Your AI coding assistant is fast. Make it accountable without slowing it down.
Vibe Code Safely is a compact operating agreement for AI-assisted development: understand the work, agree the boundary, make the smallest sufficient change, and report real evidence.
curl -fsSL https://raw.githubusercontent.com/Enesevki/vibe-code-safely/main/CLAUDE.md -o CLAUDE.mdcurl -fsSL https://raw.githubusercontent.com/Enesevki/vibe-code-safely/main/AGENTS.md -o AGENTS.mdAdd the matching file to a project. No CLI, package manager, account, or framework is required.
The reusable skill lives at
skills/vibe-code-safely/SKILL.md. Install
or copy that folder into your Codex skills directory when you want these rules
available across projects.
- The agent surfaces assumptions and asks when ambiguity matters.
- Non-trivial work gets a clear goal, acceptance criteria, non-goals, and verification before implementation.
- Scope creep, speculative complexity, and drive-by refactors are resisted.
- Risky changes trigger an explicit pause instead of silent execution.
- Handoffs contain actual verification evidence and visible limitations.
The guideline also requires a short preflight before substantial implementation
and a structured handoff afterwards. Append your project’s stack, test commands,
architecture constraints, and security boundaries to the Project-specific rules section in the same file—no additional framework required.
See EXAMPLES.md for examples of turning vague prompts into bounded, verifiable requests.
These guidelines are working when you see:
- Fewer unnecessary lines in diffs: requested changes, not drive-by cleanup.
- Clarifying questions and a preflight before substantial implementation.
- Smaller rewrites because the first solution matches explicit success criteria.
- Real commands, test output, and unverified limits in handoffs—not just “it works”.
- Explicit pauses before risky or scope-expanding work.
MIT