AgentSkillsCN

next

恢复项目工作——探查当前状态，验证运行健康状况，清理陈旧文档，提出下一步任务建议。 在以下场景中使用：(1) 每日清晨开始工作时；(2) 长期代理运行结束后；(3) 用户询问“接下来做什么？”“我们该做些什么？”或其他类似问题；(4) 用户明确执行 /next 命令。

SKILL.md
--- frontmatter
name: next
description: |
  Resume work on a project - discover state, verify health, clean up stale docs, suggest next tasks.
  Use when: (1) starting work at beginning of day, (2) after a long agent work loop completes,
  (3) user asks "what's next?", "what should we work on?", or similar,
  (4) user explicitly runs /next.

Next: Project Resume Workflow

1. Gather Context

Read project planning files (if they exist):

  • •README.md - human-centric overview
  • •PLAN.md - agent-centric plan, status, future work
  • •SPEC.md - detailed specifications
  • •TODO.md - immediate tasks

Also check git status for uncommitted work or in-progress branches.

2. Discover & Run Verification

Find verification commands from (in priority order):

  1. •README.md - look for "Verification", "Development", "Quick Start" sections
  2. •pyproject.toml - scripts section, or infer from dependencies (pytest, mypy, ruff)
  3. •Makefile - check, test, lint targets
  4. •package.json - scripts section

Run discovered commands (typically: types, lint, tests). Note failures.

3. Clean Up Planning Docs

Auto-fix (no confirmation needed):

  • •Test counts that don't match actual (N tests passing → update N)
  • •Completed items still listed as in-progress
  • •Duplicate entries

Ask user about:

  • •Items marked "in progress" that appear stale
  • •Future work items that may now be done
  • •Sections that seem outdated but require judgment

4. Summarize & Present Options

Output a brief summary:

code
## Project State
- Health: [tests/types/lint status]
- Current: [what's in progress or recently completed]
- Cleaned: [what was auto-fixed in docs]

## Next Options
1. [Most impactful/logical next task]
2. [Alternative direction]
3. [Another option if applicable]

Wait for user to choose direction before proceeding.