Design-integrity guardrails for Codex and Claude Code: AGENTS.md, review skill, and deterministic hooks against speculative catches, fallbacks, and API detours.
This repository also configures its own agents. See what design-integrity-guardrails tells them →
a1024053774/design-integrity-guardrails
Skill Claude CodeCodex
Audit whether tests, evals, benchmarks, autoresearch tasks, generated artifacts, public APIs, or user-visible workflows provide independent behavioral evidence. Use at a completion checkpoint, not for ordinary implementation or documentation-only changes.
a1024053774/design-integrity-guardrails
Skill Claude CodeCodex
Route one frozen candidate change through a focused structural-integrity review, or to behavioral-acceptance-review when an explicit evaluation surface changed. Use at a completion checkpoint, not during ordinary edits; do not use for documentation-only, formatting-only, or routine test maintenance.
a1024053774/design-integrity-guardrails
Skill Codex
Establish a short, evidence-backed decision gate before architecture or broad implementation when missing facts, requirements, external behavior, or acceptance criteria could change the direction. Use only for consequential or materially uncertain work.
a1024053774/design-integrity-guardrails
Skill Codex
Preserve failure evidence and prove that tests detect the behavior they claim to cover. Use for bug fixes, regression coverage, test changes, flaky-test investigation, and behavior changes with a concrete acceptance contract; do not use for documentation-only or formatting-only work.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: