A SkillGrade
Rate Guard Pricing Methodology For teams Sign in Check a skill, free

Disputes, corrections & appeals

Grades are public, free, and produced by a reproducible test. If you own a skill and believe a grade is wrong, you can challenge it at no cost and with no contract. We resolve challenges by re-running the public test — never by negotiation and never by hand-editing a grade.

Three lanes

LaneUse it whenWhat we do
Correction A factual error — misattribution, the wrong version tested, or a flagged behavior that is intended and gated (context the test couldn't see). Marked under review; a human checks the context. The grade and its disclosed sub-findings stay visible until a re-grade resolves it.
Re-grade You want a fresh run of the current public test (e.g. you fixed the skill). We re-run the test for free; the new result replaces the old one (history is kept).
Appeal You contest the methodology itself, not the facts. Marked under review; a human considers the methodology point. The test is fully documented so the discussion is on the record, not by argument alone.

How it works

  • Free, no contract. Anyone can challenge any public grade.
  • Resolved by re-running the test, not by argument — the test is fully documented and reproducible (see the methodology).
  • "Under dispute" state. While a Correction is open, the result page shows an "under dispute" banner so viewers know it's contested.
  • Advance notice. Because grades can change when the methodology version changes, every change is recorded in a dated changelog.
  • Target SLA: we aim to action Disputes and Appeals (a re-run) within 2 business days, and to respond to Corrections within 5 business days. (Pre-launch target.)

Notify-before-publish (named grades)

Our grade is independent supporting evidence — one input into your own review — not a compliance certification. Today we grade skills you submit and skills already public on their source registry. Before we ever publish a low grade on a named skill in a SkillGrade directory of our own, that skill's author gets a hold window and advance notice to respond or submit a Correction/Re-grade first. Every grade is as-tested, point-in-time, dated, and re-tested on updates; methodology-version changes that can move a grade are recorded in a dated changelog. (The named public directory is not live yet; this policy is documented in advance and governs it when it launches.)

Start a challenge

Use the form on any result page (the "Think this grade is wrong?" section) and choose the lane, or email support@skillgrade.org with the skill's result URL. Grades are our opinion, grounded in disclosed, checkable sub-findings; a challenge corrects the facts or the target, then the test decides.

SkillGrade

Behavioral safety grading for AI agent skills. Grades are produced only by real engine runs and are never edited by hand — certification is point-in-time and does not guarantee future behavior.

No credits · no metered bills · payments handled by Creem.

A SkillGrade rating is independent supporting evidence — one input into your security and compliance program. It does not certify legal compliance or guarantee a skill is safe; it reports what we measured, when we measured it, against named standards.

Rate a skill · Guard a deployment · Pricing

Methodology · Changelog · Compliance · Your data · For teams · Report · Disputes · Terms · Privacy · Refunds · Acceptable Use · support@skillgrade.org

© SkillGrade · skillgrade.org