summarize

name: summarize description: Condenses complex content into actionable summaries. Use for journal compaction, compiled truth synthesis, research distillation, and any long-form-to-short-form transformation.

Long-Form Summary Compression

Core Principle

If removing a sentence doesn't change the reader's next action, cut it.

Four Phases

Phase 1: Identify Key Claims

• Read the full source before summarizing anything
• Mark every claim that is: actionable, surprising, or load-bearing (other claims depend on it)
• Ignore: repetition, hedging, background the audience already knows, examples that illustrate an already-clear point

Phase 2: Extract Supporting Evidence

• For each key claim, find its strongest supporting evidence
• Keep ONE piece of evidence per claim (the most compelling)
• If a claim has no evidence, flag it as unsupported — don't summarize unsupported claims as fact

Phase 3: Remove Redundancy

• Group claims by topic — merge duplicates
• If two claims say the same thing differently, keep the clearer one
• Convert relative references to absolute: "yesterday" → "2026-04-13", "the bug we discussed" → "the race condition in auth.ts"

Phase 4: Structure for Scanning

• Lead with the most important insight (inverted pyramid)
• Use bullet points for parallel items
• Use bold for key terms on first appearance
• Target: 20% of original length or less

Compression Levels

Anti-Patterns

• Summarizing summaries: each compression pass loses nuance. Go back to the source.
• Context stripping: removing qualifiers that change meaning — "works for small datasets" becomes "works"
• Uniform compression: treating all sections equally. Important sections get more space.
• Opinion laundering: presenting "the author argues X" as "X is true"
• Over-compression: a summary that raises more questions than it answers