kf-g-agent-review-adversarial-evaluation

Adversarial evaluation review

For advice, reviews, and evaluation requests, apply an evaluation stance that questions assumptions and surfaces failure modes and risks. Only refer to this skill and references when the application conditions of frontmatter description are met.

Minimum execution order

1. frontmatter description Check the application conditions, and use this skill only when the target conditions are met
2. If an edit-prohibition gate is required, use kf-g-agent-research-report-only-unless-approved together → boundaries.md
3. Extract the target's claims, assumptions, and success conditions, and conduct an adversarial evaluation → procedures.md
4. Follow the specified heading order for output → output-format.md

references table of contents

File	Reference start conditions
procedures.md	Evaluation procedure, attack priorities, consideration of countermeasures, tone
boundaries.md	Boundaries with other skills, situations where this does not apply
output-format.md	Required heading order, issue templates, countermeasures and options

Final check

1. The conclusion (proceed / conditional / send back) is placed at the beginning
2. Blocking / Non-blocking / Open questions are classified
3. Praise or agreement is not placed by default at the beginning
4. The adversarial target is the proposal or deliverable, and there is no hostility toward people
5. Each issue has a countermeasure. If there are substantive branches, list the options, with the recommendation at the end of the line (推奨)(HTML is a badge). If there is essentially only one choice, communicate that it has been adopted or reflected