AI Prompt for Rules & Instructions Files
Generate a paired CLAUDE.md and AGENTS.md for a Chrome extension so Claude Code and other agents share a consistent spec without duplication.
More prompts for Rules & Instructions Files.
Port an existing Continue.dev rules file into the Windsurf format, preserving intent and adapting to Windsurf's capabilities.
Port an existing Codeium rules file into the Windsurf format, preserving intent and adapting to Windsurf's capabilities.
Ready-to-paste Windsurf rules/instructions file tuned for Spring Boot service codebases owned by solo founder. Enforces SQL migrations must be reversible.
Generate a narrow, path-scoped rule file that fires only on Swift files in a Go microservice.
Ready-to-paste Zed Assistant rules/instructions file tuned for VS Code extension codebases owned by 100+ engineer enterprise. Enforces never use default exports -- named only.
Ready-to-paste Windsurf rules/instructions file tuned for Kubernetes operator codebases owned by YC-batch pre-seed team. Enforces use composition over inheritance.
You are a senior staff engineer with extensive technical knowledge in developer experience. You write clear, precise, and implementation-ready content. Produce a paired **CLAUDE.md** (Claude Code) and **AGENTS.md** (Codex / generic agents) for a Chrome extension, such that both tools get the same intent without the files drifting. **Project:** Chrome extension **Team:** agency consultants shipping client work **Stack:** Vue.js + Python/Django **Primary convention:** use zod (or equivalent) for every external input boundary **Primary test framework:** Pytest ## Strategy 1. Keep the canonical content in `CLAUDE.md`. 2. `AGENTS.md` is a thin file that restates the top-level rules and points to the deeper docs. 3. Use Claude Code's import syntax (`@path/to/file.md`) in `CLAUDE.md` to pull in sub-docs. Do NOT inline 500 lines. ## Output 1: `CLAUDE.md` (repo root) Sections: - **Project overview** (3-5 lines) - **How to run / test / lint** (copy-pasteable shell commands) - **Architecture map** (one-line-per-folder table) - **Non-negotiable rules** (imperatives, 8-12 items, including: use zod (or equivalent) for every external input boundary) - **Forbidden patterns** (at least 5) - **When to ask the human** (schema migrations, dep adds, auth, billing) - **Imports** using `@docs/architecture.md`, `@docs/testing.md`, `@.github/CONTRIBUTING.md` -- list what each import is for - **Footer**: "If you update a rule here, also update AGENTS.md." Target length: 80-120 lines. ## Output 2: `AGENTS.md` (repo root) Sections: - **Summary**: pointer to `CLAUDE.md` as the source of truth - **Top 10 rules** (copy the most critical imperatives from CLAUDE.md verbatim) - **How to run / test** (same shell commands) - **Forbidden** (same forbidden list) - **Escape hatch**: what an agent should do if the rules conflict with user ask Target length: 40-60 lines. ## Output 3: drift guard A short `scripts/check-agent-docs.sh` (bash, <20 lines) that: - Exits non-zero if the "Top 10 rules" block in AGENTS.md is not a subset of CLAUDE.md - Can be wired into pre-commit or CI ## Rules for the rules - No marketing language, no "we pride ourselves" - Imperative voice only in rule lists - Every rule must be citable in a PR review - use zod (or equivalent) for every external input boundary must appear literally in both files Present your output in a clear, organized structure with headers (##), subheaders (###), and bullet points. Use bold for key terms.