-id> | head -25

Task Tracking

On invocation:

1. TaskList - find existing CI monitoring tasks for this PR
2. Update completed checks
3. Create new monitoring tasks for pending checks
4. For failures, create RCA tasks linked to rca:ci or rca:hypershift

Naming: <worktree> | <PR> | CI | Monitor | <check-name>

Troubleshooting

HyperShift not starting

• Needs /test comment on PR
• Check if previous run is still cleaning up
• Check slot availability: CI has limited concurrent HyperShift clusters

CI running on old commit

• Push triggers new run on latest commit
• Old runs may still be in progress
• Check headSha matches expected commit

Related Skills

• ci:status - One-time status check
• rca:ci - RCA when CI fails
• tdd:ci - TDD iteration using CI
• tdd:kind - Local Kind testing (faster feedback)