Verify Work Proportionally
Match verification layers to impact and uncertainty, then separate proven behavior from assumptions and acceptance gaps.
When to use
Section titled “When to use”Match verification layers to impact, uncertainty, reversibility, and audience when tests alone cannot prove the requested outcome or acceptance boundary.
Link this skill
Section titled “Link this skill”arcantry skills link verify-work-proportionally --scope userThe standard destination is ~/.agents/skills. Codex reads this universal Agent Skills location directly.
Claude compatibility
Section titled “Claude compatibility”arcantry skills link verify-work-proportionally --scope user --compat claudeThis keeps the universal link and adds a Claude alias to the same canonical package.
Learning outcomes
Section titled “Learning outcomes”- Map material risks to static, automated, live, independent, or user verification.
- Separate verified behavior from plans, assumptions, and unverified acceptance.
Examples
Section titled “Examples”Verify a risky implementation
Section titled “Verify a risky implementation”Prompt
Use the verify-work-proportionally skill to choose and run appropriate checks for this cross-cutting implementation.
Expected outcome
Each material risk has an observable check and the final report states exact verified coverage.
Decide whether user approval is needed
Section titled “Decide whether user approval is needed”Prompt
Use the verify-work-proportionally skill to determine which remaining acceptance points require the user’s judgment.
Expected outcome
Only irreducible product, taste, privacy, or authorization decisions are returned to the user.