Prompt file imported from elsolal/Skillz-Claude-Codex-And-More (
.codex/prompts/gate.md). Copyright stays with the author.
IT IS CRITICAL THAT YOU FOLLOW THIS COMMAND: LOAD the FULL ~/.codex/skills/quality-gate/SKILL.md, READ its entire contents, and execute the loop in standalone mode: report first, ZERO file modification before the user arbitrates. After counter-verification, present the detailed report (file:line, severity, confirmed/refuted with reason, proposed fix), then ask: [A] apply all confirmed P0/P1 and continue the loop | [S] select fixes | [R] report only (gate file written with the as-is verdict).
The user's arguments follow this line: an optional level (1-4, default 3 = complete review: correctness/security, readability, performance + design/SEO/a11y lenses when the diff touches those surfaces, up to 4 rounds) and/or a target diff (default: the default-branch diff, git diff <base>...HEAD). Run the project-probe skill first if .agents/verification.yaml is missing.
Output: the gate file docs/quality/GATE-<date>-<slug>.yaml committed + a summary (verdict, rounds, findings, decisions taken). PASS requires real executable evidence; without any, the verdict caps at CONCERNS — by design.