About GSD Skill Creator
Two frameworks, one integrated system for sustained AI-assisted development.
The Core Insight
Modern AI-assisted development with tools like Claude Code faces three fundamental challenges: context degradation over long sessions, knowledge loss between agent spawns, and inefficient token utilization. Two complementary open-source frameworks address these challenges from different angles.
get-shit-done (GSD) by glittercowboy provides the structural and workflow layer: spec-driven development, multi-agent orchestration, and persistent project state through structured markdown documents. gsd-skill-creator by Tibsfox provides the knowledge and optimization layer: pattern detection, skill creation, bounded learning, and cache-aware token management.
GSD solves the workflow problem — context rot, spec drift, and agent coordination. Skill-Creator solves the knowledge problem — pattern capture, token efficiency, and adaptive learning. Neither is sufficient alone; together they enable sustained, high-quality AI-assisted development.
At a Glance
| get-shit-done (GSD) | gsd-skill-creator | |
|---|---|---|
| Created by | glittercowboy | Tibsfox |
| Primary role | Workflow orchestration | Knowledge accumulation |
| Core problem | Context rot & spec drift | Knowledge loss & token waste |
| Key mechanism | Structured docs + multi-agent lifecycle | Pattern detection + bounded learning |
GSD: The Workflow Layer
GSD externalizes project state into structured, size-bounded markdown files (PROJECT.md, REQUIREMENTS.md, ROADMAP.md, STATE.md) that serve as persistent truth, replacing volatile conversation memory. It enforces a spec-driven lifecycle: nothing gets built until planned, nothing gets merged until verified.
| Phase | Purpose | Key Activities |
|---|---|---|
| Discuss | Scope the work | Review PROJECT.md, clarify requirements, identify constraints |
| Plan | Break into atomic tasks | Research codebase, generate PLAN.md, validate |
| Execute | Implement the plan | Follow PLAN.md, write code, run tests atomically |
| Verify | Confirm correctness | Check against requirements, commit to SUMMARY.md |
Skill-Creator: The Knowledge Layer
Skill-Creator captures recurring patterns as compact 300-500 token skills so agents arrive pre-loaded with institutional knowledge. The six-step flywheel — Observe, Detect, Suggest, Create, Refine, Compose — gets smarter with every session.
| Feature | Description |
|---|---|
| Bounded Learning | 20% max change per refinement, 7-day cooldown, 3+ corrections required |
| Token Budget | 2-5% of context allocated to skills with priority scoring |
| Cache-Aware Ordering | Deterministic load order maximizes prompt cache hits |
| Research Compression | 10-20x reduction from RESEARCH.md to compact skills |
| Agent Composition | Co-activation tracking auto-generates composite agents |
How They Work Together
The integration creates a self-reinforcing efficiency loop: GSD spawns agents with fresh context, Skill-Creator captures patterns and outcomes, research gets compressed into compact skills, deterministic ordering maximizes cache hits, and the system gets cheaper with every milestone.
| Approach | Context Per Agent | Total (4 Agents) |
|---|---|---|
| Traditional (independent) | 15K tokens each | 60K tokens |
| Agent Teams (distributed) | 6K + 2K relevant skills | 38K tokens |
| Savings | 37% reduction |
Documentation for AI & LLMs
This site follows the llms.txt standard to make documentation discoverable by AI agents and language models. Machine-readable documentation indexes are available at the site root:
- llms.txt — Structured index with section headings, page titles, and links to raw GitHub markdown
- llms-full.txt — Complete documentation contents concatenated into a single file for full-context ingestion
- For AI Agents — Dedicated landing page with quick install, key concepts, and documentation structure
Every documentation page on this site includes a source link block pointing to the corresponding raw markdown file on GitHub, so AI agents can follow links directly to machine-readable source material.
Learn More
- Bounded Learning — The guardrails that keep skills evolving safely
- Release History — Overview of all 33 versions
- Documentation Hub — Complete reference documentation
- For AI Agents — Machine-readable docs and llms.txt discovery
- Release Chronicle — Detailed narrative for each version