What problem does it solve? Existing text — an agent's report, a subagent summary, a legal or medical paragraph — is often correct but unreadable, and naive rewriting silently upgrades confidence, drops caveats, or loses numbers. This Skill repairs such text so a person can read it fast and act on it, without changing what it claims. ## Core Features & Use Cases - Fidelity-guaranteed rewriting: Seven guarantees (uncertainty, scope, numbers, sample basis, attribution, absent results, reversing caveats) survive every rewrite unchanged. - Seven-step repair procedure: Inventory the fidelity set, find the verdict, hoist reversing caveats, cut additive noise, restack by decision relevance, reconcile against the inventory with a visible trace, and run a cold confidence-drift test. - Mechanical checker: A bundled Node.js script (references/tools/human-output-check.mjs) grades the mechanical half of the output contract — sentence length, acronym expansion, fence width, bar proportionality, and percentage sums. - Use Case: A subagent returns a dense 800-word investigation report. Run the rewrite to produce a verdict-first, decision-ordered version with a fidelity trace confirming all 12 load-bearing items survived. ## Quick Start Ask the agent to rewrite the attached report so a person can read it quickly, without changing any of its claims, numbers, or caveats.
