AgentSkillsCN

ci:monitoring

监控正在运行的 CI 作业，等待其完成，并为结果创建任务。

SKILL.md
--- frontmatter
name: ci:monitoring
description: Monitor running CI jobs, wait for completion, create tasks for results

CI Monitoring

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

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:

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

Example:

code
#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

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

4. Report Summary

Present final status table:

#CheckStatusAction
#33Kind Deploy & Testpassed-
#34HyperShift E2Efailed→ #35 RCA created
#35Kind CIRCAAgent empty response

CI Check Types

CheckTriggerDurationManual?
CodeQLAuto on push~2 minNo
Deploy & Test (Kind)Auto on push~12 minNo
E2E HyperShift/test comment~45 minYes
Build, Lint, SecurityAuto on push~1 minNo

HyperShift E2E

HyperShift requires manual trigger:

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

Monitor:

bash
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)