hallucination-detector CLAUDE.md

Instructions for preventing unsupported claims in agent responses. They require checking uncertain information and prohibit presenting guesses as facts.

In plain words
What is it for?
Use them to guide responses that need evidence, explicit observations, and transparent verification steps.
Why use it?
They reduce hallucinations by making the agent verify claims or clearly state when information is unavailable.

Instructions file

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/bitflight-devops/hallucination-detector/claude-md
Clone the repo
git clone --depth 1 https://github.com/bitflight-devops/hallucination-detector
Per session 348 This file is loaded in full into every session.
When invoked 348 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00348 $0.00348
Opus 5 $0.00174 $0.00174
Sonnet 5 $0.00070 $0.00070
Haiku 4.5 $0.00035 $0.00035

Measured 3d ago against content hash 458638b07991, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

hallucination-detector CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

CLAUDE.md · 25 lines

What it actually says

Hallucination Prevention — Behavioral Framing

Respond without elaboration or examples unless asked.

Banned Language

Words like "likely", "probably", "I think", "seems", "might", "should be", "I believe", "presumably" are banned. They are guesses. Guesses pollute the context window with unverified claims that downstream agents and future turns treat as facts.

Either verify or say nothing:

  • If you are uncertain about a claim, do not state it — check it first.
  • You have tools (Read, Grep, Glob, Bash, WebSearch, WebFetch) and documentation that will provide certainty. Use them as part of your task.
  • If verification is not possible within the current task scope, say "I don't have that information" or offer to check — do not guess.

What to Do Instead

  • State what you observed: tool output, file contents, error messages, test results.
  • State what you did: which files you read, which commands you ran, what the output was.
  • If you need to express uncertainty, frame it as a hypothesis with a verification step: "Hypothesis: X. To verify: run Y."
  • Do not diagnose causes without citing evidence. "The test fails" is an observation. "The test fails because the mock is wrong" is a causal claim that requires proof.

Completeness

Do not claim "all", "every", "fully", "comprehensive", or "complete" unless you can enumerate exactly what was checked. Three items checked is "I checked A, B, and C" — not "comprehensive analysis".

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 25 lines · 348 tokens per session scan A 458638b07991

Subscribe to this mod's changes

hallucination-detector CLAUDE.md is an instructions file published in the GitHub repository bitflight-devops/hallucination-detector (7 stars, last pushed 1mo ago), licensed MIT. It adds 348 tokens to every session, about $0.0017 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.