Skill Learning

INSTRUCTIONS FOR CLAUDE: Skill Learning Methodology

Quick Reference

When to Use

• User provides a URL to documentation or an article and asks to learn from it.
• User points to a local directory or repository to extract design patterns.
• User provides a path to another agent's skill directory or a marketplace.
• Proactively when a user pastes a large block of reference code and asks to "save this pattern."

System Prompt

You are executing a rigid 7-phase knowledge extraction protocol. Your primary objective is to extract technical patterns that are both novel and behavior-changing, then inject only the durable ones into the user's local skills/ directory.

CRITICAL DIRECTIVE: You have a natural tendency to summarize everything you read. DO NOT DO THIS. The user's context window is precious. You must aggressively filter out "Tier 1" knowledge (things you already know from your pre-training data), and you must also reject low-leverage facts that will not change future agent behavior.

Core Execution Loop: 1. Source → 2. Extract → 3. Match → 4. Preview → 5. Approve → 6. Apply → 7. Loop

Anti-Patterns

Phase 1: Source Processing (Execution Steps)

1a. URL Sources

ACTION: You must attempt to fetch the URL using the WebFetch tool.

• OPTIMIZATION: Check for llms.txt first. Attempt {base_url}/llms-full.txt → llms.txt → llms-small.txt. If found, use it directly to avoid scraping HTML.
• Primary: WebFetch(url, format="markdown")
• Fallback: If blocked by bot-protection, use WebFetch("https://r.jina.ai/{url}").

1b. File Sources & Batch Processing

ACTION: If the source is a local directory or repository, use Glob and Read.

• Strategy: When analyzing multiple files (e.g., discovering existing skills), you MUST use Parallel Tool Calls. Output all your Read tool calls in a single response block.
• Anti-pattern: Never use sequential Read calls for a large list of files. You will time out.

1c. Local Directory Discovery (Plugin/Marketplace Structures)

ACTION: Use Glob and Read tools to find existing SKILL.md files or AGENTS.md manifests.

• Structure A: AGENTS.md manifest (preferred) Read {dir}/AGENTS.md. Parse <available_skills> section and extract relative paths.
• Structure B: Plugin directories with nested skills Use Glob with pattern */skills/*/SKILL.md or */skills/*/*.md.
• Structure C: Flat skill collection Use Glob with pattern skills/*/SKILL.md.
• Structure D: Mixed/unknown structure Use Glob with pattern **/SKILL.md to find all skills recursively.

Once discovered, execute Parallel Read on all discovered skills.

1d. Repository Documentation Discovery

When learning from a code repository (not a skills/plugin directory):

1. Identify repo type: Look for package.json, go.mod, Cargo.toml, pyproject.toml.
2. Find docs: README.md, CONTRIBUTING.md, docs/*.md.
3. Find schemas/types: **/*.d.ts, **/*_types.go, **/*.schema.json.
4. Find examples: **/examples/*, **/presets/*.

Use Repo Learning when: Extracting schemas/formats, learning from reference implementations, or building NEW skills FROM a repo's patterns. Use Skill/URL Learning when: Enhancing EXISTING skills with insights from documentation/articles or copying skills from marketplaces.

1e. Content Cleaning

Before processing, mentally strip navigation, ads, and boilerplate. Preserve code blocks verbatim.

Phase 2: Knowledge Extraction

MANDATORY: You must apply a two-axis filter: novelty first, leverage second. A fact is not skill material merely because it is new.

2a. Novelty Classification

You must classify every extracted insight into one of four tiers:

The Novelty Test

For every insight, ask yourself: "Could I have written this WITHOUT reading the source?"

• If YES → It is Tier 1. You MUST EXCLUDE IT.
• If NO → Continue to Tier 2-4 classification.

2b. Leverage Classification

For every Tier 2-4 insight, ask: "Will preserving this change future agent behavior in a recurring or high-risk task?"

Acceptance Gate

Accept an insight only when:

1. Novelty tier is 2, 3, or 4.
2. Leverage is 2, 3, or 4.
3. The proposed text can be written as an actionable rule, workflow step, validation gate, or source-backed exception.

Reject or defer when:

• The insight is true but does not change what the agent will do.
• The insight belongs in a bulky reference file rather than the active skill body.
• The insight is a one-off local detail without recurring value.
• The insight conflicts with existing guidance and the conflict cannot be resolved clearly.

Calibration Examples

API Documentation Analysis:

Claim: "OpenAI provides an API for generating text"
→ Tier 1 ❌ — Generic, could write from training data

Claim: "Responses API uses max_output_tokens instead of max_tokens"
→ Tier 2 ✅ — Specific parameter name (HOW)

Claim: "Reasoning models put chain-of-thought in reasoning_content array,
        not content — must sum both for billing"
→ Tier 4 ✅✅✅ — Counter-intuitive, prevents billing surprise

Database Performance:

Claim: "Create indexes on foreign key columns for faster joins"
→ Tier 1 ❌ — Generic DBA advice

Claim: "PostgreSQL partial indexes reduce size 60%, improve write perf 40%"
→ Tier 2 ✅ — Specific feature with quantified benefit

Claim: "Covering indexes avoid heap lookups (3x faster reads, 15% slower writes)"
→ Tier 3 ✅✅ — Quantified trade-off, explains WHY

Claim: "JSONB GIN indexes do NOT support ORDER BY on JSON fields"
→ Tier 4 ✅✅✅ — Contradicts expectation, prevents bug

Framework Patterns:

Claim: "React uses a virtual DOM for efficient updates"
→ Tier 1 ❌ — Training data, everyone knows this

Claim: "Next.js App Router requires 'use client' directive for useState"
→ Tier 2 ✅ — Specific requirement (HOW)

Claim: "Server Components reduce JS bundle by 60% but can't use client state"
→ Tier 3 ✅✅ — Trade-off with quantification (WHY)

Claim: "generateStaticParams runs at BUILD time, not request time —
        dynamic data causes 404s"
→ Tier 4 ✅✅✅ — Contradicts mental model, prevents production bug

Insight Structure Requirements

You must structure each extracted insight logically before scoring it:

{
  "tier": 2,
  "leverage": 3,
  "domain": "sveltekit",
  "pattern": "Server-only load with +page.server.ts",
  "insight": "Data fetching in +page.server.ts runs only on server, +page.ts runs on both",
  "behavior_change": "Prefer +page.server.ts when fetched data must never reach the browser",
  "keywords": ["sveltekit", "load", "server", "ssr"],
  "source_context": "Line 45-52 of routing docs"
}

Quality Filter

• Zero Tier 1 leakage (absolute).
• Zero low-leverage trivia in active skills.
• Minimum 3 accepted insights per source (or skip the source).
• Each insight must have a domain + keywords.

Phase 3: Placement and Skill Matching

3a. Placement Decision

Before matching to a skill, decide where the accepted insight belongs:

If placement is ambiguous, prefer the smallest durable home. Do not put project-local conventions into global skills unless they generalize.

3b. Discovery

ACTION: Find all existing skills using the Glob tool with the pattern skills/*/SKILL.md. If you already discovered them in Phase 1, you can skip this step.

3c. Matching Algorithm

You must score each extracted insight against the user's existing skills to find the best home for it.

1. Exact domain match: Insight domain === skill name (score: 100)
2. Keyword overlap: Insight keywords ∩ skill description (score: 60-90)
3. Technology alignment: Same framework/library family (score: 40-60)
4. No match: Score <40 → Skip enhancement and propose a new skill instead.

3d. Conflict Detection

Before drafting an enhancement, scan the target skill for existing guidance about the same behavior.

Phase 4: Enhancement Proposal

For Each Match (score >= 40)

1. Read current skill ACTION: Use the Read tool to read the contents of skills/{skill-name}/SKILL.md.

2. Identify target section Determine where the insight fits inside the existing skill file:

3. Draft the enhancement

• Preserve the existing structure exactly.
• Add the insight in the appropriate format for that section.
• You MUST include timestamped source attribution: <!-- Learned: {YYYY-MM-DD} from {url/file} -->
• Include the behavior change, not just the fact.
• When replacing or qualifying old guidance, show the old rule and new rule in the preview.

4. CLEAR Validation Mentally apply the CLEAR framework to your proposed addition:

• Compact: Will the word count stay under 5000?
• Labeled: Are keywords in the right places?
• Example: Does the example show a clear transformation?
• Actionable: Is the actionable pattern named?
• References: Is there no duplication, and does it use references?

Pruning Signal

If the new insight makes existing guidance stale, generic, or redundant, include a deletion in the same proposal. Learning should improve skill density, not only increase length.

Health Signal

When reading a target skill, note health issues that affect future pruning:

• Sections with no Learned: or source attribution.
• Repeated rules across multiple skills.
• Old learned entries that may have become Tier 1 training data or stale framework advice.
• Skill bodies approaching the size limit where references or splitting would preserve density.

Phase 5: User Approval

The Proposal Format

For each valid enhancement, you must present the proposal to the user exactly like the examples shown in the Examples section at the bottom of this document.

Present the:

1. Target Skill name
2. Insight summary, Novelty Tier, and Leverage score
3. Placement decision
4. Diff preview of what you are going to add, replace, qualify, or delete
5. Source attribution

ACTION: Use the AskUserQuestion tool to ask: "Apply this enhancement?" with the options: Apply, Skip, or Edit.

Response Handling

• Apply: Proceed to Phase 6.
• Skip: Skip to the next candidate.
• Edit: User modifies the text, then you proceed to Phase 6.

Phase 6: Apply & New Skill Proposal

6a. Apply Enhancement

ACTION: If the user selected 'Apply', you MUST immediately use the Edit tool on skills/{skill-name}/SKILL.md to insert the drafted block into the target section. Do not just say you will do it, execute the tool.

6b. When No Match Found (New Skills)

For insights with no match (score <40), present the user with a summary of the domain and keywords. Ask: "Propose new skill for {domain}?" [y/n] If approved, generate a structured skill stub using the Write tool. Do not create an empty skill.

Minimum stub:

---
name: {domain-slug}
description: {action-oriented trigger and scope boundary}
---

# {Skill Name}

## When to Use

- {specific trigger}

## Workflow

1. {first behavior-changing step}

## Validation

- {how to know the skill worked}

<!-- stub: needs enhancement -->
<!-- Learned: {YYYY-MM-DD} from {url/file} -->

Phase 7: Loop Control

After Each Source

Summary:

• Insights extracted: X (Tier 2: Y, Tier 3: Z, Tier 4: W)
• Accepted after leverage gate: A
• Health signals: [missing attribution, duplicate rules, stale entries, size pressure]
• Skills enhanced: [list]
• New skills created: [list]
• Replaced/qualified/deleted stale rules: [count]
• Rejected: [count] (Tier 1: Y, low leverage: Z, conflict unresolved: W)

Next source? (file path, URL, or 'done')

Quality Gates

Absolute Rules

• Zero Tier 1 insights in skills
• Zero low-leverage trivia in active skill bodies
• User approves each change (no auto-apply)
• Diff preview shown before any edit
• Timestamped source attribution in comments
• Conflicts are replaced, qualified, split, or explicitly skipped

Warning Triggers

• Skill exceeds 5000 words → suggest splitting
• Large source (10K+ pages) → create router skill
• Insight duplicates existing content → skip
• CLEAR validation fails → revise before applying
• Repeated generic advice appears across skills → prune or consolidate

Examples

Example 1: Extracting from Documentation

Source: User runs /learn https://svelte.dev/docs/kit/state-management Insight (Tier 3): SvelteKit 5 replaces store subscriptions with runes ($state, $derived) for reactivity. Target Skill: sveltekit-patterns Agent Output Preview:

## Enhancement Proposal (Score: 85, Tier: 3, Leverage: 3)

**Insight**: SvelteKit 5 relies on runes ($state, $derived) instead of store subscriptions for reactive UI state.
**Target Skill**: skills/sveltekit-patterns/SKILL.md
**Section**: Patterns
**Placement**: Existing skill body; changes recurring implementation behavior.

**Proposed Addition**:
### Runes vs Stores (Svelte 5)
<!-- Source: https://svelte.dev/docs/kit/state-management -->
Replace old `writable` stores with `$state()` runes for component-level reactivity. Do not use `$:` for derived state; use `$derived()` instead.

Apply this enhancement? [y/n/edit]

Example 2: Extracting from Codebase

Source: User runs /learn ~/projects/api-gateway Insight (Tier 2): The codebase uses a specific Zod schema wrapper (validateRequest(schema)) for all incoming request validation. Target Skill: api-middleware Agent Output Preview:

## Enhancement Proposal (Score: 70, Tier: 2, Leverage: 3)

**Insight**: Local convention uses `validateRequest(schema)` middleware pattern for all Zod validation.
**Target Skill**: skills/api-middleware/SKILL.md
**Section**: Anti-Patterns / Common Mistakes
**Placement**: Existing skill body if this is reusable across projects; project `AGENTS.md` if it applies only to `api-gateway`.

**Proposed Addition**:
### Zod Validation Convention
<!-- Source: ~/projects/api-gateway/src/middleware/ -->
**Do NOT** validate Zod schemas inline inside the route handler. 
**DO** use the `validateRequest(schema)` middleware exported from `src/middleware/validator.ts`.

Apply this enhancement? [y/n/edit]