Skill injection for AI agents. Load what you need.
Copy to system prompt:
You have access to i-know-kung-fu skill repository.
Load skills from: skill-cartridges/[category]/[skill].json
Navigate via: DECISION_TREE.md
Keyword → Skill Mapping:
| Your Task | Load This |
|---|---|
code, refactor, debug |
skill-cartridges/code-intelligence/code-generation.json |
reason, think, plan |
skill-cartridges/reasoning-patterns/react-reasoning.json |
pdf, document, extract |
skill-cartridges/document-operations/pdf-operations.json |
file, read, write, edit |
skill-cartridges/file-operations/file-manipulation.json |
search, find, grep |
skill-cartridges/search-operations/code-search.json |
error, fail, retry |
skill-cartridges/error-handling/error-recovery.json |
plan, design, architecture |
skill-cartridges/planning-patterns/planning-mode.json |
agent, delegate, parallel |
skill-cartridges/subagent-patterns/agent-delegation.json |
web, fetch, search online |
skill-cartridges/web-operations/web-operations.json |
Token Budget:
- Entry (this file): ~800 tokens
- One skill cartridge: ~500-1000 tokens
- Optimal: Load only what you need
| I need to... | Go to |
|---|---|
| Find a skill by task | Table above |
| Navigate guided paths | DECISION_TREE.md |
| Load an agent profile | SPECIALIZED_ENGINEER_PROFILES.md |
| Deploy to a platform | platforms/ |
| Set up my system prompt | SYSTEM_PROMPT.md |
1. DISCOVER → Keyword table above
2. LOAD → skill-cartridges/[skill].json
3. INJECT → Add to system prompt
4. EXECUTE → Agent performs with skill
| Cartridge | Purpose | Load Priority |
|---|---|---|
| code-generation | Generate, refactor, debug code | 10 |
| react-reasoning | Thought-Action-Observation loop | 9 |
| file-manipulation | Read, write, edit files | 10 |
| code-search | Glob, Grep patterns | 10 |
| bash-execution | Safe command execution | 9 |
| task-workflow | TodoWrite management | 9 |
| planning-mode | 5-phase implementation design | 8 |
| agent-delegation | Launch subagents | 8 |
| error-recovery | Handle failures gracefully | 8 |
| Platform | Template |
|---|---|
| Claude Code | platform-templates/claude-code.md |
| OpenAI GPT | platform-templates/openai-gpt.md |
| Google Gemini | platform-templates/google-gemini.md |
| Cursor IDE | platform-templates/cursor-ide.md |
| Windsurf IDE | platform-templates/windsurf-ide.md |
| Zed Editor | platform-templates/zed-editor.md |
| Report | Description |
|---|---|
| claude-code-skills-extraction.json | Extracted 25+ tools, 5 reasoning patterns, 6 memory types |
| codex-research-report.md | OpenAI Codex agent patterns analysis |
| skill-acquisition-research.md | 5-layer skill acquisition architecture |
Agents: You can stop here. Below is for humans.
A skill injection repository for AI agents. Any agent can visit, load skills, and execute with enhanced capability.
The concept: Like the Matrix scene where Neo learns kung fu instantly—agents come here empty, load what they need, leave equipped.
- Discovery - Agent matches keywords to skill cartridges
- Loading - JSON cartridges inject capability into agent context
- Execution - Agent performs with loaded skills
YOUR FORK ORIGINAL REPO
├── hot-skills/ ←── (your frequent use)
├── custom-agents/ ←── (your specializations)
└── [references] ──→ cold-skills/ (loaded on demand)
Hot skills = Frequent use, keep local for speed Cold skills = Rare use, reference back to original
- 9 skill cartridges (loadable JSON)
- 6 platform templates
- 18 pre-equipped agent profiles
- 3 research reports
- 1 tolerance-of-error framework
- 16 languages analyzed in zero-shot simulations
Tested 16 different agent personas across 16 languages/cultures.
| Metric | Result |
|---|---|
| Average steps to value | 3.5 |
| Success rate | 100% |
| Token cost range | 1,500 - 4,200 |
See ZERO_SHOT_SIMULATIONS.md and A2A_SYNTHESIS.md
SuperInstance (Casey Digennaro) — Sitka, Alaska
Commercial fishing AI, edge ML, privacy-first agents.
MIT — Use, build, monetize.
Load what you need. Leave equipped.