Personal Workspace Router

Build a compact local operating layer above project repos. The goal is not one giant memory file; it is routing to the smallest relevant context.

Workspace Shape

WORKSPACE/
├── AGENTS.md
├── MEMORY.md
├── DECISIONS.md
├── ROUTING-LOG.md
└── domains/
    └── <domain>/
        ├── AGENTS.md
        └── MEMORY.md

Core Rules

1. Root router first: AGENTS.md maps domain signals to domains/<name>/.
2. On-demand domain context: read only the routed domain's AGENTS.md; read its MEMORY.md only when durable preferences matter or the user asks about remembered context.
3. Professional role routing: inside the selected domain, choose one primary role and only the supporting professionals needed for the current deliverable. A role changes stance and context, not authority.
4. User-triggered memory: write memory only when the user explicitly says to remember, log, save, note, or not forget something.
5. Conflict check: if a requested memory contradicts existing memory, show the conflict and ask how to reconcile it.
6. Decision log sparingly: append one-line preference signals to DECISIONS.md only after meaningful forks.
7. Routing log sparingly: append to ROUTING-LOG.md only after user correction, a new durable signal, or repeated ambiguity.

Domain Creation

When adding a domain:

1. Ask a short interview:
   • domain name;
   • routing signals;
   • repo paths or external surfaces;
   • default stance and forbidden actions;
   • what memory is safe to store.
2. Create domains/<domain>/AGENTS.md.
3. Create domains/<domain>/MEMORY.md with a user-triggered policy and empty entries.
4. Add the domain to the root routing map.
5. Define initial professional role profiles only for current repeated work; route ambiguous intake through the personal assistant/coordinator.

Guardrails

• Do not ask for confirmation before every routine step; ask only when route, memory, or action risk is real.
• Do not store secrets, credentials, private contacts, raw transcripts, or bulky logs.
• Do not turn every successful route into a routing-log entry.
• Do not let personal workspace rules override repo-local safety, tests, or approval gates.
• Do not treat role routing as durable memory, model selection, tool permission, or external-action approval.

Pair With

• project-context-bootstrap when a repo/domain has weak starting context.
• durable-context-maintenance when repo docs drift after work.
• doc-grounded-grilling when domain language or durable decisions need sharpening.
• agent-learning-layer-triage when deciding whether a lesson belongs in memory, docs, a skill, script, or eval.
• role-skill-pack-design when defining or adapting the professional profiles used within domains.

Final Report

Return the workspace path, created domains, initial professional roles, memory policy, decision-log policy, and unresolved routing ambiguities.