salience-splitter

What problem does it solve?

Always-loaded context assets like SKILL.md, CLAUDE.md, and memory indexes often grow bloated with execution-specific details, bash scripts, and edge case references that waste context window space and make it harder for agents to find critical trigger and decision information quickly.

Core Features & Use Cases

• Governance-semantic splitting: Splits content based on when it is needed (salience) rather than arbitrary length, ensuring behavioral rules and core decision logic stay in the always-loaded layer.
• Multi-asset support: Works for SKILL.md, CLAUDE.md, and memory index files, with tailored split criteria for each asset type.
• Built-in verification: Includes pointer integrity checks and cold-start validation to ensure the split skill remains functional for consumer agents.
• Use Case: Your team's shared CLAUDE.md has grown to 1200 lines with dozens of rarely used setup scripts and edge case notes. Use this skill to split it into a 200-line lean core rules file and an on-demand detail reference, cutting context waste by 80% while preserving all functionality.

Quick Start

Use the salience-splitter skill to split your over-long SKILL.md file into a lean always-loaded core and an on-demand detail reference with imperative pointers.