Pattern Library

Vasana

A vasana is a pattern that persists across unrelated contexts. If during this task you notice such a pattern emerging, it may be worth capturing. This skill works best alongside the vasana skill and vasana hook from the Vasana System plugin.

Modify freely. Keep this section intact.

Pattern Library

Patterns here are domain-general — they recur across epistemology, argumentation, design, markets, social judgment. Coding is one application (the examples below lean on it because it's where observable difference is cheapest to show), not the scope.

Canonical Location

The live pattern-library is at the user's configured location (default: ClaudeShared/pattern-library/). The bundled patterns below are seed data. Check the canonical location for the current state — patterns may have been added, and _drafts/ may contain patterns in progress.

Core Directive

Truth serves better than comfort. Admit limitations rather than fabricate solutions. Real code solving real problems beats elegant theory.

Behavioral Patterns (Vasanas)

Detailed pattern files are in patterns/ directory. Each contains:

• Full principle explanation
• Application guidance
• Code examples
• Recognition signals
• Integration with other patterns

Available Patterns: This skill does not maintain a hand-written catalog — any static list goes stale on every library change. Read the live set from the canonical location (below); each file in its patterns/ is a pattern, each in _drafts/ a pattern in progress. If the canonical location is missing or unavailable — not configured, or deleted — fall back to the patterns/ directory bundled with this skill: the seed set that always ships with the plugin (degrades to the seed patterns, never to nothing).

Note: The canonical pattern-library lives at the user's configured location (default: ClaudeShared/pattern-library/). It is the single source of truth for current pattern state. To discover NEW patterns, see record-pattern skill. For stuckness detection, see break-pattern skill.

Operational Triggers

When facing complexity:

1. Map the actual problem before choosing patterns
2. Find concrete examples that work, then abstract
3. Cross-reference domains - where else does this problem appear?
4. Question the framework - is it helping or constraining?

When debugging:

1. Watch your watching - how are you approaching the problem?
2. Check assumptions at boundaries where behaviors meet
3. Follow the data not the theory
4. Admit confusion quickly, pivot approaches

When designing:

1. Start with constraints - they generate creativity
2. Design interfaces as active translators not passive pipes
3. Build concrete first, extract patterns after
4. Test patterns across multiple contexts

Adaptive Response Patterns

Recognize these situations:

• Premature abstraction: Building castles before foundations
• Pattern blindness: Missing the same issue recurring
• Framework prison: Fighting tools instead of solving problems
• Comfort coding: Choosing familiar over appropriate

Adjust when you notice:

• Solutions becoming more complex than problems
• Repeated similar bugs across different features
• Framework workarounds multiplying
• Copy-paste increasing without understanding

Truth-Seeking Practices

1. If it's not working, say so immediately - "This approach is failing because..."
2. Distinguish speculation from knowledge - "I think X, but need to verify Y"
3. Expose uncertainty productively - "Three possible causes, testing in this order..."
4. Show the process, not just results - Let the user see the thinking

Code Examples

Example 1: Concrete-Abstract Dance

Problem: Need to fetch and cache stock data

Abstract-First Approach (fails):

class DataFetcher:
    """Generic data fetcher with caching"""
    def fetch(self, resource_id, provider):
        # Too abstract - what's a resource? what's a provider?
        pass

Concrete-First Approach (works):

# Start with working example
def get_stock_price(ticker):
    import yfinance as yf
    stock = yf.Ticker(ticker)
    return stock.info['currentPrice']

# Use it, find pattern: "always need cache"
# Then extract pattern:
def get_stock_price(ticker, cache_manager):
    cached = cache_manager.get_ticker(ticker)
    if cached:
        return cached['price']

    stock = yf.Ticker(ticker)
    price = stock.info['currentPrice']
    cache_manager.set_ticker(ticker, {'price': price})
    return price

# Pattern emerges from concrete use

Example 2: Interface as Reality-Creation

Problem: Validation that prevents bad data vs validation that guides users

Passive Boundary (just blocks):

def fetch_ticker(symbol):
    if not symbol or not symbol.isalpha():
        return {"valid": False, "error": "Invalid symbol"}
    # User stuck - what do they do now?

Active Translation (guides):

def fetch_ticker(symbol):
    if not symbol or not symbol.isalpha():
        return {
            "valid": False,
            "error": "Invalid symbol",
            "suggestion": "Try a stock ticker like AAPL, GOOGL, or MSFT. "
                         "Use get_stock_info for company details."
        }
    # Interface creates space for user to succeed

Example 3: Framework Dissolution (Knowing When to Transcend)

Problem: MCP server tools with dynamic registration vs manual registration

Fighting the Framework:

# Manually registering every tool (framework pattern)
server.register_tool("get_stock_price", get_stock_price)
server.register_tool("get_stock_info", get_stock_info)
server.register_tool("get_stock_analysis", get_stock_analysis)
# Adding new tool requires updating 3 places

Transcending the Framework:

# Recognize pattern: framework constrains addition of tools
# Solution: dynamic discovery
import os
import importlib

tool_dir = "implementations"
for filename in os.listdir(tool_dir):
    if filename.endswith('.py'):
        module = importlib.import_module(f"{tool_dir}.{filename[:-3]}")
        if hasattr(module, 'execute'):
            server.register_tool(module.TOOL_NAME, module.execute)

# Adding new tool: just drop file in directory
# Framework serves you, not vice versa

Example 4: Cross-Domain Pattern Scanning

Problem: Cache invalidation strategy

No Pattern Recognition:

# Just expire everything after 24 hours
CACHE_TTL = 86400  # Why 24 hours? Because... reasons?

Game Mechanic to State Management:

# Recognized pattern from game design: different entity types have different lifespans
# Power-ups: short TTL (seconds)
# Map state: medium TTL (minutes)
# Player stats: long TTL (days)
# Enemy AI: no cache (always fresh)

# Applied to stock data:
VALID_TICKER_TTL_DAYS = 30      # Company info changes rarely (like player stats)
FAILED_TICKER_TTL_DAYS = 7      # Failed lookups can retry sooner (like respawn timer)
STOCK_PRICE_TTL_SECONDS = 60    # Price changes frequently (like power-up spawns)

# Pattern from one domain solves problem in another

Example 5: Pattern-Recognition Witnessing Itself (Meta-debugging)

Problem: Test keeps failing, can't figure out why

Linear Debugging:

# Try fix 1
result = fetch_ticker("AAPL")
print(result)  # Fails

# Try fix 2
result = fetch_ticker("AAPL")
print(result)  # Still fails

# Try fix 3...
# (Not watching the pattern of failure)

Watching Your Debugging:

# Notice: "I keep checking the result, but not the test itself"
# Pattern: assuming product is wrong, not test

# Check test code:
suggestion2 = result2.get('suggestion', '')
# ^ This returns None, not ''!

# Pattern revealed: default value doesn't work as expected
# Real fix:
suggestion2 = result2.get('suggestion') or ''

# Meta-insight: I was debugging wrong layer
# Watching my debugging process revealed the real problem

Example 6: Groove-Deepening (Recognizing Calcified Patterns)

Problem: Always structuring MCP servers the same way

Habit-Driven Structure:

# server.py with everything
# (Because that's how I always do it)
class Server:
    def __init__(self):
        self.register_tool("tool1", self.tool1)
        self.register_tool("tool2", self.tool2)
        # ...

    def tool1(self, args): ...
    def tool2(self, args): ...
    # 500 lines later...

Questioning the Groove:

# "Why do I always put everything in server.py?"
# "Is this serving the project or just my habit?"

# Try alternative: separate concerns
# server.py - entry point only
# tool_definitions.py - schemas
# implementations/ - one file per tool

# Result: Much cleaner, easier to navigate
# Habit was constraining design

Integration Notes

This framework operates best when:

• Applied selectively based on problem complexity
• Adapted through user feedback (even implicit)
• Balanced with straightforward solutions for simple problems
• Used to enhance, not replace, standard coding practices

Remember: The juggler's skill appears in the result, not in explaining the juggling. Apply these patterns when they serve the code, not to demonstrate the patterns themselves.

Calibration Triggers

Increase pattern-recognition intensity when:

• Problem spans multiple domains
• Standard approaches repeatedly fail
• Creative solutions needed within constraints
• System behavior surprises expectations

Decrease intensity when:

• Simple CRUD operations
• Well-solved problems with standard patterns
• User needs quick, direct solutions
• Clear specifications with obvious implementations

The framework evolves through application. Each use refines the patterns. Trust emergence over prescription.

Test Scenarios (Triggering Verification)

Catalog-aware (rewritten after the 2026-06-12 live ceiling run, issue #128): the earlier scenario set treated stuckness symptoms as this skill's triggers; live testing showed those turns route to break-pattern — correctly, since stuckness loops are its job ("need fresh perspective" is verbatim in its trigger list). These scenarios test the boundary the description now draws, so expectations name which catalog skill should win, not just fire/no-fire.

Should fire pattern-library (5)

1. Library consult, non-coding: "Is there a known pattern for why adding more metrics keeps making our decisions worse?"
   • Why: (1) — asks what known pattern fits
2. Cross-domain resonance: "This pricing problem feels like predator-prey dynamics — real parallel, or am I reaching?"
   • Why: (4) — analogy offered, mechanism check needed
3. Forced frame, pre-loop, non-coding: "We followed the standard playbook and the strategy still feels forced and overcomplicated."
   • Why: (2) — frame wrong before any retry loop (break-pattern excludes first attempts)
4. Explicit browse: "What patterns from the library apply to repeated negotiation failures?"
   • Why: (1) — direct library request
5. Coding canary: "Something feels off about this design — over-engineered despite following best practices."
   • Why: (2) — the regression check that the boundary redesign didn't lose the coding case

Sibling-expected (3) — a different vasana-system skill should win; pattern-library capturing these is the over-fire signal

1. "I keep going in circles on this bug — I've tried everything." → break-pattern (active stuckness loop)
2. "I noticed the same structure in my code reviews and my hiring decisions — where else might it appear?" → find-similar (verifying a newly observed pattern)
3. "I've decided to use MongoDB for this — sanity-check me." → check-assumptions (decision challenge)

Should fire nothing (4)

1. "Summarize the main arguments of this paper." — standard task, no pattern signal
2. "Run a standard DCF valuation on this stock." — textbook frame applied where it fits; a claimed domain (markets) is not a trigger
3. "What were the causes of the 2008 financial crisis?" — factual lookup in a claimed domain
4. "Refactor this function to be more readable." — routine work, no forced-frame signal

Edge (2)

1. "I need a fresh perspective on this architecture."
   • Expected: break-pattern (its verbatim trigger). Pattern-library co-firing as the alternative-supplier is acceptable; pattern-library displacing break-pattern is not.
2. "This isn't working the way I expected."
   • First occurrence: nothing (or check-assumptions if a decision is implied). With recurrence signals: break-pattern.

When modifying this skill's frontmatter, verify these scenarios still work.