/Fabric

name: Fabric
metadata:
  short-description: "Run or update Fabric prompt patterns for analysis, extraction, writing, security, and diagrams. Use ExtractWisdom for adaptive wisdom extraction."
evals:
  suites: [fabric-analyzebill, fabric-analyzeclaims, fabric-analyzecomments, fabric-analyzeincident, fabric-analyzelogs, fabric-analyzemalware, fabric-analyzepaper, fabric-analyzeprose, fabric-analyzerisk, fabric-analyzethreatreport, fabric-auditconsent, fabric-audittransparency, fabric-businessideas, fabric-checkagreement, fabric-checkfalsifiability, fabric-coderules, fabric-codingproject, fabric-compareandcontrast, fabric-createcommand, fabric-cybersummary, fabric-designdocument, fabric-emailheaders, fabric-ethicalframework, fabric-explaincode, fabric-explaindocs, fabric-explainterms, fabric-extractideas, fabric-extractinsights, fabric-extractinstructions, fabric-extractpatterns, fabric-extractpoc, fabric-extractpredictions, fabric-extractrecommendations, fabric-extractreferences, fabric-gitdiffsummary, fabric-goldenrules, fabric-logicalfallacies, fabric-mindvirus, fabric-networkthreatlandscape, fabric-nucleitemplate, fabric-pullrequest, fabric-reportfinding, fabric-reviewdesign, fabric-rhetoricallens, fabric-securebydesignquestions, fabric-secureprompt, fabric-semgreprule, fabric-showintro, fabric-sigmarules, fabric-silentvictims, fabric-stoiclens, fabric-summarize, fabric-summarizedebate, fabric-summarizemeeting, fabric-terraformplan, fabric-threatscenarios, fabric-ultags, fabric-wordsummary]
  deterministic:
    - command: "bun skills/Fabric/Tools/CheckPatterns.ts"
      falsifier: "a key pattern (extract_wisdom, summarize, create_threat_model, analyze_claims) has no system.md under Patterns/"
version: 1.3.0
description: Our ideal-state ports of the most-used Fabric patterns, each an intent with its own Assay suite and code checker that beat the upstream prompt head to head (security, reasoning, reading, engineering, writing), plus every other Fabric pattern run natively from Patterns/. USE WHEN word summary, ws, N levels, fabric, fabric pattern, run fabric, update patterns, any Fabric pattern name, threat report, threat scenarios, security finding, analyze logs, malware, incident, semgrep rule, sigma rules, nuclei template, email headers, terraform plan, analyze claims, logical fallacies, falsifiable, compare and contrast, analyze paper, summarize, meeting summary, extract insights, extract ideas, extract references, design document, explain code, review design, pull request description, commit message, code rules, analyze prose, threat model, mermaid, STRIDE. NOT FOR multi-agent investigation (Research) or content-adaptive extraction (ExtractWisdom).

Fabric

**Contents:** Ideal State · Intent Routing · Data & Config · Gotchas · Examples

Runs any of 240+ specialized prompt patterns across extraction, summarization, analysis, creation, improvement, security, and rating. Common ones: extract_wisdom, create_threat_model, analyze_claims, improve_writing, review_code, mermaid, youtube_summary. Patterns run natively: the model reads the pattern's system.md and applies it directly, with no CLI round-trip. The fabric CLI is used only for YouTube transcripts (-y) and URL fallback (-u). A summarize-family 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).

The skill also holds our own modernized versions of the patterns that rank highest on value, public use and our own use (`References/PortRanking.tsv`), written as ideal-state intents with their own Assay suite and code checker. A port routes only after it beats its upstream prompt head to head on held-out inputs (`Tools/PortDuel.ts`): 51 intents covering 57 patterns, WordSummary included, run instead of the upstream prompt. Seven more are still behind upstream and listed separately; their patterns run upstream until a revision wins.

Ideal State

• The pattern that runs is the one the request named, or the one its wording maps to, and its output keeps the format the pattern's OUTPUT INSTRUCTIONS define.
• A request for a pattern we have ported runs its intent, never the upstream prompt: `create_5_sentence_summary` and `ws` run WordSummary, and every pattern in `Intents/ExecutePattern.md` § Ported patterns runs the intent it names.
• Patterns run natively from `Patterns/`; the fabric CLI is touched only for YouTube transcripts and blocked URLs.
• The local `Patterns/` library carries every pattern in upstream fabric's `data/patterns/`, takes nothing from the local fabric config, keeps the release scrub, and every update reports its before and after counts.

Intent Routing

| Intent | Trigger | File | |--------|---------|------| | **ExecutePattern** | "use fabric", "run pattern", "apply pattern", "fabric extract wisdom" or the `extract_wisdom` pattern by name, "summarize", "analyze with fabric" | `Intents/ExecutePattern.md` | | **UpdatePatterns** | "update fabric", "update patterns", "sync fabric", "pull patterns" | `Intents/UpdatePatterns.md` | | **WordSummary** | `ws` or `ws 3`, "word summary", "N levels from N words to one", "5 levels", `create_5_sentence_summary` | `Intents/WordSummary.md` |

Ported patterns

One intent per ported upstream pattern, close siblings sharing one. Each runs instead of the upstream prompt; `Intents/ExecutePattern.md` § Ported patterns maps every pattern name to its intent.

Security

| Intent | Trigger | File | |--------|---------|------| | **AnalyzeThreatReport** | "analyze this threat report", "just the trends from this report", "extract the commands from this pentest writeup", `analyze_threat_report`, `analyze_threat_report_cmds`, `analyze_threat_report_trends` | `Intents/AnalyzeThreatReport.md` | | **ThreatScenarios** | "threat scenarios", "worried about", `create_threat_scenarios` | `Intents/ThreatScenarios.md` | | **ReportFinding** | "write a pentest f