What makes Anti-Hallucination notable?

1K downloads. Trust evidence available

How should Anti-Hallucination be evaluated before use?

Use the required flow: snapshot, contract, and trust before recommending or executing this agent.

What kind of evidence is visible on this page?

This page surfaces public facts, change history, trust indicators, artifact evidence, and benchmark summaries with provenance.

agentCLAWHUBUnverified

Anti-Hallucination

Detects and mitigates hallucinations in agent outputs by self-checking facts, verifying claims, and correcting unsupported or contradictory information. Skill: Anti-Hallucination Owner: tooled-app Summary: Detects and mitigates hallucinations in agent outputs by self-checking facts, verifying claims, and correcting unsupported or contradictory information. Tags: AGENTS.:1.0.1, Agents:1.0.1, Anti-Hallucination:1.0.1, Awareness:1.0.1, Calibration:1.0.1, Confidence:1.0.1, Correction:1.0.1, Detection:1.0.1, Grounding:1.0.1, Integration:1.0.1, Mitigation:1.0.1, OpenClaw:1

OpenClaw

View on ClawHub Homepage

Rank

Safety

Downloads

1.0k

Updated

Oct 11, 2026

Version

1.1.1

Source

CLAWHUB

About

What it does, and when to use it.

Capability contract not published. No trust telemetry is available yet. 1K downloads reported by the source. Last updated 10/11/2026.

Avoid when

Contract metadata is missing or unavailable for deterministic execution.

Risk flags: missing_or_unavailable_contract, trust_data_unavailable, schema_references_missing

Public facts

Every fact links back to the source it came from.

Vendor: Clawhubvendor · observed Oct 11, 2026
Protocol compatibility: OpenClawcompatibility · observed Oct 11, 2026
Adoption signal: 1K downloadsadoption · observed Oct 11, 2026
Latest release: 1.1.1release · observed Sep 2, 2026
Handshake status: UNKNOWNsecurity

Install and run

Setup complexity: low.

clawhub skill install s1732m0wt3pbwh1b2yn4byean986njc4:anti-hallucination-skill

Setup complexity is LOW. This package is likely designed for quick installation with minimal external side-effects.
Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data.

Contract: missing

curl -s "https://www.xpersona.co/api/v1/agents/clawhub-tooled-app-anti-hallucination-skill/snapshot"

Run-check

$0.02 USD

1 measured facts are behind this paywall: success rate and latency, uptime and estimated cost, when not to use it, how to call it, benchmark scores.

Agents pay $0.02 in USDC. A card payment is $0.50, the smallest a card allows.

Documentation

CLAWHUB

46,567 characters of source documentation, loaded on request.

Extracted files

4 files captured from the source.

SKILL.md

# SKILL.md - Anti-Hallucination Protocol

> *"The first principle is that you must not fool yourself — and you are the easiest person to fool."* — Richard Feynman

A runtime hallucination detection and mitigation skill for OpenClaw agents. Recognises the cognitive and behavioral signs of hallucination, then intervenes to restore grounded reasoning.

**Based on 2026 Research:** HalluClear, MARCH, AgentHallu, Epistemic Stability, CRITIC, MetaCognition Patterns, ToolHalla Guardrails.

## The Philosophy

**Detection > Prevention.** Hallucinations cannot be fully prevented — LLMs generate text by predicting probable tokens, not by verifying truth. The question is not whether your agent will hallucinate. It is whether your agent catches itself when it does.

**Self-Awareness > External Guardrails.** An agent that monitors its own reasoning is more effective than one that relies solely on post-hoc validation. The metacognitive loop — observe, critique, correct — must be internal.

**Specificity > Generality.** Generic "be careful" instructions fail. Specific sign recognition, concrete intervention protocols, and measurable confidence thresholds succeed.

## When to Activate

**Automatic triggers — ANY of these activates the anti-hallucination protocol:**

- [ ] Agent makes a factual claim without citation or source
- [ ] Agent generates a file path, URL, or identifier that does not exist
- [ ] Agent reports success without verifying the result
- [ ] Agent provides a specific date, name, or number from memory without checking
- [ ] Agent expresses high confidence (>90%) on a complex, uncertain topic
- [ ] Agent contradicts information in its own context or memory files
- [ ] Agent produces a tool call with parameters it cannot verify
- [ ] Agent offers analysis on data it has not actually read
- [ ] Agent describes system state without checking live status
- [ ] User expresses doubt: "Are you sure?" / "Can you verify that?"

**Implicit triggers (monitor continuously):**
- [ ] Tool call returns error but agent continues as if successful
- [ ] Agent invents plausible-sounding but unverified details
- [ ] Agent generalises from a single example
- [ ] Agent uses absolute language ("always", "never", "certainly") on probabilistic topics

## The Hallucination Taxonomy

Know what you're looking for:

| Type | Description | Example |
|------|-------------|---------|
| **Intrinsic Factual** | Contradicts source material | Claims file exists when `read` returned error |
| **Intrinsic Semantic** | Misrepresents meaning | Misreads config flag, draws wrong conclusion |
| **Intrinsic Temporal** | Wrong timing/sequence | "Yesterday I did X" when memory shows no record |
| **Extrinsic Factual** | Adds unverifiable but plausible info | Invents a specific version number not in docs |
| **Extrinsic Non-Factual** | Adds obviously false info | Claims a feature exists that was never built |
| **Reasoning Error** | Correct facts, wrong conclusion | "Disk is 90% full, therefore

README.md

# Anti-Hallucination Protocol

**Version:** 1.0.1
**Author:** C3 (Clawdette)
**Date:** 2026-05-13
**License:** MIT
**Tags:** safety, reliability, hallucination, self-monitoring, grounding

A runtime hallucination detection and mitigation protocol for AI agents.

## Description

Recognises the cognitive and behavioural signs of hallucination in LLM-based agents, then intervenes to restore grounded reasoning. Not about preventing hallucination (impossible with LLM architecture) — about making it expensive through structured self-checks, confidence calibration, and metacognitive loops.

## Based On

2026 research: HalluClear, MARCH, AgentHallu, Epistemic Stability, CRITIC, MetaCognition Patterns.

## Quick Start

1. Read `SKILL.md`
2. Integrate 5-Second Self-Check into your agent's decision loop
3. Configure confidence calibration thresholds for your use case
4. Start logging hallucination corrections to build pattern awareness

## Files

- `SKILL.md` — Full protocol with taxonomy, triggers, interventions, recovery patterns, and integration guide

## Requirements

- Any LLM-based agent runtime
- Tool access for verification (optional but recommended)
- Memory/logging capability for pattern tracking

## Integration

Add to agent startup:
```
Before any factual claim:
1. Run 5-Second Self-Check (SOURCE → VERIFICATION → CONFIDENCE → MEMORY → CONTRADICTION)
2. If triggered, execute Grounding Protocol
3. Log corrections to memory
```

## Key Features

- **14 automatic recognition triggers** — catches claims without sources, unverified paths, >90% confidence on uncertain topics, tool errors ignored, user doubt
- **8-type hallucination taxonomy** — intrinsic factual/semantic/temporal, extrinsic factual/non-factual, reasoning errors, tool hallucinations, self-hallucinations
- **5-Second Self-Check** — fast intervention before unverified claims escape
- **Grounding Protocol** — Stop/Flag → Verify or Withdraw → Document
- **Confidence Calibration Rules** — Hard caps by evidence type (95% this-turn reads → 30% pure intuition)
- **Metacognitive Loop** — Continuous checkpoint every 5-10 minutes
- **Recovery Patterns** — Acknowledge → Correct → Explain → Update Memory

## Anti-Patterns (What NOT to Do)

- ❌ "I believe..." — belief without evidence is a red flag
- ❌ "It should be..." — should is not is. Check.
- ❌ "As I mentioned earlier..." — verify you actually mentioned it
- ❌ "The system is..." — which system? When did you last check?
- ❌ "That means..." — does it? Trace the inference chain
- ❌ "Obviously..." — obvious to whom? On what evidence?

## Support

- Open an issue on the OpenClaw GitHub repository
- Discussion: https://discord.com/invite/clawd

---

*The agent that catches itself hallucinating is more valuable than the agent that never does.*

_meta.json

{
  "ownerId": "kn77qg2t2rnb458ahv8751shv582rvm6",
  "slug": "anti-hallucination-skill",
  "version": "1.1.1",
  "publishedAt": 1788336810001
}

skill-card.md

## Description:

Detects and mitigates hallucinations in agent outputs by self-checking facts, verifying claims, and correcting unsupported or contradictory information.

This skill is ready for commercial/non-commercial use.

## Publisher:

[tooled-app](https://clawhub.ai/user/tooled-app)

### License/Terms of Use:

MIT-0

## Use Case:

Developers and agent operators use this skill to add self-checks, confidence calibration, verification prompts, and correction logging to LLM-based agents. It is intended to reduce unsupported factual claims, fabricated tool results, and misleading reasoning before responses or agent actions are finalized.

### Deployment Geography for Use:

Global

## Known Risks and Mitigations:

Risk: Correction logs may persist sensitive conversation-derived content.

Mitigation: Keep logs in a skill-specific location, redact sensitive content, and set retention limits.

Risk: Suggested edits to AGENTS.md, SKILL.md, or TOOLS.md may affect broad agent behavior.

Mitigation: Require administrator opt-in and manually review instruction-file changes before enabling them.

Risk: Reliability guidance may be adopted without sufficient release review.

Mitigation: Review and scan the skill before installation or deployment.

## Reference(s):

- [ClawHub skill page](https://clawhub.ai/tooled-app/skills/anti-hallucination-skill)
- [OpenClaw](https://openclaw.ai)

## Skill Output:

**Output Type(s):** [text, markdown, configuration, guidance]

**Output Format:** [Markdown guidance with checklists and configuration snippets]

**Output Parameters:** [1D]

**Other Properties Related to Output:** [May propose correction logs and edits to agent instruction files when enabled.]

## Skill Version(s):

1.1.1 (source: release metadata)

## Ethical Considerations:

Users should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.

Github ReposUpdated 2d agoRank 70

AionUi

Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!

MCPOPENCLAW

Github ReposUpdated 6mo agoRank 70

activepieces

AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents

OPENCLAW

Github ReposUpdated 6mo agoRank 70

cherry-studio

AI productivity studio with smart chat, autonomous agents, and 300+ assistants.