AGENTS.md & CLAUDE.md Context Architect

Most auto-generated AGENTS.md/CLAUDE.md files just restate your README - folder structure, tech stack, "this is a Next.js app." Your coding agent can already see all of that by reading the repo once. A restating-only context file wastes context budget and can actively hurt agent performance.

What it does

Runs a structured interview - not a "paste your README" form - to extract the landmines a new agent session would otherwise rediscover the hard way:

• Fragile or non-standard build/test/deploy steps
• Files that must never be hand-edited (generated code, vendored deps)
• Commands that look like they'd work but don't
• Past mistakes an agent (or new hire) already made here
• Architecture decisions that look wrong but are deliberate

Output

A single markdown file structured for the AGENTS.md/CLAUDE.md convention - read natively by Claude Code, Cursor, GitHub Copilot, and Codex. Sections: Setup & Build, Testing, Do Not Touch, Landmines, Architecture Decisions, Code Style. Empty sections are omitted - filler wastes the agent's context budget on every session, which is the exact problem this skill exists to fix.

Who it's for

Developers and teams running Claude Code, Cursor, Copilot, or Codex who are tired of their coding agent repeating the same mistakes every session.

FAQ

Do I need to paste my whole codebase?
No - a README, file tree, or package manifest helps it ask sharper questions, but the value comes from your answers about what's NOT visible in the code.

Will it just make things up if I don't know an answer?
No - it asks rather than guesses. A wrong instruction in a context file is worse than a missing one.