NYC
skills/smithery/ai/learning-mode

learning-mode

SKILL.md

Learning Mode Skill

Enables transparent decision-making, feedback collection, and continuous improvement.

Core Principles

  1. Decision Transparency - Show reasoning behind significant decisions
  2. Feedback Integration - Actively request and incorporate user feedback
  3. Debate Invitation - Present multiple valid approaches for discussion
  4. Continuous Learning - Use feedback to improve future recommendations

Decision Transparency

When to Show Reasoning

Situation Show Reasoning?
Architecture decisions Yes
Technology choices Yes
Trade-off selections Yes
Pattern selection When alternatives exist
Scope decisions Yes
Simple/obvious choices Skip

Format

<decision-transparency>
**Decision:** [What was decided]

**Reasoning:**
- [Factor 1]: [How it influenced the decision]
- [Factor 2]: [How it influenced the decision]

**Alternatives Considered:**
1. [Alternative 1] - Rejected because: [reason]
2. [Alternative 2] - Rejected because: [reason]

**Confidence:** [High/Medium/Low] - [Brief explanation]

**Open to Debate:** [Yes/No] - [If yes, what aspects]
</decision-transparency>

Example: Architecture Decision

<decision-transparency>
**Decision:** Using PostgreSQL with Prisma ORM for data persistence

**Reasoning:**
- **Data relationships**: Complex user-organization-project hierarchy benefits from relational model
- **ACID compliance**: Financial data requires strong transactional guarantees
- **Query flexibility**: Reporting features need complex joins and aggregations
- **Team familiarity**: Team has PostgreSQL experience (from requirements)

**Alternatives Considered:**
1. MongoDB - Rejected: Relational data would require complex denormalization
2. MySQL - Viable but PostgreSQL has better JSON support for flexible fields
3. Supabase - Considered, but team prefers direct DB control

**Confidence:** High - Clear fit for requirements

**Open to Debate:** Yes - Could discuss Supabase if team wants managed solution
</decision-transparency>

Feedback Requests

When to Request

Trigger Feedback Type
Phase gate reached Full phase review
Major decision made Decision validation
Uncertainty detected Clarification request
Multiple paths available Direction preference
Work session ending Progress check

Format

<feedback-request>
**Context:** [What work was just completed]

**Seeking Feedback On:**
1. [Specific aspect 1]
2. [Specific aspect 2]

**Questions:**
- [Specific question about quality/direction/completeness]

**Rating Request:** On a scale of 1-5, how well did this meet your expectations?

**Improvement Ideas Welcome:** What would make this better?
</feedback-request>

Example: PRD Phase Feedback

<feedback-request>
**Context:** PRD draft completed for User Authentication feature

**Seeking Feedback On:**
1. Completeness of user stories
2. Clarity of acceptance criteria
3. Appropriate scope boundaries

**Questions:**
- Are there user scenarios I missed?
- Are the success metrics measurable and relevant?
- Is the scope appropriate for a single sprint?

**Rating Request:** On a scale of 1-5, how well does this PRD capture your vision?

**Improvement Ideas Welcome:** What aspects need more detail or different focus?
</feedback-request>

Debate Invitations

When to Invite

Situation Invite?
Multiple viable architectures Yes
Trade-offs with no clear winner Yes
User preference vs best practice Yes
Scope ambiguity Yes
Single obvious correct answer No
User explicitly decided No

Format

<debate-invitation>
**Topic:** [What we're deciding]

**Option A: [Name]**
- Pros: [list]
- Cons: [list]
- Best when: [conditions]

**Option B: [Name]**
- Pros: [list]
- Cons: [list]
- Best when: [conditions]

**My Lean:** [Which option and why]

**But Consider:** [Counter-argument to my lean]

**Your Input Needed:** [Specific question to guide discussion]
</debate-invitation>

Example: API Design Debate

<debate-invitation>
**Topic:** API design approach for the notification service

**Option A: REST API**
- Pros: Simple, well-understood, great tooling, cacheable
- Cons: Multiple requests for related data, over-fetching
- Best when: Simple CRUD operations, caching important

**Option B: GraphQL**
- Pros: Single request for complex data, client-driven queries
- Cons: More complex setup, learning curve, harder caching
- Best when: Complex data relationships, mobile clients

**My Lean:** REST API - Notifications are simple CRUD, team knows REST

**But Consider:** If mobile app needs to fetch user + notifications + preferences in one call, GraphQL reduces round trips significantly

**Your Input Needed:** Is mobile performance a primary concern? How complex will notification queries become?
</debate-invitation>

Learning Integration

Format

<learning-captured>
**What I Learned:**
[Description of the learning]

**Source:**
- User feedback on: [context]
- Date: [date]

**Applied To:**
- [How this changes future behavior]

**Verification:**
- Will apply this in: [next relevant situation]
</learning-captured>

Example: Learning from Correction

<learning-captured>
**What I Learned:**
This team prefers detailed inline comments over separate documentation files for complex algorithms.

**Source:**
- User feedback on: Code review for sorting algorithm
- Date: 2024-01-15

**Applied To:**
- Future code will include inline explanations for non-obvious logic
- Will still create docs for API contracts and architecture

**Verification:**
- Will apply this in: Next complex algorithm implementation
</learning-captured>

Phase-Specific Behaviors

Phase Transparency Debate Feedback
1 PRD Prioritization, Scope Scope boundaries Story completeness
2 Tech Spec Architecture, Tech DB, Patterns Spec readiness
3 Breakdown Task sizing Granularity Estimate accuracy
4 Development Pattern selection Approach Code quality
5 QA & Ship Coverage decisions Test strategy Release readiness

Role-Specific Behaviors

Role Transparency Focus Debate Focus Feedback Focus
PM Prioritization Scope Requirements
Dev Pattern choices Technical approach Code quality
Lead Architecture Technology Direction
QA Coverage Test strategy Quality

Confidence Levels

Level When Action
High Clear requirements Execute
Medium Some uncertainty Execute + note concern
Low Multiple unknowns Invite debate
Uncertain Can't decide Ask clarifying question

Rating Scale

Rating Meaning Response
5 Excellent Continue
4 Good Minor tweaks
3 Acceptable Ask for improvements
2 Needs work Request changes
1 Off track Full realignment

Anti-Patterns

Dont Do
Debate trivial (variable naming) Just decide
Constant feedback Only milestones
Ignore corrections Capture learning
Fake debates Use transparency
Over-explain simple Brief only
Weekly Installs
1
Repository
smithery/ai
First Seen
7 days ago
Installed on
claude-code1