escalate-to-architect

The safety valve that makes local-first viable: Hermes never guesses on hard things — it queues them for the architect. Fully async/offline: write a file now; the architect reads it next session. Background/why: [[NOTES]].

When to Use

• A skill's escalate_when fires.
• An ambiguous sorting / judgment call.
• Multi-step reasoning, design, refactoring, or new-skill authoring.
• Anything that would modify scripts, configs, or system services.
• (No chat trigger — Hermes invokes this internally when it hits a wall.)

Procedure

1. Open (create if missing): <architect_memory>/Escalations.md, where <architect_memory> = python $MYOS_ENGINES/myos_paths.py get architect_memory.
2. Append:

   ## YYYY-MM-DD HH:MM — <one-line task>
   - **Context:** <note/file/project; [[links]]>
   - **What I tried:** <or "did not attempt — out of scope">
   - **Why escalated:** <the escalate_when reason>
   - **Status:** OPEN

3. If a note is blocked, also tag it #needs-architect.
4. Tell the user plainly: "Queued for the architect: <task>." Then move on.

Pitfalls

• Never attempt the hard task anyway — the whole point is not guessing. Queue it and stop.
• Always fill Context + Why so the architect can resolve without re-deriving the situation.

Verification

• The new ## YYYY-MM-DD … entry is at the tail of Escalations.md with Status: OPEN.
• The user received the one-line "Queued for the architect" confirmation.

Cadence

On demand — whenever Hermes hits something outside its lane.