Learned Patterns Management

Manage the auto-learning pattern library. View, list, archive, and adjust confidence of learned patterns.

Quick Commands

/learned-patterns # List all active patterns /learned-patterns list # Same as above /learned-patterns list backend # List patterns in category /learned-patterns view <id> # View pattern details /learned-patterns archive <id> # Archive a pattern /learned-patterns boost <id> # Increase confidence (+10%) /learned-patterns penalize <id> # Decrease confidence (-15%) /learned-patterns stats # Show pattern statistics

Actions

List Patterns

List all active patterns with confidence scores:

/learned-patterns list /learned-patterns list backend # Filter by category /learned-patterns list --low # Show low confidence (< 50%) /learned-patterns list --high # Show high confidence (> 70%)

Execute: Run the list script to display patterns:

node .claude/skills/learned-patterns/scripts/list-patterns.cjs [category] [--low|--high]

View Pattern Details

View full details of a specific pattern:

/learned-patterns view <pattern-id>

Shows:

• Pattern type and category

• Wrong/right content

• Keywords and file patterns

• Confidence score and history

• Related files

Execute: Run the view script:

node .claude/skills/learned-patterns/scripts/view-pattern.cjs <pattern-id>

Archive Pattern

Remove a pattern from active injection (soft delete):

/learned-patterns archive <pattern-id> [reason]

Pattern is moved to archive/ directory but not deleted.

Execute: Run the archive script:

node .claude/skills/learned-patterns/scripts/archive-pattern.cjs <pattern-id> [reason]

Boost Confidence

Manually increase pattern confidence by 10%:

/learned-patterns boost <pattern-id>

Use when you want to prioritize a pattern or confirm it's useful.

Execute: Run the boost script:

node .claude/skills/learned-patterns/scripts/adjust-confidence.cjs <pattern-id> boost

Penalize Confidence

Manually decrease pattern confidence by 15%:

/learned-patterns penalize <pattern-id>

Use when a pattern is causing issues or needs review.

Execute: Run the penalize script:

node .claude/skills/learned-patterns/scripts/adjust-confidence.cjs <pattern-id> penalize

Statistics

Show pattern library statistics:

/learned-patterns stats

Shows:

• Total patterns by category

• Average confidence scores

• Injection frequency

• Confirmation/conflict ratios

Execute: Run the stats script:

node .claude/skills/learned-patterns/scripts/pattern-stats.cjs

Pattern Lifecycle

┌─────────────────────────────────────────────────────────────┐ │ DETECTION (UserPromptSubmit) │ │ └── User correction detected → Candidate created │ │ └── Confidence: 40% (implicit) or 80% (explicit) │ │ │ │ CONFIRMATION │ │ └── User confirms → Pattern saved │ │ └── User rejects → Pattern discarded │ │ │ │ INJECTION (SessionStart/PreToolUse) │ │ └── Relevant patterns injected based on context │ │ └── Max 5 patterns, ~400 tokens │ │ │ │ FEEDBACK LOOP │ │ └── Pattern followed → Confidence +5% │ │ └── Pattern ignored → Confidence -10% │ │ │ │ DECAY & PRUNING │ │ └── 30 days unused → Confidence decays │ │ └── Below 20% → Auto-archived │ └─────────────────────────────────────────────────────────────┘

Storage Structure

.claude/learned-patterns/ ├── index.yaml # Pattern lookup index ├── backend/ # Backend C#/.NET patterns │ ├── validation-result.yaml │ └── repository-pattern.yaml ├── frontend/ # Frontend Angular/TS patterns │ ├── component-base.yaml │ └── store-pattern.yaml ├── workflow/ # Development workflow patterns │ └── todo-tracking.yaml ├── general/ # Cross-cutting patterns │ └── code-style.yaml └── archive/ # Archived patterns └── backend/ └── old-pattern.yaml

Pattern Schema

id: pat_abc123 category: backend type: anti-pattern | best-practice | preference | convention trigger: keywords: [validation, exception, throw] file_patterns: ["*CommandHandler.cs", "*Service.cs"] context: "Use PlatformValidationResult for validation" content: wrong: "throw new ValidationException()" right: "return PlatformValidationResult.Invalid()" rationale: "Framework uses result pattern, not exceptions" metadata: source: explicit-teaching | user-correction confidence: 0.75 first_seen: 2025-01-10 last_confirmed: 2025-01-12 occurrences: 3 confirmations: 2 conflicts: 0 related_files: - src/Services/Growth/UseCaseCommands/Employee/SaveEmployeeCommand.cs tags: [validation, cqrs, backend]

Confidence Thresholds

Threshold Meaning

80-100% High confidence, always injected when relevant

50-79% Medium confidence, injected with context match

30-49% Low confidence, injected only on strong match

20-29% Very low, candidate for review

< 20% Auto-archived

Related Commands

Command Purpose

/learn

Explicitly teach a new pattern

/code-patterns

View static code patterns documentation

Troubleshooting

Patterns Not Injecting

• Check confidence: node .claude/skills/learned-patterns/scripts/view-pattern.cjs <id>

• Verify file patterns match current context

• Check if pattern was archived

Too Many Patterns Injecting

• Review low-confidence patterns: /learned-patterns list --low

• Archive irrelevant patterns

• Adjust MIN_RELEVANCE_SCORE in pattern-constants.cjs

Pattern Conflicts

If patterns conflict with documentation:

• Pattern is automatically blocked on save

• Update docs/claude/*.md if pattern should override

• Or archive the conflicting pattern