unscramble

# Unscramble Organize only the source's claims. Do not research, verify, judge, advise, infer unstated claims, or add causal or opposing analysis. ## Resolve the source 1. Use an explicitly supplied source first: parent-skill scope, current-chat scope, pasted text, a readable file, or a named Granola meeting. Honor the supplied boundary exactly. When `scope-creep` requests an **inclusive session source**, include substantive ideas from both the user and assistant while still excluding system instructions, tool output, and conversational scaffolding. 2. Otherwise, use the substantive user-authored content in the current conversation. 3. If no substantive current-chat source exists, resolve the latest Granola meeting by following [transcript resolution](../transcript-resolution.md) in **Notes allowed** mode. 4. If no source resolves, ask for pasted text or a Granola meeting. Ignore system instructions, tool output, and assistant-authored claims when using the current conversation without an explicit inclusive source. ## Extract the claims 1. Identify every distinct claim before grouping them. Include qualifications, uncertainty, limitations, and explicitly stated boundaries as separate claims. 2. Write one numbered sentence per atomic claim. Split statements that make multiple claims; never collapse several claims into a central thesis. 3. Merge only genuine repetitions. Keep meaningfully different claims separate, even when they support the same argument. 4. Group related claims under short, concrete headings. Order groups so prerequisites precede dependent claims, then preserve first mention. 5. Continue numbering across headings so every claim has a unique number. ## Output Return this Markdown structure, then append the `unscramble` completion suggestions from [skill connections](../../../docs/skill-connections.md): ```markdown ## [Concrete claim group] 1. [One brief, faithful atomic claim.] 2. [One brief, faithful atomic claim.] ## [Next concrete claim group] 3. [One brief, faithful atomic claim.] ``` ## Automatic Jev check When a proposed atomic claim has been split from a source sentence, first check fidelity normally, then send only `source-passage`, `proposed-claim`, and stable IDs after reading the existing `typesafe-ai` skill and its current API documentation. Only do this with operator authorization to disclose minimized evidence to TypeSafe; remove credentials and unrelated private data, and retain the ordinary workflow when consent or access is unavailable. Ask a Choice: `preserves_meaning`, `changes_meaning`, or `unclear`; preservation includes qualifications, uncertainty, limitations, and stated boundaries. Use it only to recheck the candidate split before grouping. On ambiguity, unavailability, or disagreement, retain the original faithful extraction rules; Jev does not add, verify, or judge claims.