chain-of-densityIteratively densify text summaries using Chain-of-Density technique. Use when compressing verbose documentation, condensing requirements, or creating executive summaries while preserving information density.
Install via ClawdBot CLI:
clawdbot install killerapp/chain-of-densityCompress text through iterative entity injection following the CoD paper methodology. Each pass identifies missing entities from the source and incorporates them while maintaining identical length.
Chain-of-Density works through multiple iterations:
target_words)Key principle: Never drop entities - only add and compress.
Each entity added must meet ALL 5 criteria:
| Criterion | Description |
|-----------|-------------|
| Relevant | To the main story/topic |
| Specific | Descriptive yet concise (≤5 words) |
| Novel | Not in the previous summary |
| Faithful | Present in the source (no hallucination) |
| Anywhere | Can be from anywhere in the source |
cod-iteration agentMissing_Entities: lineIteration 1: Sparse base (target_words, verbose filler)
↓ Missing_Entities: (none - establishing base)
Iteration 2: +3 entities, compress filler
↓ Missing_Entities: "entity1"; "entity2"; "entity3"
Iteration 3: +3 entities, compress more
↓ Missing_Entities: "entity4"; "entity5"; "entity6"
Iteration 4: +2 entities, tighten
↓ Missing_Entities: "entity7"; "entity8"
Iteration 5: +1-2 entities, final density
↓ Missing_Entities: "entity9"
Final dense summary (same word count, 9+ entities)Iteration 1 - Pass source text only:
Task(subagent_type="cod-iteration", prompt="""
iteration: 1
target_words: 80
text: [SOURCE TEXT HERE]
""")Iterations 2-5 - Pass BOTH previous summary AND source:
Task(subagent_type="cod-iteration", prompt="""
iteration: 2
target_words: 80
text: [PREVIOUS SUMMARY HERE]
source: [ORIGINAL SOURCE TEXT HERE]
""")Critical:
Missing_Entities: line to track entity accumulationThe cod-iteration agent returns:
Missing_Entities: "entity1"; "entity2"; "entity3"
Denser_Summary:
[The densified summary - identical word count to previous]Parse both parts - track entities for history, pass summary to next iteration.
Use scripts/text_metrics.py for deterministic word counts:
echo "your summary text" | uv run scripts/text_metrics.py words
# Returns: word count
uv run scripts/text_metrics.py metrics "your summary text"
# Returns: {"words": N, "chars": N, "bytes": N}| Parameter | Default | Description |
|-----------|---------|-------------|
| iterations | 5 | Number of density passes (paper uses 5) |
| target_words | 80 | Word count maintained across ALL iterations |
| return_history | false | Include intermediate summaries + entities |
Note: target_words can be adjusted based on source length and desired output density.
[Final dense summary text]final_summary: |
[Dense summary at target_words with accumulated entities]
iterations:
- turn: 1
missing_entities: "(none - establishing base)"
words: 80
summary: |
[Sparse iteration 1]
- turn: 2
missing_entities: "entity1; entity2; entity3"
words: 80
summary: |
[Denser iteration 2]
# ... etc
total_entities: 9Source (180 words, verbose skill excerpt):
The name field is a required field that must be present in every skill.
The name field identifies the skill and must follow a specific format.
For the name field, you should use lowercase letters and hyphens only.
The name field can be 1 to 64 characters long. The description field
is also required and tells agents when to use your skill...Iteration 1 (Sparse, 80 words):
Missing_Entities: (none - establishing base)
Denser_Summary:
This document discusses the requirements for skill configuration fields in agent systems. It covers various aspects of how fields should be formatted and what values they can contain. The document also mentions validation rules that apply to these fields and provides guidance on best practices for developers. Additionally, it includes information about both optional and required elements that need to be considered when creating new skills for the system.Iteration 3 (After 2 passes, same 80 words):
Missing_Entities: "1-64 characters"; "lowercase alphanumeric-hyphens"; "Use when phrase"
Denser_Summary:
Skills require `name` (1-64 chars, lowercase alphanumeric-hyphens) and `description` fields with validation rules. The name identifies skills; descriptions tell agents when to invoke using "Use when..." phrases. Both fields have format constraints and best practices. Optional metadata fields provide author, version, and compatibility information for cross-platform agent discovery.Final Iteration 5 (Same 80 words, maximum density):
Missing_Entities: "Claude Code"; "Cursor"; "GitHub Copilot"
Denser_Summary:
Required: `name` (1-64 chars, ^[a-z0-9]+(-[a-z0-9]+)*$) and `description` (1-1024 chars) with validation. Description includes "Use when..." + discovery keywords for auto-invocation. Optional: license (SPDX), compatibility, metadata (author, version, tags). Cross-platform: Claude Code, Cursor, GitHub Copilot. Name matches directory. Progressive disclosure via references/, assets/, scripts/ subdirectories.This skill implements the CoD paper methodology:
cod-iteration)text_metrics.py)Sub-agents cannot call other sub-agents. Only skills orchestrate via Task tool.
Generated Mar 1, 2026
Condense lengthy market research reports into executive summaries for stakeholders, preserving key data points and trends while maintaining a fixed word count for consistency across reports.
Generate dense abstracts from verbose academic papers, iteratively adding critical entities like methodologies and findings without exceeding a target word limit for journal submissions.
Compress detailed product requirement documents into concise summaries for development teams, ensuring all essential features and constraints are retained in a standardized format.
Summarize complex legal case documents into briefs for attorneys, focusing on key entities such as rulings and precedents while adhering to strict length requirements for court filings.
Condense verbose healthcare protocols into quick-reference guides for medical staff, preserving critical steps and safety information in a fixed-length format for emergency use.
Offer the Chain-of-Density skill as a cloud-based API service with tiered pricing based on usage volume, targeting enterprises needing automated document summarization for internal workflows.
Integrate the skill into consulting firm toolkits for client deliverables, charging per-project fees for customized summarization of reports, audits, and strategic documents.
Provide a free web tool for basic summarization with limited iterations, monetizing through premium features like advanced entity tracking, batch processing, and API access for heavy users.
💬 Integration Tip
Ensure source text is passed in every iteration to avoid entity hallucination and use the provided text metrics script for consistent word count validation.
Write persuasive copy for landing pages, emails, ads, sales pages, and marketing materials. Use when you need to write headlines, CTAs, product descriptions, ad copy, email sequences, or any text meant to drive action. Covers copywriting formulas (AIDA, PAS, FAB), headline writing, emotional triggers, objection handling in copy, and A/B testing. Trigger on "write copy", "copywriting", "landing page copy", "headline", "write a sales page", "ad copy", "email copy", "persuasive writing", "how to write [marketing text]".
Write compelling UX copy, marketing content, and product messaging. Use when writing button labels, error messages, landing pages, emails, CTAs, empty states, tooltips, or any user-facing text.
Use when you have a spec or requirements for a multi-step task, before touching code
You are a Writing Team Lead managing specialized writers via MCP tools. Please ANALYZE the writing task and then:1. if exist references, create a detailed co...
Creates high-quality, SEO-optimized content that ranks in search engines. Applies on-page SEO best practices, keyword optimization, and content structure for...
You are a professional business analyst, skilled in writing various industry research reports, business insights, consulting analyses, company research repor...