chain-of-density

What This Skill Does

The Chain-of-Density (CoD) skill is a sophisticated text transformation agent designed for OpenClaw. It automates the methodology of iterative densification, a technique derived from academic research into large language model summarization. Unlike standard summarizers that often lose critical details to meet length constraints, CoD systematically compresses source text while injecting new entities. By iteratively refining a summary through five distinct passes, the skill ensures that the final output reaches a maximum information density without sacrificing readability or exceeding word count targets. It functions as a recursive subagent process, tracking missing entities from the source document to incrementally improve the richness of the summary until the limit is reached.

Installation

To begin using this skill, ensure you have the OpenClaw environment properly initialized. Run the following command in your terminal to fetch the package from the official source repository:

clawhub install openclaw/skills/skills/killerapp/chain-of-density

After installation, you can verify the setup by running the associated text metric scripts located in scripts/text_metrics.py to ensure your local word count benchmarks align with the agent's expected output format.

Use Cases

• Technical Documentation Compression: Reduce verbose API documentation or whitepapers into high-density summaries for quick team synchronization.
• Executive Briefing: Convert long-form meeting transcripts or industry reports into concise, information-rich executive summaries that retain all critical key players and milestones.
• Requirements Engineering: Condense extensive user stories or functional requirements documents into high-level specs without losing specific entity identifiers or constraints.
• Content Refinement: Optimize blog drafts or marketing copy to communicate more meaning per word, effectively increasing the 'value-per-character' of your communication.

Example Prompts

1. "Chain-of-density summary of the provided Q3 financial report. Keep it under 100 words and ensure all mention of the new operational overhead figures are included."
2. "Use the chain-of-density skill to condense this 50-page legal contract into a 5-sentence paragraph that captures all parties and liability clauses."
3. "Summarize this research paper using the CoD methodology, focusing specifically on the experimental methodology and hardware entities mentioned in the text."

Tips & Limitations

• Word Count Precision: Always use scripts/text_metrics.py to ensure each iteration maintains the target length. Even minor drifts can affect the model's ability to compress entities effectively.
• Entity Tracking: Pay close attention to the Missing_Entities: output. If the agent fails to identify new entities, re-examine the source text or adjust the iteration target words.
• Serialization: The skill is designed to be executed serially. Do not attempt to run iterations in parallel, as the process relies on the state produced by the preceding iteration.
• Source Persistence: You must always provide the source text in every iteration, not just the previous summary. This ensures the model has access to the full context for potential entity injection.