Onboarding Guides for AI Agents
dimonb19a · reddit · 2026-07-13
A recent post argues that onboarding an AI agent should be treated just like onboarding a new human colleague.
The core idea is that agents lack context every time they launch, and guessing isn't reliable. Therefore, project rules should be structured into distinct tiers:
- Tier 1: Essential
- AGENTS.md acts as a universal entry point across tools, documenting tech stacks, commands, and hard rules.
- CLAUDE.md serves as a supplement since Claude Code doesn't read AGENTS.md directly.
- Multi-step workflows can be written as SKILL.md files for multiple tools to load.
- Tier 2: Tool-Specific Supplements
- Copilot and Cursor have their own additional config files, mainly for optimization rather than overriding.
- Tier 3: Optional Enhancements
- MCP servers, Gemini CLI configs, and system prompt docs for raw APIs.
The author stresses that these files should not be maintained manually. They should be generated from a single source of truth via deterministic scripts to prevent drift, as incorrect documentation is worse than having none at all.
More from coding & agent
- Codex helps build Valdiluce, an open-world game with climbing, gliding and gondolas — Dimillian · 2026-07-22
- HeyGen adds a media-sourcing skill for coding agents with 75k images and 10k tracks — HeyGen · 2026-07-22
- Agent search bottlenecks are now about variance, not raw latency — rohanpaul_ai · 2026-07-22
- LangSmith adds tracing for Pipecat, LiveKit, OpenAI Realtime, and Gemini Live — LangChain · 2026-07-22
- An MCP server signs every AI agent tool call into a verifiable Merkle chain — Funky_Chicken_22 · 2026-07-22
- Annotated transcript of a Claude Code team interview is now available — trq212 · 2026-07-22