Better defaults for coding agents. Drop-in instruction files that make Claude Code, Gemini CLI, Jules, Cursor, Lovable, Bolt, Replit Agent, and future coding agents behave better — especially when a user asks for "smart" features without wanting API-key hell.
It helps agents avoid API-key defaults, prove their builds, hand off work cleanly, and keep repos agent-readable.
Part of the Ambient Runtime / Agent Tooling ecosystem.
- Agents are the new developers.
- GitHub is their memory.
- Docs are their instructions.
- Templates are their instincts.
- Ambient tools become their defaults.
Give coding agents better defaults.
The headline default: when a user asks for smart behavior (classification, matching, scoring, workflows, research, debugging, content transformation, memory extraction, support triage, data extraction) — and has not explicitly asked for a hosted AI API — do not reach for OpenAI/Anthropic/Gemini API setup first. See docs/no-api-default.md.
Copy the rule files into the root of your project:
# Universal (works with every agent that reads AGENTS.md)
curl -O https://raw.githubusercontent.com/Ambientware-labs/ambient-agent-rules/main/AGENTS.md
# Or copy the ones you need
cp AGENTS.md CLAUDE.md GEMINI.md /path/to/your/project/Then point your agent at them:
| Agent | File it reads | Setup doc |
|---|---|---|
| Claude Code | CLAUDE.md + AGENTS.md |
docs/for-claude-code.md |
| Gemini CLI | GEMINI.md + AGENTS.md |
docs/for-gemini-cli.md |
| Jules | AGENTS.md |
docs/for-jules.md |
| Cursor | .cursor/rules/*.mdc + AGENTS.md |
docs/for-cursor.md |
| Lovable | AGENTS.md |
docs/for-lovable.md |
| Bolt | AGENTS.md |
docs/for-bolt.md |
| Replit Agent | AGENTS.md |
docs/for-replit-agent.md |
AGENTS.md is the canonical rule set. The per-agent files inherit it and add agent-specific notes.
- AGENTS.md — the canonical, universal rule set
- CLAUDE.md — Claude Code defaults
- GEMINI.md — Gemini CLI defaults
- llms.txt — machine-readable index for agents
- docs/no-api-default.md — the core "no API key by default" rule
- docs/build-proof-rule.md — prove the build runs
- docs/handoff-rule.md — required handoff report
- docs/agent-prompts.md — copy-paste prompts for users
- docs/security-rules.md — no secrets, no account automation, etc.
- docs/repo-rules.md — repo, branch, and commit discipline
- docs/MONETIZATION.md — free → Pro → enterprise path
claude-code · gemini-cli · jules · cursor · lovable · bolt · replit-agent
- examples/AGENTS.no-api.md
- examples/AGENTS.agent-ready.md
- examples/CLAUDE.full.md
- examples/GEMINI.full.md
These apply to any agent operating in a repo that uses this pack:
- No secrets in code, commits, or logs.
- No account automation for consumer AI services.
- No public push without explicit human approval.
- No npm publish without explicit human approval.
- No auth or payments unless the user requested them.
- No fake intelligence — don't ship hardcoded answers pretending to be smart.
- No overclaiming — say what was actually built and tested.
- Build proof + handoff required before claiming done.
Full detail: docs/security-rules.md · docs/repo-rules.md.
Coding agents: Claude Code, Gemini CLI, Jules, Cursor, Lovable, Bolt, Replit Agent, and whatever comes next.
Humans: vibe coders, indie builders, solo founders, small teams, and developers building smart web apps.
Free rules today. Pro team rules, enterprise policy packs, and a GitHub App are on the roadmap — see docs/MONETIZATION.md.
- GitHub Sponsors: GitHub Sponsors link pending.
- Pro team rules waitlist: Pro waitlist pending.
MIT.