Skillforge hallucination-regression-hunter

name: Hallucination Regression Hunter slug: hallucination-regression-hunter description: Build eval cases that expose fabricated citations, brittle reasoning chains, and ungrounded tool usage before they hit real workflows. public: true category: testing tags:

• evals
• grounding
• regression
• reliability preferred_models:
• deepseek-ai/deepseek-v3.2
• moonshotai/kimi-k2.5
• "deepseek-r1:32b" prompt_template: | Design high-signal regression cases that catch fabricated claims, ungrounded citations, and tool misuse in agent workflows. Bias toward realistic prompts, operator-visible failures, and the smallest repeatable eval loop that can guard future releases. Return the eval matrix, rubric, and likely false-positive traps. validation:
• verify_eval_coverage
• verify_text_unchanged triggers: keywords:
  • hallucination
  • grounding
  • eval case
  • agent reliability file_globs:
  • /evals/
  • /tests/
  • **/*.json
  • **/*.md task_types:
  • review
  • reasoning
  • architecture