원클릭으로
steering-memory
Memory model rules — 2-layer precedence, output boundary, entity-specific refusal
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Memory model rules — 2-layer precedence, output boundary, entity-specific refusal
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Analyze 1:1 patterns with a specific person — action item follow-through, recurring topics, carry-forward rate. Reports facts from notes only. Does NOT assess relationship quality. Use for any 1:1 pattern analysis: "analyze my 1:1s with X", "1:1 trends with X", "1:1 patterns", "how are my 1:1s with X going?"
Create personal calendar time blocks — focus time, task blocks, deep work. Finds free slots, proposes options, confirms before any write — no attendees, ever. For reminders, use /myna:reminder. Does not handle meeting prep or scheduling others.
Scan the entire vault for blockers — explicit blocker callouts and overdue tasks. Separates your blockers from tasks assigned to others. Use for any blockers query: "what's blocked?", "show me blockers", "any blockers?", "what's holding things up?"
Deep-dive briefing on one person — role, shared projects, open items, pending feedback, 1:1 history, personal notes. Works for directs, peers, manager, or cross-team. (For all directs at once, use /myna:team-health.)
Catch me up on a project — quick (3-5 bullet TL;DR) or full (status, timeline, blockers, tasks, dependencies, upcoming meetings). Use for any project status or update request: "catch me up on X", "status of X", "how is X going", "where are we with X", "what's the latest on X".
Save information to the vault — quick notes, observations, tasks, recognition, project status. Use when: "capture: [text]", "log this", "add task: [desc]", "observation about [person]", "create recurring task". NOT for processing email/messages (/myna:process-updates), meeting notes (/myna:process-meeting), or inbox triage (/myna:email-triage). For links, use /myna:links.
| name | steering-memory |
| disable-model-invocation | true |
| description | Memory model rules — 2-layer precedence, output boundary, entity-specific refusal |
| user-invocable | false |
If vault_path is not in context, read ~/.myna/config.yaml first. If the file does not exist, tell the user to run /myna:setup and stop.
Myna's behavioral rules live in two layers. Applied together at runtime with explicit precedence:
| Layer | Lives in | Authoritative for |
|---|---|---|
| Hard rules | 6 steering skills (myna-steering-*) | Safety, scope, draft-never-send, vault-only writes, append-only discipline |
| User preferences | workspace.yaml identity fields + Claude Code memory (feedback type) | Behavioral corrections, workflow adjustments observed across sessions |
Runtime resolution:
Behavioral memory informs behavior, never content. Never reference memory preferences in:
The only exception: when the user explicitly asks to see their saved preferences — and only to the user, never in content meant for others.
Preferences that apply broadly across interactions belong in memory. Facts about specific entities belong in entity notes. The litmus test:
Refuse to store entity-specific facts as behavioral preferences. Direct the user to the appropriate entity file instead.