Automate rubric and bucket boundary upgrades with auditable governance.
npx skills add https://github.com/Jason5330/ai-self-eval --skill cheat-bump
Please help me install this Agent Skill. Skill: cheat-bump Source: https://github.com/Jason5330/ai-self-eval/tree/main/skills/cheat-bump Command: npx skills add https://github.com/Jason5330/ai-self-eval --skill cheat-bump
This Skill automates the process of upgrading calibration rubrics and bucket boundaries, enabling controlled, auditable changes to scoring systems and reducing drift in model evaluation.
Trigger a rubric bump by issuing /cheat-bump --propose "<new formula>" or run /cheat-bump --bucket-only to perform a lightweight bucket recalibration.
High-intent search queries and answers about installing and using this skill.
You can automate calibration rubric upgrades by triggering a formal proposal mode that enforces cross-model audits, blind sub-agent scoring workflows, and updates state tracking files to ensure auditable governance over the new formulas.
Bucket-only recalibration is a lightweight adjustment process that modifies bucket boundaries without altering the core rubric formula, providing a quick method to correct scoring drift in evaluation pools.
Cross-model auditing works by running blind sub-agent scoring workflows across different models to validate proposed rubric formulas, ensuring the upgraded boundaries maintain evaluation consistency and reduce drift.
You need formal input when executing a complete rubric bump, which requires submitting a proposed formula to trigger enforced validation protocols, cross-model reviews, and state updates for full auditability.
The automated rubric bump process updates rubric_notes.md, rubric-memo.md, and .cheat-state.json to track state changes, document validation protocols, and maintain auditable governance records for the scoring system.
Yes, you can adjust bucket boundaries independently by running a bucket-only mode, which performs lightweight recalibration on the evaluation pools while leaving the underlying rubric formula completely intact.
Owl-ListenerDefine structured rubrics for evaluating AI outputs across seven quality dimensions.
mifunedevDetect and score documentation drift between the docs wiki and live application.
Jason5330Generate immutable blind-prediction logs with YAML headers and hook-based locking.
nicholls-incVerify repository assurance documentation presence and analyze drift and coverage issues.
