feat: agent-friendly zero-interactive setup

[the-herbalist]

What This Does

Makes MemPalace fully setup-able by AI agents with zero interactive prompts. Three non-breaking additions:

1. MEMPALACE_NONINTERACTIVE env var — Respects this env var (in addition to existing --yes flag) to skip all input() prompts in confirm_entities(). Works as a global kill switch for interactive mode.

2. mempalace mine --auto — Auto-detects whether the target directory contains project files or conversation exports by scanning for known patterns (Claude/ChatGPT transcripts, Slack exports, OpenAI JSON format, etc.). Picks the right mining mode automatically.

3. mempalace agent-setup <dir> command — One-shot non-interactive setup: init + auto-detect content + mine + status verification. Forces non-interactive mode internally. Returns exit code 0 on success.

4. AGENT.md — Agent discovery document at repo root. Written for LLM parsing — any AI agent finding this repo can execute the setup flow autonomously.

Usage

bash
Existing behavior unchanged
mempalace init ~/project --yes

New — agent-friendly
MEMPALACE_NONINTERACTIVE=1 mempalace init ~/project
mempalace mine ~/project --auto
mempalace agent-setup ~/project

Backward Compatibility

• All existing commands work identically
• --yes flag on init is unchanged
• --mode auto is opt-in (default is still projects)
• agent-setup is a new subcommand, no collision
• Zero breaking changes

Files Changed

• mempalace/cli.py (+182 lines: _detect_mine_mode, cmd_agent_setup, --mode auto support)
• mempalace/entity_detector.py (+3 lines: env var check in confirm_entities)
• AGENT.md (new file, agent setup documentation)

[bensig]

Fixed in #123 — mempalace init --yes is now fully non-interactive. Closing as duplicate.