AgentSkillsCN

innovate

识别对当前应用或计划具有高影响力的创新补充。适用于规划下一项功能、进行战略评审或当用户寻求创意时使用。

SKILL.md
--- frontmatter
name: innovate
description: Identify high-impact innovative additions to the current app or plan. Use when planning next features, during strategic reviews, or when the user asks for creative ideas.
allowed-tools: Read, Glob, Grep, Bash, WebSearch, WebFetch

Innovate

Identify the single highest-impact, feasible addition to make to the current app or plan. Not a list of 10 safe ideas -- one bold, high-leverage move backed by research and cross-referenced evidence.

Context Gathering

Before proposing anything, deeply understand the current state:

1. Read the Project

code
Read: README.md, CLAUDE.md, AGENTS.md
Glob: src/**/*.{ts,tsx,js,jsx,py,go,rs,swift}
Grep: "TODO|FIXME|HACK|FUTURE"

Understand:

  • •What the project does and who it's for
  • •Current architecture and tech stack
  • •Existing features and capabilities
  • •Known gaps, TODOs, and pain points

2. Read Any Specs or Plans

code
Read: PRODUCT_SPEC.md, TECHNICAL_SPEC.md, EXECUTION_PLAN.md, FEATURE_SPEC.md, VISION.md

Understand where the project is headed and what's already planned.

3. Assess the Landscape

Use WebSearch to research:

  • •What competitors or similar projects are doing
  • •Emerging technologies relevant to this domain
  • •User expectations in this space (current year)

4. Verify Research Findings

Cross-reference every key claim from Step 3 against at least 2 independent sources before using it to support a recommendation. Discard any finding that cannot be corroborated. Note which sources confirmed each claim in your internal evaluation (not in the final output).

Innovation Criteria

The proposal must score HIGH on ALL of these:

CriterionQuestion
LeverageDoes a small implementation unlock disproportionate value?
SurpriseWould this make someone say "I didn't know that was possible"?
FeasibilityCan this be built in days, not months?
FitDoes it align with the project's direction and users?
DefensibilityIs this hard to copy or does it create a moat?

Anti-Patterns (Do NOT Propose)

  • •Generic AI features ("add AI chat", "use LLMs for X")
  • •Incremental improvements that are obvious next steps
  • •Features that require massive infrastructure changes
  • •Ideas that sound cool but don't serve the actual users
  • •Anything already on the roadmap or in the execution plan

Analysis Process

Step 1: Identify Latent Potential

Look for:

  • •Underused data — What data does the app collect that it doesn't fully exploit?
  • •Workflow friction — Where do users leave the app to accomplish something?
  • •Combinatorial opportunities — What existing features could be combined in unexpected ways?
  • •Platform capabilities — What OS/browser/runtime features could be leveraged?
  • •Network effects — What becomes more valuable as usage grows?

Step 2: Generate Candidates (Internal)

Brainstorm 5-7 candidates internally. Do NOT output these. Evaluate each against the criteria table silently.

Step 3: Select the Winner

Pick the single strongest candidate. If no candidate scores HIGH on all five criteria, say so honestly rather than forcing a weak idea.

Output Format

code
╔══════════════════════════════════════════════════════════════════╗
║                        THE INNOVATION                           ║
╚══════════════════════════════════════════════════════════════════╝

{One-sentence pitch — what it is and why it matters}

────────────────────────────────────────────────────────────────────

WHY THIS, WHY NOW
─────────────────
{2-3 sentences on why this is the right move at this moment.
Reference specific things discovered in context gathering.}

WHAT IT CHANGES
───────────────
Before: {Current state — what users deal with today}
After:  {Future state — what becomes possible}

HOW IT WORKS
────────────
{3-5 bullet points on the core mechanism. Be specific enough
that a developer could start building from this description.}

• ...
• ...
• ...

IMPLEMENTATION SKETCH
─────────────────────
Effort: {LOW / MEDIUM / HIGH}
Files:  {Key files to create or modify}
Dependencies: {New libraries or services needed, if any}

Steps:
1. {First concrete step}
2. {Second concrete step}
3. {Third concrete step}

RISK & MITIGATION
─────────────────
Risk: {The main thing that could go wrong}
Mitigation: {How to handle it}

────────────────────────────────────────────────────────────────────

CRITERIA SCORECARD
──────────────────
Leverage:      ██████████ HIGH — {one-line justification}
Surprise:      ██████████ HIGH — {one-line justification}
Feasibility:   ██████████ HIGH — {one-line justification}
Fit:           ██████████ HIGH — {one-line justification}
Defensibility: ██████████ HIGH — {one-line justification}

Edge Cases

SituationAction
No README or project contextAsk the user to describe the project before proceeding
Project is too early stage (no code)Focus innovation on architecture/approach rather than features
Project is a library/SDK (not an app)Focus on DX innovations, API design, or ecosystem integrations
All ideas feel incrementalBe honest: "This project is well-optimized. Here's the best marginal gain I see:" and lower expectations
The argument is optionalIf invoked with an argument (e.g., /innovate payments), constrain the search to that domain

Review Your Output

Before presenting:

  • • Proposal is ONE idea, not a list
  • • Scores HIGH on all five criteria (or honestly notes where it doesn't)
  • • Implementation sketch is specific enough to act on
  • • Not something already planned or obvious
  • • Aligned with project direction