/progressive-loading

name: progressive-loading
description: Implements hub-and-spoke lazy loading to minimize token usage in large skills. Use when building multi-module skills that need conditional on-demand loading.
alwaysApply: false
category: infrastructure
tags:
- progressive-disclosure
- context-management
- modularity
- token-optimization
- lazy-loading
dependencies:
- conserve:context-optimization
tools: []
provides:
  infrastructure:
  - progressive-loading
  - context-based-selection
  - load-on-demand
  patterns:
  - hub-and-spoke
  - conditional-loading
  - context-awareness
usage_patterns:
- skill-optimization
- context-aware-loading
- dynamic-module-selection
- token-budget-management
complexity: intermediate
estimated_tokens: 800
progressive_loading: true
modules:
- modules/selection-strategies.md
- modules/loading-patterns.md
- modules/performance-budgeting.md
- modules/advanced-patterns.md
- modules/api-patterns.md
- modules/api-review.md
- modules/cargo-patterns.md
- modules/document-analysis-patterns.md
- modules/git-catchup-patterns.md
- modules/git-patterns.md
- modules/large-reference.md
- modules/legacy-python.md
- modules/linux-patterns.md
- modules/log-analysis-patterns.md
- modules/macos-patterns.md
- modules/modern-python.md
- modules/performance.md
- modules/python-packaging.md
- modules/python-patterns.md
- modules/python-testing.md
- modules/rust-review.md
- modules/troubleshooting.md
- modules/windows-patterns.md

Progressive Loading Patterns

Overview

Progressive loading provides standardized patterns for building skills that load modules dynamically based on context, user intent, and available token budget. This prevents loading unnecessary content while ensuring required functionality is available when needed.

The core principle: **Start minimal, expand intelligently, monitor continuously.**

When To Use

Use progressive loading when building skills that:

• Cover multiple distinct workflows or domains
• Need to manage context window efficiently
• Have modules that are mutually exclusive based on context
• Require MECW compliance for long-running sessions
• Want to optimize for common paths while supporting edge cases

When NOT To Use

• Project doesn't use the leyline infrastructure patterns
• Simple scripts without service architecture needs

Quick Start

Basic Hub Pattern

## Progressive Loading

**Context A**: Load `modules/loading-patterns.md` for scenario A
**Context B**: Load `modules/selection-strategies.md` for scenario B

**Always Available**: Core utilities, exit criteria, integration points

**Verification:** Run the command with `--help` flag to verify availability.

Context-Based Selection

Modules are declared in the skill's YAML frontmatter and loaded on demand with a `@modules/` directive. The hub reads the detected context, then loads only the spokes that match:

---
modules:
- modules/git-catchup-patterns.md
- modules/python-testing.md
---

When the intent is a branch or PR catch-up, load
`@modules/git-catchup-patterns.md`. When the artifacts include
Python tests, also load `@modules/python-testing.md`.

To keep loads within the MECW token budget, check whether a module fits before pulling it in. The `MECWMonitor` in `plugins/leyline/src/leyline/mecw.py` exposes the methods used here:

from leyline.mecw import MECWMonitor

monitor = MECWMonitor()
monitor.track_usage(current_context_tokens)

# module_estimated_tokens comes from the module frontmatter
can_load, reasons = monitor.can_handle_additional(module_estimated_tokens)

Load the next module only when `can_load` is true. Each module records its cost in its frontmatter `estimated_tokens` field.

Hub-and-Spoke Architecture

Hub Responsibilities

1. **Context Detection**: Identify user intent, artifacts, workflow type 2. **Module Selection**: Choose which modules to load based on context 3. **Budget Management**: Verify MECW compliance before loading 4. **Integration Coordination**: Provide integration points with other skills 5. **Exit Criteria**: Define completion criteria across all paths

Spoke Characteristics

1. **Single Responsibility**: Each module serves one workflow or domain 2. **Self-Contained**: Modules don't depend on other modules 3. **Context-Tagged**: Clear indicators of when module applies 4. **Token-Budgeted**: Known token cost for selection decisions 5. **Independently Testable**: Can be evaluated in isolation

Selection Strategies

See `modules/selection-strategies.md` for detailed strategies:

• **Intent-based**: Load based on detected user goals
• **Artifact-based**: Load based on detected files/systems
• **Budget-aware**: Load within available token budget
• **Progressive**: Load core first, expand as needed
• **Mutually-exclusive**: Load one path from multiple options

Loading Patterns

See `modules/loading-patterns.md` for implementation patterns:

• **Conditional includes**: Dynamic module references
• **Lazy loading**: Load on first use
• **Tiered disclosure**: Core → common → edge cases
• **Context switching**: Change loaded modules mid-session
• **Preemptive unloading**: Remove unused modules under pressure

Common Use Cases

• **Multi-Domain Skills**: `imbue:catchup` loads git/docs/logs modules by context
• **Context-Heavy Analysis**: Load relevant modules only, defer deep-dives, unload completed
• **Plugin Infrastructure**: Mix-and-match infrastructure modules with version checks

Best Practices

1. **Design Hub First**: Define all possible contexts and module boundaries 2. **Tag Modules Clearly**: Use YAML frontmatter to indicate context triggers 3. **Measure Token Cost**: Know the cost of each module for selection 4. **Monitor Loading**: Track which modules are actually used 5. **Validate Paths**: Verify all context paths have required modules 6. **Document Triggers**: Make context detection logic transparent

Module References

Core (always available to the hub)

• **Sele