What problem does it solve? Teams writing release notes, READMEs, blog posts, or competitor comparisons risk publishing unverified or overstated claims. This Skill decides what Sonder Runtime may publicly claim by tying every statement to grounded code evidence, a proven benchmark run at the current revision, or an explicit open/candidate status. ## Core Features & Use Cases - Claim ledger with proof obligations: Six candidate differentiators (execution-grounded learning, rollback-proven self-modification, privacy-first local learning, bounded fleets, signed TUF releases, measured moat) each mapped to grounding files and required proof runs. - Evidence standards for quoted numbers: Enforces the identity rule (suite, suite version, model digest) and documents exactly what scripts/benchmark_moat.py does and does not prove. - Release-gate workflow: Sequences version identity checks, history privacy scans, artifact integrity, and the TUF signing ceremony before any versioned claim ships. - Use Case: Before publishing release notes claiming a retrieval-quality improvement, run the moat benchmark at the current revision, cite the suite and model digest, and label ecosystem comparisons as unverified background knowledge. ## Quick Start Ask the assistant to review this draft announcement and tell me which claims are grounded, which need a benchmark run, and which must be removed.
