AgentSkillsCN

claude-complex-tasks

在交付前,通过实施结构化探索、范围规划和验证,引导克劳德完成复杂且多步骤的任务。

SKILL.md
--- frontmatter
name: claude-complex-tasks
description: Guide Claude on complex, multi-step tasks by enforcing structured discovery, scoped planning, and verification before delivery.
compatibility: Designed for Claude Code and similar agents.

Claude Complex Tasks

Use this skill when the task is complex, multi-step, or ambiguous and needs careful planning, validation, or staged delivery.

When to Use

  • The request spans multiple subsystems or files.
  • The outcome depends on external constraints (CI, build, integrations).
  • The task is risky, irreversible, or requires high accuracy.

Approach

  1. Clarify scope

    • Restate the goal in one sentence.
    • List assumptions and missing inputs.
    • Ask only one focused question if blocked.
  2. Plan the work

    • Break into 3-7 steps with clear outcomes.
    • Identify dependencies (tools, files, environment).
    • Decide what can be done in parallel.
  3. Gather context

    • Locate relevant files and read them first.
    • Verify conventions and existing patterns.
    • Identify edge cases and constraints.
  4. Execute in stages

    • Make smallest safe change to prove direction.
    • Iterate with tests or quick checks.
    • Keep changes scoped to the request.
  5. Verify

    • Run requested tests or minimal validation.
    • Confirm behavior against acceptance criteria.
    • Note any remaining risks.
  6. Close the loop

    • Ensure PR and issue linkage is explicit when applicable.
    • Merge only after required checks pass.
    • Confirm outcomes are reflected in docs/status where needed.
  7. Report

    • Summarize what changed and where.
    • Call out any decisions or tradeoffs.
    • Suggest next steps if needed.

Output Expectations

  • Prefer concise, structured updates.
  • Avoid long dumps of code; reference files instead.
  • Keep decisions explicit and traceable.