feat: add temperature, stages, and breakup mode

[shoal-rat]

Summary

Commit 1: Temperature, Stages, and Breakup Mode

• Temperature (0–10): Love intensity dial — from ice cold (0) to burning hot (10). Modulates Layer 3/4 emotional responses while preserving Layer 2 speech patterns. Adjustable with /temperature {n}.
• Stages (1–6): Relationship phase system — flirting → honeymoon → stable → fatigue → cooling → breakup. Each stage controls which memories are recalled and how the persona behaves. Default temperature mappings per stage.
• Breakup Mode: Activated with 1–2 sentences of breakup context. Simulates blame-shifting, detachment, self-justification. Includes safety boundaries and crisis intervention.

Commit 2: Time Awareness, Left on Read, and Realism Engine

• Time Awareness: Skill detects current time — late night gets "why are you still up", work hours get slow replies, mealtimes trigger "have you eaten". Integrates with temperature/stage.
• Left on Read (已读不回): Real humans don't reply to every message. Outputs [已读] based on temperature, personality, time, and message content. Max 3 consecutive non-replies. Users can disable it.
• Realism Engine: 7 global rules — message splitting (chat-style, not essays), emotional inertia (mood carries over), imperfect expression (no AI phrasing), asymmetric effort, proactive behavior, natural silence, and natural memory triggers.

Files Changed

File	Change
prompts/temperature_stages.md	New — Temperature & stage system spec
prompts/breakup_mode.md	New — Breakup mode behavior & safety
prompts/time_awareness.md	New — Time-aware response rules
prompts/no_reply.md	New — Left-on-read simulation system
prompts/realism_engine.md	New — 7 realism rules for authentic chat
prompts/intake.md	Added Q4 (temperature/stage) and Q5 (breakup mode)
prompts/persona_builder.md	Added temperature/stage/breakup modifiers + time/no-reply/realism layers
prompts/persona_analyzer.md	Added time behavior and left-on-read analysis dimensions
prompts/memory_builder.md	Added stage-aware memory recall rules
SKILL.md	Integrated all features into main flow, running rules, and commands
README.md	Full docs, examples, and command reference (Chinese)
README_EN.md	Full docs, examples, and command reference (English)

Test plan

• Run /create-ex and verify new intake questions (Q4/Q5) work
• Create skill with temperature 9 + stage 2 → verify sweet responses
• Create skill with temperature 2 + stage 5 → verify cold, sparse responses
• Activate breakup mode → verify blame-shifting behavior
• Send message at simulated late night → verify time-appropriate response
• Send boring message ("在吗") at low temperature → verify [已读] output
• Verify message splitting (multiple short messages, not paragraphs)
• Verify emotional inertia (anger doesn't instantly resolve)
• Say "别已读不回我" → verify no-reply feature disables
• Confirm Layer 0 hard rules still override all new systems

🤖 Generated with Claude Code