Require verifiable evidence before answering fact-sensitive requests.
npx skills add https://github.com/vulikjulik/DeepLom --skill aif-grounded-vulikjulik
Please help me install this Agent Skill. Skill: aif-grounded Source: https://github.com/vulikjulik/DeepLom/tree/main/.opencode/skills/aif-grounded Command: npx skills add https://github.com/vulikjulik/DeepLom --skill aif-grounded-vulikjulik
This Skill eliminates the risk of AI-generated hallucinations, unsubstantiated guesses, and incorrect assumptions that lead to unreliable outputs, especially for high-stakes, fact-sensitive requests where accuracy is non-negotiable.
Invoke the aif-grounded skill when you need a fully verified answer to a high-stakes question, and it will only respond if it can confirm 100% confidence with available evidence.
High-intent search queries and answers about installing and using this skill.
To prevent AI hallucinations during codebase analysis, you need an evidence-based confidence gate that blocks unsubstantiated claims. This approach requires explicit sourcing from local codebases or command outputs, returning a missing evidence checklist instead of guessing when information is insufficient.
An evidence-first workflow for fact verification classifies requests, identifies required evidence sources, and verifies changeable facts before generating any output. It enforces a 100% confidence gate by matching claims against authoritative documentation or user-provided sources.
To enforce no assumptions for AI responses in financial compliance checks, apply guardrails that require explicit evidence sourcing. If full confidence cannot be achieved, the system must return a structured insufficient information response detailing the missing evidence.
Yes, 100% confidence fact verification works with user-provided documentation by treating it as an explicit evidence source. The system verifies all changeable facts against these provided sources before delivering a final answer, ensuring zero hallucinations.
When there is insufficient information for a fully verified answer, the system returns a structured insufficient information response. This response includes a clear checklist of missing evidence, explicitly blocking hallucinations and unsubstantiated guesses.
nicograefFact-check repository documentation against code, configuration, and command outputs.
silva2kandGenerate legal research memos with claim-specific sources and confidence ratings.
burgebjAutomate security audits and vulnerability checks across a codebase.
n-marusovEnforce evidence-based reasoning for AI answers in legal, financial, or medical contexts.
