A small set of language-agnostic, portable guidance documents for agentic AI coding assistants and the people working alongside them. Drop any of them into a repository as-is, or copy the parts you need into an AGENTS.md, CLAUDE.md, or a style guide. Each document records its sources inline so a reader can audit the basis for every rule.
-
Programming best practices for agentic AI coding
Twenty-six research-gated practices covering small focused changes, reviewable code, deterministic validation, tests as behavior contracts, secure-by-design defaults, dependency and supply-chain discipline, grounding claims instead of fabricating, treating external content as untrusted input, safe use of powerful tools, performance and resource budgets, concurrency safety, resilient error handling, data privacy, safe migrations, and threat modeling. It adds nine user-mandated design principles, a change checklist, a review rubric, a source matrix, and a copy-paste agent-instruction template. Each research-gated practice is backed by at least three independent sources (three separately operated publishing organizations).
-
A field guide to writing patterns that read as machine-generated: non-ASCII punctuation, the "AI style" word cluster, promotional inflation, negative parallelism, rule-of-three filler, transition pileups, hedging, vague attribution, sycophancy, and low-density genericism. Every entry carries an evidence tier (A peer-reviewed, B documented, C convergent). The target is clear prose that carries information, not text that fools a detector.
-
Web UI/UX guidelines for agents
Web UI/UX heuristics spanning usability foundations, visual hierarchy and typography, responsive layout, forms and data entry, data tables, navigation, dashboards, feedback and loading states, reduced motion, internationalization, and WCAG-aligned accessibility. A heuristic is admitted only when at least three independent reputable publishers corroborate it, with the sources listed under each entry.
These are guidance, not code. Pick the document that fits the task, follow the repository's own conventions first, and treat this set as a default the project can override. The practices are written to be enforceable in review: each one states what good looks like and what to watch for.
Released under the MIT License.