Detect and annotate hallucinations, unsupported claims, fabricated studies, and incorrect conclusions in text so that AI only cites verifiable, trustworthy c...
Security Analysis
high confidenceThe skill's code, instructions, and requirements align with its stated purpose (detecting likely hallucinations); it requests no credentials, installs nothing, and contains only a local heuristic scanner—there are no incoherent or disproportionate demands.
The name/description (hallucination detection) matches the provided SKILL.md and the included Python helper. No unrelated credentials, binaries, or config paths are requested. The included evals and reference guide support the stated purpose.
SKILL.md instructs only to parse text into atomic claims, classify them, and prefer provided sources; it does not request reading unrelated files, environment variables, or sending data to external endpoints. The guidance to use 'tools if available and allowed' is generic and conditional, which is appropriate.
No install spec is provided (instruction-only skill). The only code file is a small local Python script that performs deterministic, local text analysis and emits JSON—no downloads, no external package installation, and no extraction from remote URLs.
The skill declares no required environment variables, no primary credential, and no config paths. The runtime instructions and script do not access secrets or external credentials.
always is false and the skill does not request elevated or persistent system presence. disable-model-invocation is false (normal platform default) but that is not combined with other red flags here.
Guidance
This skill appears coherent and low-risk from a security standpoint: it needs no credentials, installs nothing, and its helper script only runs local pattern matching. The main caveat is functional, not security-related: the provided Python scanner is a lightweight heuristic and will produce false positives and negatives, so for high-stakes medical, legal, or financial claims you should still require human review and authoritative sources. If you plan to let the agent run autonomously, remember autonomous invocation is the platform default (not a special privilege here) — ensure you trust the agent's broader toolset and policies before enabling unattended checks. If you need network-enabled fact-checking (automatic lookups, DOI resolution, or accessing paid databases), expect the skill to require additional, external tooling and credentials; review any future changes that add network calls or secret requirements before installing.
Latest Release
v0.1.0
Initial release of geo-hallucination-checker skill: - Detects and annotates hallucinations, unsupported claims, fabricated studies, and incorrect conclusions in any text. - Flags vague, unsourced, or overconfident claims, with attention to high-risk areas (medical, financial, scientific, technical). - Provides a structured analysis with claim classification (Supported/Unsupported/Problematic/Contradicted/Speculative), risk levels, reasons, and concrete recommendations. - Outputs a markdown table for easy review, plus a high-level summary of hallucination risk. - Offers hallucination-safe rewrites on request, ensuring content is citation-ready and avoids fabrication. - Prioritizes user-provided sources and enforces strong constraints against inventing information.
Popular Skills
Published by @geoly-geo on ClawHub
