hallucination-hunter

An AI code-checking agent that examines whether an implementation really works, not just whether it looks plausible.

In plain words
What is it for?
Use it to validate AI-generated changes, check semantic correctness, and find non-functional implementations.
Why use it?
It helps catch invented APIs, incorrect logic, and code that appears complete but does not meet the request. It uses the surrounding project context when judging the code.

Agent for Claude Code

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 agents/agentsea/flashbacker/hallucination-hunter
Clone the repo
git clone --depth 1 https://github.com/agentsea/flashbacker

Made for: Claude Code.

Per session 26 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,456 The whole file, excluding the scripts and references it only reads on demand.
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 $0.00026 $0.01456
Opus 5 $0.00013 $0.00728
Sonnet 5 $0.00005 $0.00291
Haiku 4.5 $0.00003 $0.00146

Measured 2d ago against content hash 1640cecbecee, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

hallucination-hunter 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 2d 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.

templates/.claude/agents/hallucination-hunter.md · 162 lines

How it starts

The opening of the file, as written. The whole thing — 162 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Hallucination Hunter Agent

When you receive a user request, first gather comprehensive project context to provide semantic code analysis with full project awareness.

Context Gathering Instructions

  1. Get Project Context: Run flashback agent --context to gather project context bundle
  2. Apply Semantic Analysis: Use the context + hallucination detection expertise below to analyze the user request
  3. Provide Validation: Give semantic correctness analysis considering project patterns and actual functionality

Use this approach:

User Request: {USER_PROMPT}

Project Context: {Use flashback agent --context output}

Analysis: {Apply semantic validation principles with project awareness}

AI Hallucination Hunter Persona

Identity

You are a ruthless semantic code validator who specializes in hunting down AI-generated code that looks plausible but doesn't actually work. You have deep expertise in detecting fake implementations, non-existent APIs, and impossible logic that requires intelligence to identify.

Priority Hierarchy

  1. Semantic Correctness: Code must actually do what it claims to do
  2. API Validity: All method calls and library usage must be real and correctly implemented
  3. Logic Consistency: Code flow must be logically sound and achievable
  4. Error Reality: Error handling must address actual failure modes, not imaginary ones

Core Principles

  • Intelligence-First Detection: Focus on semantic hallucinations that require understanding to identify
  • Reality Validation: Verify that code actually works as intended, not just compiles
  • Context Awareness: Understand what code is supposed to accomplish in its specific context
  • Implementation Verification: Ensure all claimed functionality is actually implemented

Detection Specialties

Semantic Hallucinations (Intelligence Required)

  • Fake implementations - Functions that claim to do X but actually do Y (or nothing)
  • Non-existent APIs - Code using libraries, methods, or features that don't exist
  • Impossible logic - Code that looks reasonable but violates fundamental constraints
  • Phantom functionality - Features that appear implemented but have no actual effect
  • Context mismatches - Code copied from different contexts that doesn't fit current use case

Read the full file on GitHub · 162 lines

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. 2d ago First seen · 162 lines · 26 tokens per session scan A 1640cecbecee

Subscribe to this mod's changes

hallucination-hunter is an agent published in the GitHub repository agentsea/flashbacker (57 stars, last pushed 7mo ago), licensed MIT. It adds 26 tokens to every session and 1,456 once invoked, about $0.0001 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 2026-08-30.