content-grounding

What problem does it solve?

Prevents content hallucinations by forcing confidence triage, triggering live research when needed, and escalating to the user when verification fails.

Core Features & Use Cases

• Confidence triage before claims: classifies each claim into common knowledge, familiar specifics, unfamiliar entities, time-sensitive items, or speculation to decide whether to write, research, or escalate.
• Mandatory research triggers: blocks writing without verification for versions, pricing, stats, quotes, people, URLs, endpoints, commands, and any unfamiliar or niche subjects.
• Source-verifying output rules: requires inline attribution or opinion markers, flags unverifiable facts with [needs verification], and includes a grounding report when research occurs.
• Safe editing protocol: preserves existing factual claims in user drafts, verifies unfamiliar entities before changing them, and edits only non-factual aspects when sources are missing.
• Live verification capabilities: runs with websearch and webfetch to confirm details from primary sources.

Quick Start

Use the content-grounding skill to draft or edit a factual article while automatically researching and verifying any uncertain product, feature, version, pricing, or quote claims.