Documentation > Architecture > Developer Guide

Developer Guide

Extension points, architecture deep-dives, and contribution guidelines for developers building on or extending GSD Skill Creator. This guide covers the internal pipeline architecture, the Amiga-chipset agent orchestration model, custom topology integration, and the full source tree navigation. Whether you are adding a new skill validator, implementing a custom MCP tool integration, or contributing to the core skill loading pipeline, this guide maps the architecture and shows you where to plug in.

Source & Code Navigation

• Source Code Browser — Navigate the full source tree with annotated module descriptions, cross-references between capabilities and their implementations, and architectural decision records linked to the code they influenced. Start here for any implementation-level investigation.
• Skill Format Guide — The specification for skill YAML format, covering every field and validation rule. Essential reading before implementing a custom skill validator or extending the skill loading pipeline.

Integration Tutorials

• CI/CD Integration Tutorial — Step-by-step guide for embedding skill-creator into continuous integration pipelines. Covers the skill-creator verify hook for pre-commit validation, automated pattern detection in CI context, and skill promotion workflows triggered by test suite results.

Pipeline Architecture

The 6-stage skill loading pipeline (Score → Resolve → ModelFilter → CacheOrder → Budget → Load) is the central architectural element governing how skills are selected, filtered, and loaded into context. The full specification, including the scoring formula, resolver logic, and budget enforcement algorithm, is documented in the capability page:

• CAP-016: Skill Loading Pipeline — The 6-stage pipeline architecture. Documents stage interfaces, configuration parameters for each stage, ModelFilter compatibility rules, CacheOrder eviction policy, and the token budget enforcement algorithm. The authoritative reference for understanding skill selection behavior.

Agent Orchestration & Chipset Model

GSD Skill Creator uses an Amiga chipset-inspired agent orchestration model where specialized agents (Agnus=memory-coordinator, Denise=display-engine, Paula=io-manager, Gary=glue-logic) collaborate through a 32-bit signal format with source/target/type/payload fields. The architecture is documented across two capability pages:

• CAP-052: Team-as-Chip Framework — The Amiga chipset model for agent composition. Covers chip role assignments, signal bus protocol, crew formation patterns, and the copper-list effect for visual output coordination. Foundation for any multi-agent integration work.
• CAP-053: Exec Kernel — Execution scheduling and task prioritization within the chipset model. Documents the exec.library API analogs, priority queue management, interrupt handling for time-critical skill activations, and the message port system for inter-agent communication.

Extension Points

Custom integrations can hook into skill-creator at three levels: (1) Skill validators — add custom YAML validation logic to the Score stage; (2) Pipeline stages — implement custom Resolve or Budget stage overrides; (3) MCP tools — register new tool definitions for the agent orchestration layer. Each extension point is documented in the Source Code Browser with interface definitions and integration examples.