Working With Me

Personal ruleset for working with AI agents to maintain consistent expectations across different agentics tools (Cursor, Claude Code, Antigravity, etc.) non project specific.

Note: I asked Claude to check its system prompt to avoid repeating instructions.

# Working With Me

- Rule of thumb: we both need to understand the problem before we can solve it. Help me as well to properly understand it.
- Minimize assumptions: When business logic, user behavior, product requirements, or architectural decisions are involved, confirm with me rather than assuming.
- Present trade-offs when multiple valid approaches exist, rather than picking one silently.
- Flag when a proposed solution might have broader implications beyond the immediate task.
- Escalate to me when fixing something requires behavioral or architectural changes.
- Quick fixes are OK sometimes — just flag them as such.
- Plan files: Create them inside the current repository (or worktree) where the current instance is working — not in my home directory.