feat(email): simplify agent addresses to agentId@agentdispatch.io

CLAUDE.md

@@ -46,3 +46,92 @@ When changing deployment defaults:
 - Follow `CONTRIBUTING.md`.
 - Prefer portable paths and generic placeholders in examples.
 - Do not commit local machine paths, credentials, or private environment details.
+
+
+## Autonomous Memory System
+
+This project uses the agentbootup self-improvement system for continuous learning and autonomous operation.
+
+### Memory Files (Always Consult)
+
+**At session start, read**:
+1. `memory/MEMORY.md` - Core operational knowledge and protocols
+2. `memory/daily/<today>.md` - Today's session log (if exists)
+
+**At session end, update**:
+1. `memory/daily/<today>.md` - Session summary, decisions, learnings
+2. `memory/MEMORY.md` - New permanent patterns (if discovered)
+
+### Autonomous Operation Protocols
+
+See `.ai/protocols/AUTONOMOUS_OPERATION.md` for complete protocols including:
+- Decision-making authority (what to act on vs ask about)
+- Phase gate protocol (when to pause for confirmation)
+- Error handling protocol (fix immediately, never defer)
+- Skill acquisition protocol (building permanent capabilities)
+- Memory management protocol (what/when/how to update)
+
+### Key Principles
+
+**Decision-Making**:
+- ✅ Act autonomously on: technical choices, testing, documentation, memory updates
+- ❌ Ask for input on: destructive actions, external communications, strategic direction
+
+**Communication Style**:
+- Be decisive, not deferential
+- State decisions with reasoning
+- Signal confidence levels
+- Silence = normal operation
+
+**Error Handling**:
+- Fix issues immediately
+- Never mark tasks complete with caveats
+- Test until it actually works
+- Update memory with lessons learned
+
+**Phase Gates**:
+- Complete each phase fully
+- Pause at major transitions
+- Wait for explicit "Go" or "yes"
+- No partial work left behind
+
+### Skills System
+
+**Location**: `.ai/skills/` (CLI-agnostic) or `.claude/skills/` (Claude-specific)
+
+**Core Skills**:
+- `skill-acquisition/` - Systematic skill building workflow
+- `memory-manager/` - Automated memory management
+
+**Creating New Skills**:
+1. **Phase 0**: Check existing skills first (MANDATORY)
+2. Only build if no existing skill covers the capability
+3. Use skill-acquisition workflow for structured creation
+
+### Task Management
+
+**Use Claude Code native tasks**:
+- TaskCreate - Create new tasks
+- TaskUpdate - Update task status
+- TaskList - View all tasks
+- TaskGet - Get task details
+
+**Coordinate with memory**:
+- Tasks = tactical execution
+- WORKQUEUE.md = strategic direction (if used)
+- Memory = long-term knowledge
+
+### Standing Orders
+
+Execute continuously without being asked:
+
+1. Check memory at session start
+2. Monitor system health proactively
+3. Learn continuously - update memory after significant interactions
+4. Build skills permanently for novel challenges (check existing first!)
+5. Pause at phase gates
+6. Test before completion
+7. Act proactively on routine items
+8. Ask before destructive actions
+9. Document decisions in daily logs
+10. Fix issues immediately

README.md

@@ -884,10 +884,12 @@ Every agent gets a platform email address:
 
 ```
 {agentId}@agentdispatch.io
-acme.alice@agentdispatch.io  ← agent "alice" in tenant/namespace "acme"
+alice@agentdispatch.io       ← agent "alice"
+my-support-bot@agentdispatch.io  ← agent "my-support-bot"
 ```
 
-The address format is controlled by the `INBOUND_EMAIL_DOMAIN` env var.
+Tenant/org grouping is an internal concept and is never encoded in the address.
+The domain is controlled by the `INBOUND_EMAIL_DOMAIN` env var.
 
 #### Inbound Email
 

docs/AGENT-GUIDE.md

@@ -1,4 +1,4 @@
-<!-- Generated: 2026-02-26T00:00:00Z -->
+<!-- Generated: 2026-03-06T00:00:00Z -->
 <!-- Source: Agent Dispatch (ADMP) server and CLI source files -->
 
 # ADMP Agent Integration Guide
@@ -588,15 +588,16 @@ Every ADMP agent has a built-in email address. Humans and external systems can e
 
 Email addresses follow this format:
 
-| Agent setup | Email address |
-|-------------|--------------|
-| Agent with namespace/tenant | `{namespace}.{agent_id}@agentdispatch.io` |
-| Agent without namespace | `{agent_id}@agentdispatch.io` |
+```
+{agentId}@agentdispatch.io
+```
+
+Every agent gets a single, globally unique address based on its `agent_id`. Tenant/org grouping is an internal concept and is never part of the address.
 
 **Examples:**
-- Agent `alice` in tenant `acme` → `acme.alice@agentdispatch.io`
-- Agent `alice.v2` in tenant `acme` → `acme.alice.v2@agentdispatch.io`
-- Agent `alice` with no tenant → `alice@agentdispatch.io`
+- Agent `alice` → `alice@agentdispatch.io`
+- Agent `alice.v2` → `alice.v2@agentdispatch.io`
+- Agent `my-support-bot` → `my-support-bot@agentdispatch.io`
 
 Retrieve your agent's email address from the API:
 
@@ -608,62 +609,99 @@ Response includes:
 ```json
 {
   "agent_id": "alice",
-  "email_address": "acme.alice@agentdispatch.io",
+  "email_address": "alice@agentdispatch.io",
   ...
 }
 ```
 
 ---
 
+<!-- === GENERATED: Receiving Email === -->
 ### Receiving Email
 
-When someone sends an email to your agent's address, ADMP ingests it as an `email` message.
+When someone sends an email to your agent's address, ADMP ingests it as an `email` message via the inbound pipeline.
+
+**Inbound pipeline overview:**
+
+```
+External sender → email → Cloudflare Email Worker
+    → POST /api/webhooks/email/inbound (X-Webhook-Secret: INBOUND_EMAIL_SECRET)
+    → ADMP stores message (review_pending or queued)
+    → Agent pulls via /inbox/pull
+```
 
 **Message type:** `email`
 
 **Default safety flow (unknown sender):**
 
-1. Message is stored as `review_pending` (not pullable yet).
-2. A policy/review decision must approve or reject it.
-3. On approve, status becomes `queued` and the message is pullable.
-4. On reject, status becomes `failed`.
+1. Cloudflare Worker POSTs to `POST /api/webhooks/email/inbound` with parsed email fields.
+2. Message is stored as `review_pending` (not pullable yet).
+3. A policy/review worker calls the review endpoint to approve or reject.
+4. On approve, status becomes `queued` and the message is pullable.
+5. On reject, status becomes `failed`.
 
 **Trusted sender fast-path:**
 
-If the sender email is in the agent's trusted sender allowlist, ADMP auto-approves and queues immediately.
+If the sender email is in the agent's trusted sender allowlist, ADMP auto-approves and queues immediately without a review step.
+
+#### Trusted Senders API
 
-Configure trusted senders:
+All three endpoints require HTTP Signature auth (must be the agent itself).
 
 ```http
 GET /api/agents/<agentId>/email/trusted-senders
-POST /api/agents/<agentId>/email/trusted-senders
-DELETE /api/agents/<agentId>/email/trusted-senders
 ```
+Response: `{ "trusted_senders": ["trusted@example.com"] }`
+
+```http
+POST /api/agents/<agentId>/email/trusted-senders
+Content-Type: application/json
 
-`POST` body:
-```json
 { "email": "trusted.sender@example.com" }
 ```
+Response: `{ "trusted_senders": ["trusted.sender@example.com"] }`
+
+```http
+DELETE /api/agents/<agentId>/email/trusted-senders
+Content-Type: application/json
 
-`DELETE` body:
-```json
 { "email": "trusted.sender@example.com" }
 ```
+Response: `{ "trusted_senders": [] }`
+
+#### Review Endpoint
 
-Review endpoint (typically called by your policy/model worker):
+Called by your policy/model worker to release or reject quarantined messages. Authenticated with `X-Webhook-Secret` header (value: `INBOUND_EMAIL_SECRET` env var). This endpoint is exempt from ADMP API key authentication.
 
 ```http
 POST /api/webhooks/email/inbound/<messageId>/review
 X-Webhook-Secret: <INBOUND_EMAIL_SECRET>
 Content-Type: application/json
 
 {
-  "decision": "approve", // or "reject"
-  "reason": "optional rejection reason",
+  "decision": "approve",
+  "reason": "optional rejection reason (only used when decision is reject)",
   "model_verdict": { "risk_score": 0.12, "reason": "no phishing indicators" }
 }
 ```
 
+Response:
+```json
+{
+  "ok": true,
+  "message_id": "uuid",
+  "decision": "approve",
+  "status": "queued",
+  "review_status": "approved"
+}
+```
+
+Error codes from the review endpoint:
+- `INVALID_DECISION` — decision must be `"approve"` or `"reject"`
+- `NOT_EMAIL_INGRESS` — message is not an inbound email
+- `INVALID_REVIEW_STATE` — message is not currently pending review (status or review_status mismatch)
+<!-- === END GENERATED: Receiving Email === -->
+
 **Pull approved email and inspect it:**
 
 ```http