AgentSkillsCN

local-ci-simulation

通过将工作流任务映射至终端命令、逐步执行各阶段、隔离失败环节，并附上修复建议来模拟本仓库的 GitHub Actions CI。适用于 CI 处于红色状态时、在推送前验证修复效果时，或当用户要求分阶段进行本地 CI 验证时使用。务必先提出完整模拟方案，并在用户明确确认后再执行，因为该过程资源消耗较大。

SKILL.md
--- frontmatter
name: local-ci-simulation
description: Simulate GitHub Actions CI locally for this repository by mapping workflow jobs to terminal commands, running phases step-by-step, isolating failing stages, and reporting exact root causes with fix suggestions. Use when CI is red, when validating a fix before push, or when user asks for phased local CI verification. Always propose full simulation first and run it only after explicit user confirmation because it is resource-intensive.

Local Ci Simulation

Core Policy

  • •Offer full local CI simulation before running it.
  • •Start full simulation only after explicit user confirmation.
  • •Prefer lightweight targeted checks first when user has not requested full run.
  • •Preserve command order from workflow files and report exact failing phase.

Workflow

  1. •Inspect .github/workflows/tests.yml and .github/workflows/release.yml.
  2. •Build a phase list that mirrors CI jobs and step order.
  3. •Ask whether to run:
    • •focused mode: failing phase only;
    • •full mode: all heavy phases sequentially.
  4. •Run selected phases and stop at first failing command unless user asks to continue.
  5. •Distinguish non-fatal warnings from fatal errors.
  6. •Report:
    • •failing phase name;
    • •command;
    • •key error lines;
    • •minimal fix;
    • •post-fix recheck commands.

Repo Command Map

Use these commands as default simulation phases for this repository:

bash
make .yarnrc.yml
yarn install
yarn workspaces foreach -A --topological-dev run build
yarn eslint
yarn test
yarn test:coverage
yarn vitest run -c packages/v1-contexts/vitest.config.ts --typecheck.only --typecheck.checker tsc --typecheck.tsconfig packages/v1-contexts/tsconfig.json
yarn workspace @retailcrm/embed-ui-v1-components run storybook:build

Practical Notes

  • •Re-run the exact failing command after applying a fix.
  • •Run dependent follow-up checks for confidence (eslint, test, relevant build/storybook step).
  • •Mention when output is truncated and provide the most diagnostic lines.