Anti-Hallucination
A verification-hierarchy checklist that catches guessed facts before they're stated.
Testowano · Działa
Co robi ten skill
Anti-Hallucination is a pure prompting methodology (no scripts, no external files) that ranks claims by evidence strength -- direct file read, search result, inference, general knowledge, or unverified guess -- and requires Claude to say 'I don't know, let me check' rather than assert unverified specifics. It triggers whenever accuracy matters: verifying file contents, function signatures, dependency versions, or any claim about a specific codebase before stating it as fact.
Raport z testu
Concrete A/B: asked an unverified guess about a CLI's flag-combination behavior vs. actually reading the 228-line script -- the guess ('probably ignores --title') was wrong, the file shows it throws a hard usage error (exit 2) when --validate is combined with --title/--out-dir/--mode/--input; the skill's own discipline is what catches that gap.
Testowano: 2026-07-15 · Claude Code 2.x (agent harness)
Instalacja
git clone https://github.com/a-ariff/ariff-claude-plugins mkdir -p ~/.claude/skills cp -r ariff-claude-plugins/plugins/anti-hallucination/skills/anti-hallucination ~/.claude/skills/anti-hallucination
Komendy i przykładowe prompty
/anti-hallucinationA verification-hierarchy checklist that catches guessed facts before they're stated.
Skille uruchamiają się na zwykłe polecenia — bez komend do zapamiętania. Po instalacji aktywują go prompty takie jak te (po angielsku):
Before you answer, please verify that claim against the actual fileDon't guess the function's signature, go check the source file firstWhat does this CLI flag actually do? Confirm it, don't just assume