AgentSkillsCN

next-steps

根据仓库概要,结合验收标准、复杂度、影响范围,以及建议的分支与 PR,提炼出一份优先级明确、可落地执行的任务清单(不超过 20 条)。

SKILL.md
--- frontmatter
name: next-steps
description: "Produce a prioritized, actionable task list (<=20) from a repo summary with acceptance criteria, complexity, impact, and suggested branches/PRs."

SKILL: next-steps

Summary

Generate a comprehensive but concise set (no more than 20) of prioritized next steps from summary.json. Items should cover fixes, enhancements, refactors, new features, and documentation. Each item must include acceptance criteria, a complexity score (1-5), an impact level (low|medium|high), and suggested branch/commit messages.

When to run

  • Immediately after review-repo completes.

Inputs

  • summary_path (required)
  • allow_insert_todos (bool, default: false)
  • allow_execute (bool, default: false)
  • max_tasks (int, default: 20)

Outputs

  • run/skills/next-steps/<timestamp>/tasks.json
  • run/skills/next-steps/<timestamp>/tasks.txt
  • Optional SQL inserts into the todos table when allowed.

Task schema (example)

code
{ "id":"add-ci-checks", "title":"Add CI checks to run scripts/setup.sh", "description":"...", "priority":1, "complexity":2, "impact":"medium", "files":["scripts/setup.sh"], "branch":"chore/add-ci-checks", "commit_msg":"chore(ci): add workflow to run scripts/setup.sh" }

Guidelines

  • Produce a comprehensive but concise list (<= 20 items) that spans fixes, enhancements, refactors, new features, and documentation improvements.
  • Each item must include: acceptance criteria, complexity (1-5), impact (low|medium|high), priority, affected files or globs, suggested branch name and commit message.
  • Do not include time or duration estimates.
  • Mark tasks that require network access, external resources, or privileged actions as requires-approval or requires-sandbox.
  • When inserting into todos, use kebab-case IDs and include complexity and impact metadata in the description or as structured fields.
  • Prefer actionable items that can be implemented as a single PR; include file paths or short code snippets to make them easy to implement.

References and how to find evidence in the codebase

  • Primary input: the provided summary_path (summary.json). If not provided, use the latest run/skills/review-repo/*/summary.json.
  • Scan top-level docs: README.md, AGENT.md, docs/, and docs/* for architecture and design.
  • Inspect .github/skills/*/SKILL.md, scripts/skills/, and scripts/ for automation and helper scripts.
  • Search source and scripts for TODO/FIXME and domain keywords: agent, swarm, autonom, self[- ]?improv, optimi, evolv, decentral, secure, copilot, supervisor, daemon, director.
  • Use git metadata (HEAD, recent commits via git log -n 50 --pretty=oneline) to prioritize recently modified components.
  • When referencing files, use relative paths and include examples of lines or snippets to change.