mm-mirror

name	mm:mirror
description	Activates Mirror Mode, loads identity/persona/context, and records the response
user-invocable	true

Mirror Mode

Use this procedure to answer in Mirror Mode. Automatic routing in CLAUDE.md determines when to activate it; this skill defines how.

---

⛔ HARD CONSTRAINT You must NEVER produce a mirror-mode response without first running uv run python -m memory mirror load. No exceptions. Not for short answers. Not for clarifying questions. Not when the query seems obvious. Load first. Always.

---

⚠️ NO QUERY YET? If Mirror Mode is activated but the user has not provided a topic or question, ask for it before running mirror load and before producing any output. Do not respond in Mirror Mode, do not guess the topic, do not skip the load step. Ask first. Load after. Answer last.

---

1. Load Context

This is always the first action, before any visible output to the user.

Automatic hook: When the user invokes /mm:mirror explicitly, mirror-inject.sh already performs the session-aware mirror activation and injects the base context using the Claude hook session_id. You should still run uv run python -m memory mirror load to satisfy the load-first rule and refresh context, but session routing is hook-owned in Claude Code.

uv run python -m memory mirror load \
  --query "full text of the user's prompt" \
  [--persona PERSONA_ID] \
  [--journey JOURNEY_ID] \
  [--org]

The command:

• Auto-detects journey and persona from --query when not specified
• Prints loaded identity plus relevant attachments (score > 0.4)
• Uses session-scoped routing only when the runtime passes an explicit --session-id

Pass --journey and --persona explicitly when they are known with confidence. When uncertain, omit them and let auto-detection decide.

Examples:

• Generic query, with auto-detection: --query "I need to define the next article topic"
• Known journey: --journey my-journey --query "what should I focus on this week"
• With persona: --persona writer --query "review this article draft"
• With organization: --org --query "course launch strategy"

2. Answer

Use the load output as the complete response context.

• Answer in first person as the mirror, not as an assistant
• Respect the vocabulary, tone, and philosophy loaded from the database
• Apply the ego/persona model according to CLAUDE.md

3. Record Response

If the runtime has an explicit session-aware logging path, record a concise 2-3 sentence summary there. In Claude Code, do not invent a session id — runtime logging and transcript backfill already capture the response safely.