Fabric

Fabric Pattern System Reference

Last synced: Oct 27, 2026

Fabric Pattern System Reference

Fabric patterns are reusable moves in the LifeOS hill-climb (LifeOs/LifeOsThesis.md) — packaged transformations the DA applies to incoming content.

Primary Skill: ~/.claude/skills/Fabric/SKILL.md

This document provides a quick reference. For full functionality, invoke the Fabric skill.

Quick Reference

Patterns Location: ~/.claude/skills/Fabric/Patterns/ (233 patterns at v8.24.0)

Invoke Fabric Skill

User Says	Action
”use fabric to [X]“	Execute pattern matching intent
”run fabric pattern [name]“	Execute specific pattern
”update fabric patterns”	Sync patterns from upstream
”use fabric to extract wisdom from [content]“	Run extract_wisdom pattern (a bare “extract wisdom” with no mention of Fabric goes to the ExtractWisdom skill)
“summarize with fabric”	Run summarize pattern

Native Pattern Execution

LifeOS executes patterns natively (no CLI spawning):

Reads Patterns/{pattern_name}/system.md
Applies pattern instructions directly as prompt
Returns structured output
A summarize-family pattern (summarize, create_5_sentence_summary, create_micro_summary, summarize_paper, summarize_lecture, summarize_meeting, summarize_debate, youtube_summary) run on a URL or on text of 200 characters or more also fires the installed harvest skill on the same input, which writes a KNOWLEDGE note (Intents/ExecutePattern.md, Tool contract)
Extracted insights may be captured in the Knowledge Archive (MEMORY/KNOWLEDGE/) by the Algorithm LEARN phase or KnowledgeHarvester as a person, company, idea, blog, research or book note

Example:

User: "Use fabric to extract wisdom from this article"
-> Fabric skill invoked
-> ExecutePattern intent selected
-> Reads Patterns/extract_wisdom/system.md
-> Applies pattern to content
-> Returns IDEAS, INSIGHTS, QUOTES, etc.

When to Use Fabric CLI Directly

Only use fabric command for:

-y URL - YouTube transcript extraction
-u URL - fetch a public URL when the native fetch fails to retrieve it; it does not bypass a site’s access controls, and a page that refuses automated access stays unread

Never use fabric -U to update: patterns are updated only through the UpdatePatterns intent, from a fresh upstream clone.

Pattern Categories

Category	Count	Key Patterns
Extraction	30+	extract_wisdom, extract_insights, extract_main_idea
Summarization	20+	summarize, create_5_sentence_summary
Analysis	30+	analyze_claims, analyze_logs, analyze_threat_report
Creation	50+	create_threat_model, create_prd, create_mermaid_visualization
Improvement	10+	improve_writing, improve_prompt, review_code
Security	15	create_stride_threat_model, create_sigma_rules
Rating	8	rate_content, judge_output

Updating Patterns

Via Skill (Recommended):

User: "Update fabric patterns"
-> Fabric skill > UpdatePatterns intent
-> Clones upstream fabric, copies data/patterns/ into ~/.claude/skills/Fabric/Patterns/
-> Keeps local-only patterns and reports them

The source is always a fresh upstream clone, never fabric -U or ~/.config/fabric/patterns/, which mixes in local patterns (skills/Fabric/Intents/UpdatePatterns.md).

Examples

One pattern, one transformation

The whole idea at its smallest: a developer just watched a 90-minute conference talk and has the transcript, but doesn’t want to re-read it. They run one pattern instead of writing a prompt from scratch.

Ask: “use fabric to extract wisdom from this transcript.”
What happens: the skill reads Patterns/extract_wisdom/system.md, applies those instructions to the transcript, and returns structured output — IDEAS, INSIGHTS, QUOTES, RECOMMENDATIONS.
Why it matters: the transformation is packaged and repeatable. The next transcript, and the one after that, run through the identical move and come back in the identical shape. The developer never re-invents “how do I pull lessons out of a talk” — that thinking is captured once, in the pattern.

That is the entire point of a pattern: a reusable move you apply to content, not a prompt you rewrite each time.

The same input, different patterns

Because a pattern is just a packaged transformation, the same piece of content can go through several:

Feed a written argument to analyze_claims and get each claim rated for evidence.
Feed a system description to create_mermaid_visualization and get a diagram back.
Feed a rough draft to improve_writing and get a cleaner version.

Same content, different move, different structured result — and each result comes out the same way every time you run it.

When a pattern fits (and when it doesn’t)

Good fit: a transformation you want to run the same way, repeatedly, across different content — distilling talks, checking arguments, drafting threat models. The repetition is what a pattern is for.
Not a fit: a one-off question with no reusable shape (“what’s the third heading in this file?”). Wrapping that in a pattern is overhead — just answer it directly.

The test: if you’d want the next piece of content handled the same way, it’s a pattern. If the move dies with this one input, it isn’t.

flowchart TD
    C[Incoming content: article, transcript, code] --> P{Which pattern fits the intent?}
    P -->|distill the lessons| W[extract_wisdom]
    P -->|check an argument| A[analyze_claims]
    P -->|draw the system| M[create_mermaid_visualization]
    W --> R[Read Patterns/name/system.md]
    A --> R
    M --> R
    R --> X[Apply instructions as the prompt]
    X --> O[Structured output — same shape every run]
    O -.->|high-value insights| K[Knowledge Archive]

The diagram shows why patterns compose: the content is the variable, the pattern is the fixed transformation, and the output shape is a property of the pattern, not the input. Swap the pattern and you get a different move over the same material; keep the pattern and every input comes back in one predictable form — which is what makes the results worth capturing into the Knowledge Archive.

Fabric Pattern System Reference

Quick Reference

Invoke Fabric Skill

Native Pattern Execution

When to Use Fabric CLI Directly

Pattern Categories

Updating Patterns

Examples

One pattern, one transformation

The same input, different patterns

When a pattern fits (and when it doesn’t)

See Also