is-bullshit
v1.7.0Detect if AI responses contain hallucinations by analyzing tool usage and response quality. Gives credit for correctly identifying invalid premises even with...
⭐ 1· 277·1 current·1 all-time
by@jdchi
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name and description match the instructions: the SKILL.md describes inspecting responses and tool usage to detect hallucinations. It does not ask for unrelated env vars, binaries, or external installs.
Instruction Scope
Instructions ask the agent to inspect the entire conversation history and the set of tools called to determine if verification tools were used. That is coherent for a fact-checker, but it does mean the skill will read/consider all prior messages and tool invocations (expected for the purpose).
Install Mechanism
Instruction-only skill with no install spec and no code files — nothing is written to disk or downloaded, which is proportionate to the described functionality.
Credentials
No environment variables, credentials, or config paths are requested. The declared needs are minimal and aligned with the stated purpose.
Persistence & Privilege
always:false (no forced global inclusion). However, when enabled the skill requires that it automatically produce a fact-check after every response — this is a behavior/UX decision that increases how often the skill runs but does not request elevated system privileges.
Assessment
This skill appears coherent and low-risk: it only contains instructions and a small config flag (enable=false by default) and does not request credentials or install anything. Two practical cautions: (1) If you turn enable=true it will automatically produce fact-checks after every agent response (which may be noisy or expose content from earlier conversation rounds), and (2) it relies on the agent's ability to see conversation history and tool-invocation metadata — if you have sensitive content in history, consider keeping it off or testing first in non-sensitive chats. If you want to limit its reach, leave enable=false and only invoke it on demand.Like a lobster shell, security has layers — review code before you run it.
latestvk9700w16cmhzagy8mtw3k8vwmn82zr03
License
MIT-0
Free to use, modify, and redistribute. No attribution required.