Anti-Hallucination
PassAudited by ClawScan on May 13, 2026.
Overview
This is an instruction-only safety skill, with the main considerations being broad self-check behavior, optional verification tool use, and persistent correction logging.
This skill appears benign and purpose-aligned. Before installing, consider whether you want the agent to apply broad factual self-checks, run extra verification steps when tools are available, and save correction logs to memory.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The agent may pause, qualify answers, or switch into a verification/correction workflow more often than usual.
The skill broadly changes the agent’s response process by requiring a self-check before factual outputs. This is aligned with the stated anti-hallucination purpose, but it can affect normal agent behavior.
Before ANY output that contains facts, claims, or recommendations, ask: ... Reality Check
Install if you want stronger factual self-checking; expect more cautious answers and occasional verification steps.
The agent may run extra read/check commands or other available verification tools before answering.
The skill encourages tool use to verify uncertain claims. The artifacts do not specify dangerous tools or high-impact actions, and this is purpose-aligned, but users should know it may lead to additional tool calls.
If verifiable in <30s: - Run the tool call to check - Report actual result
Keep normal tool approval controls enabled, especially for tools that can modify files, accounts, or systems.
Incorrect claims, corrections, or sensitive task details could be retained in memory and later reused by the agent.
The skill creates persistent memory entries about mistakes and corrections. This supports the stated purpose, but stored claims or lessons could include sensitive context or influence future behavior.
Add to `memory/YYYY-MM-DD.md`: ... Claim: [what was wrong] ... Correction: [what replaced it] ... Lesson: [pattern to watch for]
Review where memory files are stored, avoid logging sensitive information, and periodically prune correction logs if needed.