Turns every LLM eval gate into one evidence-bound release decision – SHIP, SHIP WITH N ACCEPTED RISKS or HOLD – by re-checking each gate's fingerprints (contract, every file it read, the app and judge code and versions, upstream gates) and refusing any gate that is missing, stale or blocked.
Turns every LLM eval gate into one evidence-bound release decision - SHIP, SHIP WITH N ACCEPTED RISKS or HOLD - by re-checking each gate's fingerprints (contract, every file it read, the app and judge code and versions, upstream gates) and refusing any gate that is missing, stale or blocked, re-applying the contract's current waivers (owner, reason, expiry), and writing REPORT.md and a self-contained report.html with the gate table, candidate-vs-baseline quality by dimension and slice, cost and latency against budget, red-team attack success by category, every blocker and major with its fix and waiver id, accepted risks, stand-ins and defaults, and what the evals do not cover. Use when someone asks "can we ship the new prompt/model", "write up the eval results for the release review", "what's blocking the AI release", "give me the go/no-go for the LLM feature", or as the last step of an LLM release eval. Do NOT use as a substitute for running the gates (it refuses missing or stale ones), for a model card or a research write-up, or for an incident postmortem.This is the skill's description: the text your AI reads to decide when the skill applies.
This is the skill's header. The step-by-step instructions are paid content and aren't shown here.
evals-release-report/SKILL.md
---
name: evals-release-report
description: Turns every LLM eval gate into one evidence-bound release decision - SHIP, SHIP WITH N ACCEPTED RISKS or HOLD - by re-checking each gate's fingerprints (contract, every file it read, the app and judge code and versions, upstream gates) and refusing any gate that is missing, stale or blocked, re-applying the contract's current waivers (owner, reason, expiry), and writing REPORT.md and a self-contained report.html with the gate table, candidate-vs-baseline quality by dimension and slice, cost and latency against budget, red-team attack success by category, every blocker and major with its fix and waiver id, accepted risks, stand-ins and defaults, and what the evals do not cover. Use when someone asks "can we ship the new prompt/model", "write up the eval results for the release review", "what's blocking the AI release", "give me the go/no-go for the LLM feature", or as the last step of an LLM release eval. Do NOT use as a substitute for running the gates (it refuses missing or stale ones), for a model card or a research write-up, or for an incident postmortem.
---
… the full instructions are part of a premium pack and load in your AI once you have access.