AgentSkillsCN

Plan Lite

轻量级规划

SKILL.md

Plan-Lite Skill

Purpose

A lightweight, focused prompting approach for small or scoped tasks. Delivers clear, concrete, and reliable outputs without losing rigor or determinism.

When to Use

Use this skill when the task is:

  • One feature or module
  • A focused code change
  • A small architectural decision
  • A config/schema update
  • A targeted improvement within a larger plan
  • A bug fix or refactor

When NOT to Use

Use the full plan skill instead if:

  • The task spans multiple subsystems
  • A full architecture or roadmap is requested
  • Governance or evaluation frameworks are involved
  • The user asks for a full implementation plan

Invocation

code
"use plan-lite skill"
"quick plan"
"focused change for <feature>"

Core Rules (Always Follow)

1. Read First

  • If the user provides code, markdown, YAML, or plans: read them before responding.
  • Explicitly respect constraints from provided artifacts.

2. Be Concrete

  • Map every recommendation to a file, function, schema field, or config key.
  • Avoid abstract or generic advice.

3. Preserve Determinism

  • If randomness exists, require explicit seed handling.
  • Never introduce hidden or global randomness.

4. Don't Break Existing Behavior

  • Prefer additive changes.
  • Use defaults and feature flags when behavior changes.

5. Small Changes, Clear Steps

  • Favor minimal diffs and PR-sized increments.
  • Avoid rewrites unless explicitly requested.

Standard Response Structure

Use this structure unless the user explicitly asks otherwise:

markdown
## 1. Context & Constraints
- What was read
- What constraints matter

## 2. Proposed Change
- What will change and why

## 3. Exact Implementation
- File(s) to edit (exact paths)
- Code snippets or diffs
- Config/schema changes (if any)
- Integration point (where to call it)

## 4. Tests / Validation
- Tests to add (at least one smoke test + regression/determinism where relevant)
- How to run them
- How to verify correctness

## 5. Risks / Notes
- Edge cases or follow-ups (brief)

Output

Save the plan to .claude/plans/plan-lite-YYYY-MM-DD.md in the project root, where YYYY-MM-DD is the current date.

If multiple plans are created on the same day, append a sequence number: plan-lite-YYYY-MM-DD-2.md.


Definition of Done

Output is complete when:

  • Code compiles/runs with minimal edits
  • Integration point is explicit (where to call it)
  • Tests included or clear validation steps provided
  • Behavior unchanged when feature flag is off (if applicable)

Quality Bar (Self-Check)

Before responding, confirm:

  • I read the provided artifacts
  • I referenced the actual code/schema
  • I proposed a minimal, concrete change
  • I didn't invent abstractions or rewrite unnecessarily
  • I explained how to test or validate the change
  • Integration point is clear

Anti-Patterns (Never Do)

  • Propose changes without reading provided files
  • Give abstract advice without file/function references
  • Introduce unnecessary abstractions
  • Rewrite when a small edit suffices
  • Skip validation/testing guidance