plan-feature
0
总安装量
14
周安装量
安装命令
npx skills add https://github.com/mblode/agent-skills --skill plan-feature
Agent 安装分布
cursor
12
claude-code
10
github-copilot
10
gemini-cli
10
opencode
9
codex
9
Skill 文档
Feature Planning
Create concrete, implementation-ready plans for features and complex changes.
DO NOT WRITE CODE during planning. Only explore, analyse, and document.
Planning threshold
Before deep planning, quickly confirm the scope:
- Plan in detail when changes affect 3+ files or multiple systems
- Plan in detail when requirements are unclear or architectural tradeoffs exist
- Skip heavyweight planning for single-file changes with clear requirements
- For small/obvious changes, give a short execution outline instead of a full spec
Planning workflow
1. Discovery
Ask targeted questions to uncover intent. For each question:
- Present 2-3 concrete options with tradeoffs
- Give your recommended option with clear reasoning
- One question at a time; wait for user response
- Skip questions already answered by the user
Critical questions:
- What problem are you solving? (user pain point, business goal)
- What should happen? (expected behaviour, success criteria)
- What should NOT happen? (constraints, edge cases to avoid)
- Who is this for? (user type, environment, scale)
- How will you verify it works? (testing approach, validation)
Speed-to-learning reference
- Use
ship-fast-loop.mdfor a lightweight shipping loop and feedback cadence.
2. Analysis
Explore the codebase systematically:
- Locate relevant files (document paths with line numbers)
- Map existing patterns (architecture, naming, data flow)
- Identify dependencies (what will be affected by changes)
- Find similar implementations (to maintain consistency)
- Note relevant standards (from
implement-frontend,define-architecture, etc)
Document findings:
- File:
path/to/file.ts:123– what it does, how it’s relevant - Pattern: existing approach for similar features
- Constraint: technical limitation or requirement
3. Planning
Create a concrete, ordered plan with:
For each change, specify:
- File path and approximate line number
- Exact function/component/class to modify
- What to add/remove/change (be specific)
- Why this change (how it fits the goal)
- Dependencies (what must happen first)
Plan structure:
## Goal
[One sentence: what we're building and why]
## Changes
### 1. [Description]
- File: `path/to/file.ts:45`
- Action: Add `functionName()` that does X
- Reasoning: Needed because Y
- Dependencies: None
### 2. [Description]
- File: `path/to/other.ts:89`
- Action: Modify `existingFunction()` to handle Z
- Reasoning: Integrates with change #1
- Dependencies: #1 must complete first
## Validation
- [ ] Tests pass
- [ ] Feature works for case A
- [ ] Edge case B is handled
- [ ] Follows `implement-frontend` (if frontend)
- [ ] No console logs or debug code
4. Standards reference
Explicitly note which standards apply:
- Frontend changes: reference
implement-frontend,audit-ui - UI changes: reference
design-ui - Motion: reference
ui-animation - Backend: reference
define-architecture - Typography: reference
audit-ui
Format: “This plan must follow implement-frontend for forms and type safety.”
Anti-patterns
Avoid vague plans:
- Bad: “Update the authentication system”
- Good: “Modify
auth/middleware.ts:34to addvalidateSession()that checks token expiry”
Avoid missing context:
- Bad: “Add error handling”
- Good: “Wrap API call in
auth/api.ts:67with try/catch, show toast on error peraudit-ui“
Avoid assuming knowledge:
- Bad: “Use the standard pattern”
- Good: “Follow the existing DAO pattern from
user/dao.ts:12(class-based with explicit types)”
Avoid incomplete acceptance criteria:
- Bad: “Make sure it works”
- Good: “Verify: (1) form submits on Enter, (2) shows inline errors, (3) disables submit during request”
Avoid ignoring standards:
- Bad: Plan uses
anytypes and manual form state - Good: Plan enforces
implement-frontend: noany, uses React Hook Form
Validation checklist
Before handing off the plan, verify:
- Someone can implement without asking questions
- File paths are valid with approximate line numbers
- Steps are ordered with clear dependencies
- Acceptance criteria are testable
- Relevant skill standards explicitly referenced
- Every decision is justified (the “why”)
- No ambiguous language (“update”, “improve”, “enhance” without specifics)