fact-check-before-trust

verification-before-completion checks that a task was done. This skill checks that the facts are correct. An agent confidently reporting a £716 visa fee as £70,000 will pass completion verification — this skill catches it.

When to invoke

Invoke this skill before treating any agent output as authoritative when the output contains:

• Numbers or money (prices, quantities, measurements, statistics)
• Dates and deadlines (filing deadlines, release dates, expiry dates)
• Named entities (people, organisations, laws, product names)
• Causal claims ("X causes Y", "because of Z")
• Superlatives ("the largest", "the only", "the most recent")

Skip for: code output, file system operations, and clearly self-contained tasks (renaming a variable, formatting a document).

Verification protocol

Step 1 — Extract claims Identify every verifiable claim in the output. List them explicitly: