dansasser

Context Arbitration Layer — Development skill for Claude Code

Development community

Context Arbitration Layer (CAL) extends claude-code with persistent, multi-subprocess context management.

How to install Context Arbitration Layer

This entry records only its repository, not the path inside it, so there is no exact command to give. Open dansasser/context-arbitration-layer and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Context Arbitration Layer does

Context Arbitration Layer (CAL) extends claude-code with persistent, multi-subprocess context management. It offloads heavy code analysis to external executors like ollama-code, gemini-cli, or forgecode, maintaining coherence across large projects while reducing token load.

Alternatives in Development

MCP Builder — Guide for creating high-quality MCP servers to integrate external APIs and services 94.1k ★
Claude Token Efficient — One CLAUDE.md file 3.8k ★
Claw Code Agent — Claw Code No Rust No TypeScript Only Python 543 ★

README

🧠 Context Arbitration Layer (CAL)

**Version:** 1.3 **Status:** Specification & Skill Implementation **Author:** Dan Sasser / Gorombo Labs

CAL extends `Claude Code` into a multi-process reasoning system that orchestrates subprocess executors (like `ollama-code`, `claude -p`, `claude --agent`) while maintaining persistent context through a memory ledger.

**Proven Results:** 5.5× token efficiency | 92.25% context retention | 69,500 tokens saved per session

The Problem CAL Solves

Large codebases and deep reasoning chains push `Claude Code` beyond its context window limits. Even with compression, you face:

Problem	Cause	Impact
Context Loss	Finite context window	Re-analyzing same files
Token Waste	Redundant file ingestion	Inefficient resource use
Semantic Drift	Truncated memory	Inconsistent reasoning
Session Amnesia	No persistent state	Lost work between sessions

**CAL's Solution:** Delegate heavy analysis to subprocess executors, persist results in a ledger, reuse context across sessions.

How CAL Works

Developer Request
   ↓
Claude Code (Controller)
   ├─ Analyzes task
   ├─ Decides: Sequential? Concurrent? Staggered?
   └─ Spawns ollama-code subprocess(es)
       ↓
   Executors analyze in parallel
       ↓
   Return structured JSON summaries
       ↓
Claude Code validates checksums
       ↓
   Commits to .cal_memory.json ledger
       ↓
Reintegrates summaries into context
       ↓
Developer receives result with preserved context

**Three Core Components:**

Controller (Claude Code) - Orchestrates tasks, validates outputs, manages ledger
Executors (ollama-code, claude -p, claude --agent, gemini-cli, forgecode) - Perform heavy analysis