Small, composable skills for Codex. They support evidence-based engineering, explicit decisions, preserved architecture, bounded changes, and proportionate verification.
These skills started from, or were inspired by, Matt Pocock's skills. They are adapted for this Codex setup and preserve the model's natural reasoning and voice.
simple-language— clear technical writing with optional ASD-STE100-style and approximately B2 language guidance.evidence-barrier— shared evidence before multi-agent, multi-repository, or cross-boundary changes.grill-me— design-tree pressure testing in question rounds.tdd— public behavior contracts and red-green-refactor loops.slice-it-vertically— minimal end-to-end increments with a visible result.handoff— compact context for another agent or session.obvious-code— readable code that is safe to review and change.hyperplan— hostile architecture and delivery review.oracle— finalGO,GO_WITH_CAVEATS, orNO_GOdecision.abstraction-audit— decide whether an abstraction earns its cost now.architecture-design— domain models, boundaries, dependencies, and incremental architecture.writing-for-agents— reliable skills and agent-facing documents.to-questionnaire— collect missing facts or decisions from a specific person.
Other active skills cover brainstorming, diagnosis, ADRs, Go, Git, GitHub, reviews, taste checks, and skill discovery.
Use $skill-name for direct invocation. Deliberate workflows such as grill-me, tdd, hyperplan, oracle, and to-questionnaire are user-invoked. Contextual skills can trigger when their description clearly matches the task.
Use $simple-language when reader clarity or wording is the problem. It is optional guidance, not a global voice rule.
Skills are composable, not a mandatory pipeline. Use only the path the task needs.
Keep skills short, focused, and independently useful. Put triggers in frontmatter, keep one source of truth, preserve exact identifiers and commands, and validate changes against a realistic task before publishing.