fix(org_helpers): correct duplicate phrase in loadWorkspaceEnv comment

The comment had the phrase "the workspace-specific .env" duplicated. Removed the redundant repetition. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
fix(workspace-server): inject /configs token files agent-owned, not root
2026-05-16 10:27:13 +00:00 · 2026-05-16 02:19:11 -07:00 · 2026-05-16 07:32:27 +00:00 · 2026-05-16 06:58:48 +00:00 · 2026-05-15 23:08:56 -07:00 · 2026-05-15 22:27:51 +00:00
40 changed files with 1450 additions and 100 deletions
@@ -0,0 +1,225 @@
+name: E2E Peer Visibility (literal MCP list_peers)
+
+# WHY A DEDICATED WORKFLOW (not folded into e2e-staging-saas.yml)
+# --------------------------------------------------------------
+# This is the systemic fix for a real trust failure. Hermes and OpenClaw
+# were reported "fleet-verified / cascade-complete" because the *proxy*
+# signals were green (registry registration + heartbeat for Hermes; model
+# round-trip 200 for OpenClaw). A freshly-provisioned workspace asked on
+# canvas "can you see your peers" actually FAILS:
+#   - Hermes: 401 on the molecule MCP `list_peers` call
+#   - OpenClaw: native `sessions_list` fallback, sees no platform peers
+# Tasks #142/#159 were even marked "completed" under this proxy flaw.
+#
+# A dedicated workflow (vs extending e2e-staging-saas.yml) because:
+#   - It must provision MULTIPLE distinct runtimes (hermes, openclaw,
+#     claude-code) in ONE org and assert each sees the others. The
+#     full-saas script is single-runtime-per-run (E2E_RUNTIME) and folding
+#     a multi-runtime matrix into it would conflate concerns and bloat its
+#     already-45-min run.
+#   - It needs its own concurrency group so it doesn't fight full-saas /
+#     canvas for the staging org-creation quota.
+#   - It needs an independent, non-required status-context name so it can
+#     be RED today (the in-flight Hermes-401 / OpenClaw-MCP-wiring fixes
+#     have not landed) WITHOUT wedging unrelated merges — and flipped to
+#     REQUIRED in one branch-protection edit once it goes green
+#     (flip-to-required checklist: molecule-core#1296).
+#
+# THE ASSERTION IS NOT A PROXY. The driving script
+# tests/e2e/test_peer_visibility_mcp_staging.sh issues the byte-for-byte
+# JSON-RPC `tools/call name=list_peers` envelope to `POST
+# /workspaces/:id/mcp` using each workspace's OWN bearer token, through
+# the real WorkspaceAuth + MCPRateLimiter middleware chain — the exact
+# call mcp_molecule_list_peers makes from a canvas agent. It does NOT
+# read a registry row, /health, the heartbeat table, or
+# GET /registry/:id/peers.
+#
+# HONEST GATE — NO continue-on-error. Per feedback_fix_root_not_symptom a
+# fake-green mask would defeat the entire purpose. This workflow goes red
+# on today's broken behavior and green only when the root-cause fixes
+# actually land. It is intentionally NOT in branch_protections — see PR
+# body for the required-vs-not decision + flip tracking issue.
+#
+# Gitea 1.22.6 / act_runner notes honored:
+#   - No cross-repo `uses:` (feedback_gitea_cross_repo_uses_blocked). The
+#     actions/checkout SHA is the one e2e-staging-canvas.yml already uses
+#     successfully (a mirrored SHA — see #1277/PR#1292 root-cause).
+#   - Per-SHA concurrency, not global (feedback_concurrency_group_per_sha).
+#   - Workflow-level GITHUB_SERVER_URL pinned
+#     (feedback_act_runner_github_server_url).
+#   - pr-validate posts a status under the same check name so a
+#     workflow-only PR is not silently statusless and the context is
+#     flip-to-required-ready (mirrors e2e-staging-saas.yml's proven shape;
+#     real EC2-provisioning E2E is push/dispatch/cron only — it is 30+ min
+#     and cannot run per-PR-update).
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - 'workspace-server/internal/handlers/mcp.go'
+      - 'workspace-server/internal/handlers/mcp_tools.go'
+      - 'workspace-server/internal/middleware/**'
+      - 'workspace-server/internal/handlers/registry.go'
+      - 'workspace-server/internal/handlers/workspace.go'
+      - 'workspace/a2a_mcp_server.py'
+      - 'workspace/platform_tools/registry.py'
+      - 'tests/e2e/test_peer_visibility_mcp_staging.sh'
+      - '.gitea/workflows/e2e-peer-visibility.yml'
+  pull_request:
+    branches: [main]
+    paths:
+      - 'workspace-server/internal/handlers/mcp.go'
+      - 'workspace-server/internal/handlers/mcp_tools.go'
+      - 'workspace-server/internal/middleware/**'
+      - 'workspace-server/internal/handlers/registry.go'
+      - 'workspace-server/internal/handlers/workspace.go'
+      - 'workspace/a2a_mcp_server.py'
+      - 'workspace/platform_tools/registry.py'
+      - 'tests/e2e/test_peer_visibility_mcp_staging.sh'
+      - '.gitea/workflows/e2e-peer-visibility.yml'
+  workflow_dispatch:
+  schedule:
+    # 07:30 UTC daily — catches AMI / template-hermes / template-openclaw
+    # drift even on quiet days. Offset 30m from e2e-staging-saas (07:00)
+    # so the two don't collide on the staging org-creation quota.
+    - cron: '30 7 * * *'
+
+concurrency:
+  # Per-SHA (feedback_concurrency_group_per_sha). A single global group
+  # would let a queued staging/main push behind a PR run get cancelled,
+  # leaving any gate that reads "completed run at SHA" stuck.
+  group: e2e-peer-visibility-${{ github.event.pull_request.head.sha || github.sha }}
+  cancel-in-progress: false
+
+env:
+  GITHUB_SERVER_URL: https://git.moleculesai.app
+
+jobs:
+  # PR path: post a real status under the required-ready check name so a
+  # workflow-only PR is never silently statusless. The actual EC2 E2E is
+  # push/dispatch/cron only (30+ min). This is NOT a fake-green mask of
+  # the real assertion — it validates the driving script's bash syntax
+  # and inline-python so a broken test script fails at PR time.
+  pr-validate:
+    name: E2E Peer Visibility
+    runs-on: ubuntu-latest
+    if: github.event_name == 'pull_request'
+    timeout-minutes: 5
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+      - name: Validate driving script
+        run: |
+          bash -n tests/e2e/test_peer_visibility_mcp_staging.sh
+          echo "test_peer_visibility_mcp_staging.sh — bash syntax OK"
+          echo "Real fresh-provision MCP list_peers E2E runs on push to"
+          echo "main / workflow_dispatch / daily cron (30+ min EC2 boot)."
+
+  # Real gate: provisions a throwaway org + sibling-per-runtime, drives
+  # the LITERAL list_peers MCP call per runtime, asserts 200 + expected
+  # peer set, then scoped teardown. push(main)/dispatch/cron only.
+  peer-visibility:
+    name: E2E Peer Visibility
+    runs-on: ubuntu-latest
+    if: github.event_name != 'pull_request'
+    timeout-minutes: 60
+
+    env:
+      MOLECULE_CP_URL: https://staging-api.moleculesai.app
+      MOLECULE_ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
+      # LLM provider key so each runtime can authenticate at boot.
+      # Priority MiniMax → direct-Anthropic → OpenAI matches
+      # test_staging_full_saas.sh's secrets-injection chain.
+      E2E_MINIMAX_API_KEY: ${{ secrets.MOLECULE_STAGING_MINIMAX_API_KEY }}
+      E2E_ANTHROPIC_API_KEY: ${{ secrets.MOLECULE_STAGING_ANTHROPIC_API_KEY }}
+      E2E_OPENAI_API_KEY: ${{ secrets.MOLECULE_STAGING_OPENAI_API_KEY }}
+      E2E_RUN_ID: "${{ github.run_id }}-${{ github.run_attempt }}"
+      PV_RUNTIMES: "hermes openclaw claude-code"
+
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+
+      - name: Verify admin token present
+        run: |
+          if [ -z "$MOLECULE_ADMIN_TOKEN" ]; then
+            echo "::error::CP_STAGING_ADMIN_API_TOKEN secret not set (Railway staging CP_ADMIN_API_TOKEN)"
+            exit 2
+          fi
+          echo "Admin token present"
+
+      - name: Verify an LLM key present
+        run: |
+          if [ -z "${E2E_MINIMAX_API_KEY:-}" ] && [ -z "${E2E_ANTHROPIC_API_KEY:-}" ] && [ -z "${E2E_OPENAI_API_KEY:-}" ]; then
+            echo "::error::No LLM provider key set — workspaces fail at boot with 'No provider API key found'. Set MOLECULE_STAGING_MINIMAX_API_KEY (or ANTHROPIC / OPENAI)."
+            exit 2
+          fi
+          echo "LLM key present"
+
+      - name: CP staging health preflight
+        run: |
+          code=$(curl -sS -o /dev/null -w "%{http_code}" --max-time 10 "$MOLECULE_CP_URL/health")
+          if [ "$code" != "200" ]; then
+            echo "::error::Staging CP unhealthy (HTTP $code) — infra, not a workspace bug. Failing loud per feedback_fix_root_not_symptom."
+            exit 1
+          fi
+          echo "Staging CP healthy"
+
+      - name: Run fresh-provision peer-visibility E2E (literal MCP list_peers)
+        run: bash tests/e2e/test_peer_visibility_mcp_staging.sh
+
+      # Belt-and-braces scoped teardown: the script installs an EXIT/INT/
+      # TERM trap, but if the runner itself is cancelled the trap may not
+      # fire. This always() step deletes ONLY the e2e-pv-<run_id> org this
+      # run created — never a cluster-wide sweep
+      # (feedback_never_run_cluster_cleanup_tests_on_live_platform). The
+      # admin DELETE is idempotent so double-invoking is safe;
+      # sweep-stale-e2e-orgs is the final net (slug starts with 'e2e-').
+      - name: Teardown safety net (runs on cancel/failure)
+        if: always()
+        env:
+          ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
+        run: |
+          set +e
+          orgs=$(curl -sS "$MOLECULE_CP_URL/cp/admin/orgs?limit=500" \
+            -H "Authorization: Bearer $ADMIN_TOKEN" 2>/dev/null \
+            | python3 -c "
+          import json, sys, os, datetime
+          run_id = os.environ.get('GITHUB_RUN_ID', '')
+          try:
+              d = json.load(sys.stdin)
+          except Exception:
+              print(''); sys.exit(0)
+          # ONLY sweep slugs from THIS run. e2e-pv-<YYYYMMDD>-<run_id>-...
+          # Sweep today AND yesterday's UTC date so a midnight-crossing run
+          # still matches its own slug (same bug class as the saas/canvas
+          # safety nets).
+          today = datetime.date.today()
+          yest = today - datetime.timedelta(days=1)
+          dates = (today.strftime('%Y%m%d'), yest.strftime('%Y%m%d'))
+          if run_id:
+              prefixes = tuple(f'e2e-pv-{dt}-{run_id}-' for dt in dates)
+          else:
+              prefixes = tuple(f'e2e-pv-{dt}-' for dt in dates)
+          orgs = d if isinstance(d, list) else d.get('orgs', [])
+          cands = [o['slug'] for o in orgs
+                   if any(o.get('slug','').startswith(p) for p in prefixes)
+                   and o.get('instance_status') not in ('purged',)]
+          print('\n'.join(cands))
+          " 2>/dev/null)
+          for slug in $orgs; do
+            echo "Safety-net teardown: $slug"
+            set +e
+            curl -sS -o /tmp/pv-cleanup.out -w "%{http_code}" \
+              -X DELETE "$MOLECULE_CP_URL/cp/admin/tenants/$slug" \
+              -H "Authorization: Bearer $ADMIN_TOKEN" \
+              -H "Content-Type: application/json" \
+              -d "{\"confirm\":\"$slug\"}" >/tmp/pv-cleanup.code
+            set -e
+            code=$(cat /tmp/pv-cleanup.code 2>/dev/null || echo "000")
+            if [ "$code" = "200" ] || [ "$code" = "204" ]; then
+              echo "[teardown] deleted $slug (HTTP $code)"
+            else
+              echo "::warning::pv teardown for $slug returned HTTP $code — sweep-stale-e2e-orgs will catch it within MAX_AGE_MINUTES. Body: $(head -c 300 /tmp/pv-cleanup.out 2>/dev/null)"
+            fi
+          done
+          exit 0
@@ -18,6 +18,10 @@ permissions:
  pull-requests: read
  statuses: write

+concurrency:
+  group: ${{ github.repository }}-${{ github.workflow }}-${{ github.event.issue.number || github.ref }}
+  cancel-in-progress: true
+
 jobs:
  dispatch:
    runs-on: ubuntu-latest
@@ -70,7 +70,7 @@ name: sop-checklist
 # Cancel any in-progress runs for the same PR to prevent
 # stale runs from overwriting newer status contexts.
 concurrency:
-  group: ${{ github.repository }}-${{ github.event.pull_request.number }}
+  group: ${{ github.repository }}-${{ github.workflow }}-${{ github.event.pull_request.number || github.event.issue.number || github.ref }}
  cancel-in-progress: true

 # bp-required: yes  ← emits sop-checklist / all-items-acked (pull_request)
@@ -61,6 +61,10 @@ on:
  pull_request_review:
    types: [submitted, dismissed, edited]

+concurrency:
+  group: ${{ github.repository }}-${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
 jobs:
  tier-check:
    runs-on: ubuntu-latest
@@ -962,6 +962,32 @@ function MyChatPanel({ workspaceId, data }: Props) {
          </div>
        </div>
      )}
+      {/* talk_to_user disabled banner — shown when the workspace has
+           talk_to_user_enabled=false. The agent cannot send canvas messages;
+           the user can re-enable the ability from here without opening settings. */}
+      {data.talkToUserEnabled === false && (
+        <div className="flex items-center gap-2 px-3 py-2 bg-surface-sunken border-b border-line/40 shrink-0">
+          <svg width="14" height="14" viewBox="0 0 16 16" fill="none" aria-hidden="true" className="shrink-0 text-ink-mid">
+            <path d="M8 1a7 7 0 1 0 0 14A7 7 0 0 0 8 1Zm0 10.5a.75.75 0 1 1 0-1.5.75.75 0 0 1 0 1.5ZM8 4a.75.75 0 0 1 .75.75v4a.75.75 0 0 1-1.5 0v-4A.75.75 0 0 1 8 4Z" fill="currentColor"/>
+          </svg>
+          <span className="text-[10px] text-ink-mid flex-1">
+            Agent is not enabled to chat with you.
+          </span>
+          <button
+            onClick={async () => {
+              try {
+                await api.patch(`/workspaces/${workspaceId}/abilities`, { talk_to_user_enabled: true });
+                useCanvasStore.getState().updateNodeData(workspaceId, { talkToUserEnabled: true });
+              } catch {
+                // ignore — user will see no change and can retry
+              }
+            }}
+            className="px-2 py-0.5 text-[10px] font-medium bg-accent/10 hover:bg-accent/20 text-accent rounded border border-accent/30 transition-colors shrink-0"
+          >
+            Enable
+          </button>
+        </div>
+      )}
      {/* Messages */}
      <div ref={containerRef} className="flex-1 overflow-y-auto p-3 space-y-3">
        {loading && (
@@ -176,7 +176,7 @@ export function deriveProvidersFromModels(models: ModelSpec[]): string[] {
 // exactly the point of the platform adaptor. The deep `~/.hermes/
 // config.yaml` on the container is a separate runtime-internal file,
 // not this one.
-const RUNTIMES_WITH_OWN_CONFIG = new Set<string>(["external", "kimi", "kimi-cli"]);
+const RUNTIMES_WITH_OWN_CONFIG = new Set<string>(["external", "kimi", "kimi-cli", "openclaw"]);

 const FALLBACK_RUNTIME_OPTIONS: RuntimeOption[] = [
  { value: "", label: "LangGraph (default)", models: [], providers: [] },
@@ -8,14 +8,18 @@ import { getTenantSlug } from "./tenant";
 export const PLATFORM_URL =
  process.env.NEXT_PUBLIC_PLATFORM_URL ?? "http://localhost:8080";

-// 15s is long enough for slow CP queries but short enough that a
-// hung backend doesn't leave the UI spinning forever. The abort
-// propagates through AbortController so React components can observe
-// the error and render a retry affordance. Callers that know the
-// endpoint is intentionally slow (org import walks a tree of
-// workspaces with server-side pacing) can pass `timeoutMs` to
-// override.
-const DEFAULT_TIMEOUT_MS = 15_000;
+// 35s is long enough for the slowest server-side path (EIC SSH
+// tunnel for tenant EC2 file operations, bounded server-side by
+// `eicFileOpTimeout = 30 * time.Second` in
+// workspace-server/internal/handlers/template_files_eic.go) so the
+// canvas surfaces the server's real error instead of aborting first
+// with a generic timeout. Shorter values caused "Save & Restart" to
+// time out at the client before the backend returned its 5xx. The
+// abort still propagates through AbortController so React components
+// can render a retry affordance. Callers that know an endpoint is
+// intentionally slow (org import walks a tree of workspaces with
+// server-side pacing) can pass `timeoutMs` to override.
+const DEFAULT_TIMEOUT_MS = 35_000;

 export interface RequestOptions {
  timeoutMs?: number;
@@ -519,6 +519,10 @@ export function buildNodesAndEdges(
        // #2054 — server-declared per-workspace provisioning timeout.
        // Falls through to the runtime profile when null/absent.
        provisionTimeoutMs: ws.provision_timeout_ms ?? null,
+        // Workspace abilities — defaults preserved for old platform versions
+        // that don't yet include these columns in the GET response.
+        broadcastEnabled: ws.broadcast_enabled ?? false,
+        talkToUserEnabled: ws.talk_to_user_enabled ?? true,
      },
    };
    if (hasParent) {
@@ -99,6 +99,13 @@ export interface WorkspaceNodeData extends Record<string, unknown> {
   *  @/lib/runtimeProfiles. Lets a slow runtime declare its cold-boot
   *  expectation without a canvas release. */
  provisionTimeoutMs?: number | null;
+  /** When true the workspace may POST /broadcast to send org-wide messages.
+   *  Default false. Toggled by user/admin via PATCH /workspaces/:id/abilities. */
+  broadcastEnabled?: boolean;
+  /** When false the workspace cannot deliver canvas chat messages.
+   *  send_message_to_user / POST /notify return 403 and the canvas
+   *  shows a "not enabled" state with a button to re-enable. Default true. */
+  talkToUserEnabled?: boolean;
 }

 export type PanelTab = "details" | "skills" | "chat" | "terminal" | "config" | "schedule" | "channels" | "files" | "memory" | "traces" | "events" | "activity" | "audit";
@@ -299,6 +299,9 @@ export interface WorkspaceData {
   *  `@/lib/runtimeProfiles` when absent (the default behavior for any
   *  template that hasn't yet declared the field). */
  provision_timeout_ms?: number | null;
+  /** Workspace ability flags (migration 20260514). */
+  broadcast_enabled?: boolean;
+  talk_to_user_enabled?: boolean;
 }

 let socket: ReconnectingSocket | null = null;
@@ -0,0 +1,376 @@
+#!/usr/bin/env bash
+# Staging E2E — fresh-provision peer-visibility gate via the LITERAL MCP path.
+#
+# WHY THIS EXISTS
+# ---------------
+# Hermes and OpenClaw were repeatedly reported "fleet-verified / cascade-
+# complete" because the *proxy* signals were green:
+#   - registry-registration + heartbeat (Hermes), and
+#   - model round-trip 200 (OpenClaw).
+# But a freshly-provisioned workspace, asked on canvas "can you see your
+# peers", actually FAILS:
+#   - Hermes: 401 on the molecule MCP `list_peers` call,
+#   - OpenClaw: falls back to native `sessions_list`, sees no platform peers.
+# Tasks #142/#159 were even marked "completed" under this same proxy flaw.
+#
+# This script codifies the LITERAL user-facing path so it can never silently
+# regress: it provisions a brand-new throwaway org + sibling workspaces via
+# the real control-plane provisioning path, then for each runtime that should
+# have platform peer-visibility it drives the EXACT MCP call the canvas agent
+# makes — `POST /workspaces/:id/mcp` JSON-RPC tools/call name=list_peers,
+# authenticated by that workspace's own bearer token through the real
+# WorkspaceAuth + MCPRateLimiter middleware chain. It then asserts:
+#   (1) HTTP 200,
+#   (2) JSON-RPC `result` present (NOT an `error` object — a -32000
+#       "tool call failed" or a 401 from WorkspaceAuth fails here),
+#   (3) the returned peer set CONTAINS the other provisioned sibling
+#       workspace IDs — not an empty list, not a native-sessions fallback.
+#
+# This is NOT a proxy. It does not look at a registry row, /health, the
+# heartbeat table, or `GET /registry/:id/peers`. It drives the byte-for-byte
+# JSON-RPC envelope that mcp_molecule_list_peers issues from a real agent.
+#
+# It is written to FAIL on today's broken Hermes/OpenClaw behavior and go
+# green only when the in-flight root-cause fixes (Hermes-401, OpenClaw MCP
+# wiring) actually land. That is the point: it is the objective proof gate.
+#
+# AUTH MODEL (mirrors tests/e2e/test_staging_full_saas.sh)
+# --------------------------------------------------------
+#   Single MOLECULE_ADMIN_TOKEN (= CP_ADMIN_API_TOKEN on Railway staging)
+#   drives: POST /cp/admin/orgs (provision), GET
+#   /cp/admin/orgs/:slug/admin-token (per-tenant token), DELETE
+#   /cp/admin/tenants/:slug (teardown). The per-tenant admin token drives
+#   tenant workspace creation; each workspace's OWN auth_token (returned by
+#   POST /workspaces) drives its MCP call.
+#
+# Required env:
+#   MOLECULE_ADMIN_TOKEN   CP admin bearer — Railway staging CP_ADMIN_API_TOKEN
+# Optional env:
+#   MOLECULE_CP_URL        default https://staging-api.moleculesai.app
+#   E2E_RUN_ID             slug suffix; CI passes ${GITHUB_RUN_ID}
+#   PV_RUNTIMES            space list; default "hermes openclaw claude-code"
+#   E2E_PROVISION_TIMEOUT_SECS  default 1800 (hermes/openclaw cold EC2 budget)
+#   E2E_MINIMAX_API_KEY / E2E_ANTHROPIC_API_KEY / E2E_OPENAI_API_KEY
+#                          LLM provider key injected so the runtime can boot
+#   E2E_KEEP_ORG           1 → skip teardown (local debugging only)
+#
+# Exit codes:
+#   0  every runtime saw its peers via the literal MCP call
+#   1  generic failure
+#   2  missing required env
+#   3  provisioning timed out
+#   4  teardown left orphan resources
+#   10 peer-visibility regression reproduced (the gate firing as designed)
+
+set -uo pipefail
+
+CP_URL="${MOLECULE_CP_URL:-https://staging-api.moleculesai.app}"
+ADMIN_TOKEN="${MOLECULE_ADMIN_TOKEN:?MOLECULE_ADMIN_TOKEN required — Railway staging CP_ADMIN_API_TOKEN}"
+RUN_ID_SUFFIX="${E2E_RUN_ID:-$(date +%H%M%S)-$$}"
+PV_RUNTIMES="${PV_RUNTIMES:-hermes openclaw claude-code}"
+PROVISION_TIMEOUT_SECS="${E2E_PROVISION_TIMEOUT_SECS:-1800}"
+
+# Slug MUST start with 'e2e-' so the sweep-stale-e2e-orgs safety net
+# (EPHEMERAL_PREFIXES) catches any leak this run fails to tear down.
+SLUG="e2e-pv-$(date +%Y%m%d)-${RUN_ID_SUFFIX}"
+SLUG=$(echo "$SLUG" | tr '[:upper:]' '[:lower:]' | tr -cd 'a-z0-9-' | head -c 32)
+
+ORG_ID=""
+TENANT_URL=""
+TENANT_TOKEN=""
+
+log()  { echo "[$(date +%H:%M:%S)] $*"; }
+fail() { echo "[$(date +%H:%M:%S)] ❌ $*" >&2; exit 1; }
+ok()   { echo "[$(date +%H:%M:%S)] ✅ $*"; }
+
+admin_call() {
+  local method="$1" path="$2"; shift 2
+  curl -sS -X "$method" "$CP_URL$path" \
+    -H "Authorization: Bearer $ADMIN_TOKEN" \
+    -H "Content-Type: application/json" "$@"
+}
+tenant_call() {
+  local method="$1" path="$2"; shift 2
+  curl -sS -X "$method" "$TENANT_URL$path" \
+    -H "Authorization: Bearer $TENANT_TOKEN" \
+    -H "X-Molecule-Org-Id: $ORG_ID" \
+    -H "Content-Type: application/json" "$@"
+}
+
+# ─── Scoped teardown ───────────────────────────────────────────────────
+# Deletes ONLY the org this run created (DELETE /cp/admin/tenants/$SLUG
+# with the {"confirm":$SLUG} fat-finger guard). Never a cluster-wide
+# sweep — honors feedback_cleanup_after_each_test and
+# feedback_never_run_cluster_cleanup_tests_on_live_platform. The
+# workflow's always() step + sweep-stale-e2e-orgs are the outer nets.
+teardown() {
+  local rc=$?
+  set +e
+  if [ "${E2E_KEEP_ORG:-0}" = "1" ]; then
+    echo ""
+    log "[teardown] E2E_KEEP_ORG=1 — leaving $SLUG for debugging (REMEMBER TO DELETE)"
+    exit $rc
+  fi
+  echo ""
+  log "[teardown] DELETE /cp/admin/tenants/$SLUG (scoped to this run only)"
+  admin_call DELETE "/cp/admin/tenants/$SLUG" --max-time 120 \
+    -d "{\"confirm\":\"$SLUG\"}" >/dev/null 2>&1
+  for j in $(seq 1 24); do
+    LIST=$(admin_call GET "/cp/admin/orgs?limit=500" 2>/dev/null)
+    LEAK=$(echo "$LIST" | python3 -c "
+import sys, json
+try: d = json.load(sys.stdin)
+except Exception: print(1); sys.exit(0)
+orgs = d if isinstance(d, list) else d.get('orgs', [])
+print(sum(1 for o in orgs if o.get('slug') == '$SLUG' and o.get('instance_status') not in ('purged',) and o.get('status') != 'purged'))
+" 2>/dev/null || echo 1)
+    if [ "$LEAK" = "0" ]; then
+      log "[teardown] ✓ $SLUG purged (after ${j}x5s)"
+      exit $rc
+    fi
+    sleep 5
+  done
+  echo "::warning::[teardown] $SLUG still present after 120s — sweep-stale-e2e-orgs will catch it within MAX_AGE_MINUTES" >&2
+  [ $rc -eq 0 ] && rc=4
+  exit $rc
+}
+trap teardown EXIT INT TERM
+
+# ─── 1. Provision the throwaway org ────────────────────────────────────
+log "1/6 POST /cp/admin/orgs — slug=$SLUG"
+CREATE=$(admin_call POST /cp/admin/orgs \
+  -d "{\"slug\":\"$SLUG\",\"name\":\"E2E peer-visibility $SLUG\",\"owner_user_id\":\"e2e-runner:$SLUG\"}")
+ORG_ID=$(echo "$CREATE" | python3 -c "import sys,json; print(json.load(sys.stdin).get('id',''))" 2>/dev/null)
+[ -n "$ORG_ID" ] || fail "org creation failed: $(echo "$CREATE" | head -c 300)"
+log "    ORG_ID=$ORG_ID"
+
+# ─── 2. Wait for tenant EC2 + DNS ──────────────────────────────────────
+log "2/6 waiting for tenant instance_status=running (cold EC2 + cloudflared)..."
+DEADLINE=$(( $(date +%s) + PROVISION_TIMEOUT_SECS ))
+while true; do
+  [ "$(date +%s)" -gt "$DEADLINE" ] && fail "tenant never came up within ${PROVISION_TIMEOUT_SECS}s"
+  STATUS=$(admin_call GET "/cp/admin/orgs?limit=500" 2>/dev/null | python3 -c "
+import sys, json
+try: d = json.load(sys.stdin)
+except Exception: sys.exit(0)
+orgs = d if isinstance(d, list) else d.get('orgs', [])
+for o in orgs:
+    if o.get('slug') == '$SLUG':
+        print(o.get('instance_status') or o.get('status') or 'unknown'); break
+" 2>/dev/null)
+  case "$STATUS" in running|online|ready) break ;; esac
+  sleep 10
+done
+log "    tenant status=$STATUS"
+
+# ─── 3. Per-tenant admin token + tenant URL ────────────────────────────
+log "3/6 fetching per-tenant admin token..."
+TT_RESP=$(admin_call GET "/cp/admin/orgs/$SLUG/admin-token")
+TENANT_TOKEN=$(echo "$TT_RESP" | python3 -c "import sys,json; print(json.load(sys.stdin).get('admin_token',''))" 2>/dev/null)
+[ -n "$TENANT_TOKEN" ] || fail "tenant token fetch failed: $(echo "$TT_RESP" | head -c 200)"
+
+CP_HOST=$(echo "$CP_URL" | sed -E 's#^https?://##; s#/.*$##')
+case "$CP_HOST" in
+  api.*)         DERIVED_DOMAIN="${CP_HOST#api.}" ;;
+  staging-api.*) DERIVED_DOMAIN="staging.${CP_HOST#staging-api.}" ;;
+  *)             DERIVED_DOMAIN="$CP_HOST" ;;
+esac
+TENANT_URL="https://${SLUG}.${DERIVED_DOMAIN}"
+log "    tenant url: $TENANT_URL"
+
+log "3b. waiting for tenant /health (TLS/DNS, up to 10min)..."
+for i in $(seq 1 120); do
+  curl -fsS "$TENANT_URL/health" -m 5 -k >/dev/null 2>&1 && { log "    /health ok (attempt $i)"; break; }
+  sleep 5
+done
+
+# ─── 4. Provision the parent + one sibling per runtime under test ──────
+# Inject the LLM provider key so each runtime can authenticate at boot.
+# Priority: MiniMax → direct-Anthropic → OpenAI (mirrors
+# test_staging_full_saas.sh's secrets-injection chain).
+SECRETS_JSON='{}'
+if [ -n "${E2E_MINIMAX_API_KEY:-}" ]; then
+  SECRETS_JSON=$(python3 -c "import json,os;k=os.environ['E2E_MINIMAX_API_KEY'];print(json.dumps({'ANTHROPIC_BASE_URL':'https://api.minimax.io/anthropic','ANTHROPIC_AUTH_TOKEN':k,'MINIMAX_API_KEY':k}))")
+elif [ -n "${E2E_ANTHROPIC_API_KEY:-}" ]; then
+  SECRETS_JSON=$(python3 -c "import json,os;k=os.environ['E2E_ANTHROPIC_API_KEY'];print(json.dumps({'ANTHROPIC_API_KEY':k}))")
+elif [ -n "${E2E_OPENAI_API_KEY:-}" ]; then
+  SECRETS_JSON=$(python3 -c "import json,os;k=os.environ['E2E_OPENAI_API_KEY'];print(json.dumps({'OPENAI_API_KEY':k,'OPENAI_BASE_URL':'https://api.openai.com/v1','MODEL_PROVIDER':'openai:gpt-4o','HERMES_INFERENCE_PROVIDER':'custom','HERMES_CUSTOM_BASE_URL':'https://api.openai.com/v1','HERMES_CUSTOM_API_KEY':k,'HERMES_CUSTOM_API_MODE':'chat_completions'}))")
+fi
+
+log "4/6 provisioning parent (claude-code) + one sibling per runtime under test..."
+P_RESP=$(tenant_call POST /workspaces \
+  -d "{\"name\":\"pv-parent\",\"runtime\":\"claude-code\",\"tier\":3,\"secrets\":$SECRETS_JSON}")
+PARENT_ID=$(echo "$P_RESP" | python3 -c "import sys,json; print(json.load(sys.stdin).get('id',''))" 2>/dev/null)
+[ -n "$PARENT_ID" ] || fail "parent create failed: $(echo "$P_RESP" | head -c 300)"
+log "    PARENT_ID=$PARENT_ID"
+
+# WS_IDS[runtime]=id ; WS_TOKENS[runtime]=auth_token (the MCP bearer)
+declare -A WS_IDS WS_TOKENS
+ALL_WS_IDS="$PARENT_ID"
+for rt in $PV_RUNTIMES; do
+  R=$(tenant_call POST /workspaces \
+    -d "{\"name\":\"pv-$rt\",\"runtime\":\"$rt\",\"tier\":2,\"parent_id\":\"$PARENT_ID\",\"secrets\":$SECRETS_JSON}")
+  WID=$(echo "$R" | python3 -c "import sys,json; print(json.load(sys.stdin).get('id',''))" 2>/dev/null)
+  # auth_token is top-level for container runtimes; external-like nest it
+  # under connection.auth_token (verified vs staging response shape).
+  WTOK=$(echo "$R" | python3 -c "
+import sys, json
+try: d = json.load(sys.stdin)
+except Exception: print(''); sys.exit(0)
+print(d.get('auth_token') or d.get('connection', {}).get('auth_token') or '')
+" 2>/dev/null)
+  [ -n "$WID" ] || fail "$rt workspace create failed: $(echo "$R" | head -c 300)"
+  [ -n "$WTOK" ] || fail "$rt workspace did not return an auth_token — cannot drive its MCP call (resp: $(echo "$R" | head -c 300))"
+  WS_IDS[$rt]="$WID"
+  WS_TOKENS[$rt]="$WTOK"
+  ALL_WS_IDS="$ALL_WS_IDS $WID"
+  log "    $rt → $WID"
+done
+
+# ─── 5. Wait for every sibling online ──────────────────────────────────
+log "5/6 waiting for all workspaces status=online (up to ${PROVISION_TIMEOUT_SECS}s — cold boot)..."
+WS_DEADLINE=$(( $(date +%s) + PROVISION_TIMEOUT_SECS ))
+for rt in $PV_RUNTIMES; do
+  wid="${WS_IDS[$rt]}"
+  LAST=""
+  while true; do
+    [ "$(date +%s)" -gt "$WS_DEADLINE" ] && fail "$rt ($wid) never reached online (last=$LAST)"
+    S=$(tenant_call GET "/workspaces/$wid" 2>/dev/null | python3 -c "
+import sys, json
+try: d = json.load(sys.stdin)
+except Exception: sys.exit(0)
+w = d.get('workspace') if isinstance(d.get('workspace'), dict) else d
+print(w.get('status') or '')
+" 2>/dev/null)
+    [ "$S" != "$LAST" ] && { log "    $rt → $S"; LAST="$S"; }
+    case "$S" in
+      online) break ;;
+      failed) sleep 10 ;;   # transient: bootstrap-watcher 5-min deadline, heartbeat recovers
+      *)      sleep 10 ;;
+    esac
+  done
+  ok "    $rt online"
+done
+
+# ─── 6. THE GATE — literal mcp_molecule_list_peers via POST /:id/mcp ────
+# This is the byte-for-byte user-facing call. NOT GET /registry/:id/peers,
+# NOT /health, NOT the heartbeat table. JSON-RPC 2.0 tools/call,
+# name=list_peers, authenticated by the workspace's OWN bearer token
+# through WorkspaceAuth + MCPRateLimiter.
+log "6/6 driving the LITERAL list_peers MCP call per runtime..."
+echo ""
+RPC_BODY='{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_peers","arguments":{}}}'
+REGRESSED=0
+declare -A VERDICT
+
+for rt in $PV_RUNTIMES; do
+  wid="${WS_IDS[$rt]}"
+  wtok="${WS_TOKENS[$rt]}"
+  # The expected peer set = every OTHER provisioned workspace (parent +
+  # the sibling runtimes), excluding the caller itself.
+  EXPECT_IDS=$(echo "$ALL_WS_IDS" | tr ' ' '\n' | grep -v "^${wid}$" | grep -v '^$')
+
+  set +e
+  RESP=$(curl -sS -X POST "$TENANT_URL/workspaces/$wid/mcp" \
+    -H "Authorization: Bearer $wtok" \
+    -H "X-Molecule-Org-Id: $ORG_ID" \
+    -H "Content-Type: application/json" \
+    -d "$RPC_BODY" \
+    -o /tmp/pv_mcp_body.json -w "%{http_code}" 2>/dev/null)
+  set -e
+  HTTP_CODE="$RESP"
+  BODY=$(cat /tmp/pv_mcp_body.json 2>/dev/null || echo '')
+
+  echo "--- $rt (ws=$wid) ---"
+  echo "    HTTP $HTTP_CODE"
+  echo "    body: $(echo "$BODY" | head -c 600)"
+
+  # (1) HTTP 200 — a 401 (WorkspaceAuth reject, the Hermes symptom) fails here.
+  if [ "$HTTP_CODE" != "200" ]; then
+    echo "  ✗ $rt: list_peers MCP call returned HTTP $HTTP_CODE (expected 200)"
+    VERDICT[$rt]="FAIL(http=$HTTP_CODE)"
+    REGRESSED=1
+    continue
+  fi
+
+  # (2) JSON-RPC result present, not an error object.
+  PARSE=$(echo "$BODY" | python3 -c "
+import sys, json
+expect = set(filter(None, '''$EXPECT_IDS'''.split()))
+try:
+    d = json.load(sys.stdin)
+except Exception as e:
+    print('PARSE_ERROR:' + str(e)); sys.exit(0)
+if isinstance(d, dict) and d.get('error') is not None:
+    print('RPC_ERROR:' + json.dumps(d['error'])[:200]); sys.exit(0)
+res = d.get('result') if isinstance(d, dict) else None
+if res is None:
+    print('NO_RESULT'); sys.exit(0)
+# MCP tools/call result shape: {content:[{type:text,text:'<json or prose>'}]}
+text = ''
+if isinstance(res, dict):
+    for c in res.get('content', []):
+        if c.get('type') == 'text':
+            text += c.get('text', '')
+text_l = text.lower()
+# Native-sessions fallback signature (the OpenClaw symptom): the agent
+# answered from its own runtime session list, not the platform peer set.
+if 'sessions_list' in text_l or 'no platform peers' in text_l or 'native session' in text_l:
+    print('NATIVE_FALLBACK:' + text[:200]); sys.exit(0)
+# The expected sibling IDs must literally appear in the returned peer text.
+found = sorted(i for i in expect if i in text)
+missing = sorted(expect - set(found))
+if not expect:
+    print('NO_EXPECTED_PEERS_CONFIGURED'); sys.exit(0)
+if missing:
+    print('MISSING_PEERS:found=%d/%d missing=%s' % (len(found), len(expect), ','.join(m[:8] for m in missing)))
+    sys.exit(0)
+print('OK:found=%d/%d' % (len(found), len(expect)))
+" 2>/dev/null)
+
+  case "$PARSE" in
+    OK:*)
+      echo "  ✓ $rt: list_peers returned 200 and contains all expected peers ($PARSE)"
+      VERDICT[$rt]="OK"
+      ;;
+    NATIVE_FALLBACK:*)
+      echo "  ✗ $rt: list_peers fell back to NATIVE sessions — sees no platform peers ($PARSE)"
+      VERDICT[$rt]="FAIL(native-fallback)"
+      REGRESSED=1
+      ;;
+    RPC_ERROR:*|NO_RESULT|PARSE_ERROR:*)
+      echo "  ✗ $rt: list_peers MCP call did not return a usable result ($PARSE)"
+      VERDICT[$rt]="FAIL(rpc=$PARSE)"
+      REGRESSED=1
+      ;;
+    MISSING_PEERS:*)
+      echo "  ✗ $rt: list_peers returned 200 but peer set is wrong/empty ($PARSE)"
+      VERDICT[$rt]="FAIL(peers=$PARSE)"
+      REGRESSED=1
+      ;;
+    *)
+      echo "  ✗ $rt: unexpected verdict '$PARSE'"
+      VERDICT[$rt]="FAIL(unknown)"
+      REGRESSED=1
+      ;;
+  esac
+  echo ""
+done
+
+echo "=== SUMMARY — fresh-provision peer-visibility (literal MCP list_peers) ==="
+for rt in $PV_RUNTIMES; do
+  printf '  %-14s %s\n' "$rt" "${VERDICT[$rt]:-NO_RUN}"
+done
+echo ""
+
+if [ "$REGRESSED" -ne 0 ]; then
+  echo "✗ GATE FAILED — at least one runtime cannot see its peers via the"
+  echo "  literal mcp_molecule_list_peers call. This is the real user-facing"
+  echo "  failure the proxy signals (registry row / heartbeat / model 200)"
+  echo "  were hiding. Expected RED until the Hermes-401 + OpenClaw-MCP-wiring"
+  echo "  root-cause fixes land; goes green only when they actually do."
+  exit 10
+fi
+
+ok "GATE PASSED — every runtime under test sees its platform peers via the literal MCP call."
+exit 0
@@ -0,0 +1,296 @@
+#!/usr/bin/env bash
+# E2E test: workspace broadcast and talk-to-user platform abilities.
+#
+# What this proves:
+#   1. talk_to_user_enabled (default true) — POST /notify works out-of-the-box.
+#   2. PATCH /workspaces/:id/abilities { talk_to_user_enabled: false } disables
+#      delivery: /notify → 403 with error="talk_to_user_disabled" + delegate hint.
+#   3. Re-enabling talk_to_user_enabled restores delivery.
+#   4. broadcast_enabled (default false) — POST /broadcast → 403 when disabled.
+#   5. PATCH { broadcast_enabled: true } enables fan-out.
+#   6. POST /broadcast delivers to all non-sender, non-removed workspaces:
+#      - Returns {"status":"sent","delivered":N}
+#      - Receiver's activity log has a broadcast_receive entry with the message.
+#      - Sender's activity log has a broadcast_sent entry.
+#   7. The sender itself does NOT receive a broadcast_receive entry.
+#
+# Usage:  tests/e2e/test_workspace_abilities_e2e.sh
+# Prereqs: workspace-server on http://localhost:8080, MOLECULE_ENV != production
+
+set -euo pipefail
+
+source "$(dirname "$0")/_lib.sh"
+
+PASS=0
+FAIL=0
+SENDER_ID=""
+RECEIVER_ID=""
+
+cleanup() {
+  for wid in "$SENDER_ID" "$RECEIVER_ID"; do
+    if [ -n "$wid" ]; then
+      curl -s -X DELETE "$BASE/workspaces/$wid?confirm=true" > /dev/null || true
+    fi
+  done
+}
+trap cleanup EXIT INT TERM
+
+assert() {
+  local label="$1" actual="$2" expected="$3"
+  if [ "$actual" = "$expected" ]; then
+    echo "  PASS — $label"
+    PASS=$((PASS+1))
+  else
+    echo "  FAIL — $label"
+    echo "         expected: $expected"
+    echo "         actual:   $actual"
+    FAIL=$((FAIL+1))
+  fi
+}
+
+assert_contains() {
+  local label="$1" haystack="$2" needle="$3"
+  if echo "$haystack" | grep -qF "$needle"; then
+    echo "  PASS — $label"
+    PASS=$((PASS+1))
+  else
+    echo "  FAIL — $label"
+    echo "         needle:   $needle"
+    echo "         haystack: $haystack"
+    FAIL=$((FAIL+1))
+  fi
+}
+
+assert_not_contains() {
+  local label="$1" haystack="$2" needle="$3"
+  if ! echo "$haystack" | grep -qF "$needle"; then
+    echo "  PASS — $label"
+    PASS=$((PASS+1))
+  else
+    echo "  FAIL — $label (unexpected match)"
+    echo "         needle:   $needle"
+    echo "         haystack: $haystack"
+    FAIL=$((FAIL+1))
+  fi
+}
+
+# ── Pre-sweep: remove any stale leftover workspaces from a prior aborted run ──
+echo "=== Setup ==="
+for NAME in "Abilities Sender" "Abilities Receiver"; do
+  PRIOR=$(curl -s "$BASE/workspaces" | python3 -c "
+import json, sys
+try:
+    print(' '.join(w['id'] for w in json.load(sys.stdin) if w.get('name') == '$NAME'))
+except Exception:
+    pass
+")
+  for _wid in $PRIOR; do
+    echo "Sweeping leftover '$NAME' workspace: $_wid"
+    curl -s -X DELETE "$BASE/workspaces/$_wid?confirm=true" > /dev/null || true
+  done
+done
+
+R=$(curl -s -X POST "$BASE/workspaces" -H "Content-Type: application/json" \
+  -d '{"name":"Abilities Sender","tier":1}')
+SENDER_ID=$(echo "$R" | python3 -c 'import json,sys;print(json.load(sys.stdin)["id"])' 2>/dev/null || true)
+[ -n "$SENDER_ID" ] || { echo "Failed to create sender workspace: $R"; exit 1; }
+echo "Created sender   workspace: $SENDER_ID"
+
+R=$(curl -s -X POST "$BASE/workspaces" -H "Content-Type: application/json" \
+  -d '{"name":"Abilities Receiver","tier":1}')
+RECEIVER_ID=$(echo "$R" | python3 -c 'import json,sys;print(json.load(sys.stdin)["id"])' 2>/dev/null || true)
+[ -n "$RECEIVER_ID" ] || { echo "Failed to create receiver workspace: $R"; exit 1; }
+echo "Created receiver workspace: $RECEIVER_ID"
+
+# Mint workspace-scoped bearer tokens (test-only endpoint, disabled in prod).
+SENDER_TOKEN=$(e2e_mint_test_token "$SENDER_ID")
+[ -n "$SENDER_TOKEN" ] || { echo "Failed to mint sender token"; exit 1; }
+SENDER_AUTH="Authorization: Bearer $SENDER_TOKEN"
+
+# Admin token — any live workspace bearer satisfies AdminAuth in local dev.
+# In production-like envs, set MOLECULE_ADMIN_TOKEN.
+ADMIN_TOKEN="${MOLECULE_ADMIN_TOKEN:-$SENDER_TOKEN}"
+ADMIN_AUTH="Authorization: Bearer $ADMIN_TOKEN"
+
+# ─────────────────────────────────────────────────────────────────────────────
+echo ""
+echo "=== Part 1: talk_to_user ability ==="
+
+echo ""
+echo "--- 1a: /notify works with default talk_to_user_enabled=true ---"
+CODE=$(curl -s -o /dev/null -w "%{http_code}" -X POST "$BASE/workspaces/$SENDER_ID/notify" \
+  -H "Content-Type: application/json" -H "$SENDER_AUTH" \
+  -d '{"message":"Hello from sender"}')
+assert "POST /notify returns 200 when talk_to_user_enabled=true (default)" "$CODE" "200"
+
+echo ""
+echo "--- 1b: Disable talk_to_user ---"
+CODE=$(curl -s -o /dev/null -w "%{http_code}" -X PATCH "$BASE/workspaces/$SENDER_ID/abilities" \
+  -H "Content-Type: application/json" -H "$ADMIN_AUTH" \
+  -d '{"talk_to_user_enabled": false}')
+assert "PATCH /abilities talk_to_user_enabled=false returns 200" "$CODE" "200"
+
+# Verify the flag is reflected in the workspace GET response.
+WS=$(curl -s "$BASE/workspaces/$SENDER_ID" -H "$SENDER_AUTH")
+FLAG=$(echo "$WS" | python3 -c 'import json,sys;print(json.load(sys.stdin).get("talk_to_user_enabled","MISSING"))')
+assert "GET /workspaces/:id reflects talk_to_user_enabled=false" "$FLAG" "False"
+
+echo ""
+echo "--- 1c: /notify blocked when talk_to_user disabled ---"
+BODY=$(curl -s -w "" -X POST "$BASE/workspaces/$SENDER_ID/notify" \
+  -H "Content-Type: application/json" -H "$SENDER_AUTH" \
+  -d '{"message":"Should be blocked"}')
+CODE=$(curl -s -o /dev/null -w "%{http_code}" -X POST "$BASE/workspaces/$SENDER_ID/notify" \
+  -H "Content-Type: application/json" -H "$SENDER_AUTH" \
+  -d '{"message":"Should be blocked"}')
+assert "POST /notify returns 403 when talk_to_user_enabled=false" "$CODE" "403"
+
+ERR=$(echo "$BODY" | python3 -c 'import json,sys;print(json.load(sys.stdin).get("error",""))' 2>/dev/null || echo "")
+assert_contains "403 body contains talk_to_user_disabled error code" "$ERR" "talk_to_user_disabled"
+
+HINT=$(echo "$BODY" | python3 -c 'import json,sys;print(json.load(sys.stdin).get("hint",""))' 2>/dev/null || echo "")
+assert_contains "403 body contains delegate_task hint" "$HINT" "delegate_task"
+
+echo ""
+echo "--- 1d: Re-enable talk_to_user and verify /notify works again ---"
+CODE=$(curl -s -o /dev/null -w "%{http_code}" -X PATCH "$BASE/workspaces/$SENDER_ID/abilities" \
+  -H "Content-Type: application/json" -H "$ADMIN_AUTH" \
+  -d '{"talk_to_user_enabled": true}')
+assert "PATCH /abilities talk_to_user_enabled=true returns 200" "$CODE" "200"
+
+CODE=$(curl -s -o /dev/null -w "%{http_code}" -X POST "$BASE/workspaces/$SENDER_ID/notify" \
+  -H "Content-Type: application/json" -H "$SENDER_AUTH" \
+  -d '{"message":"Re-enabled, should work"}')
+assert "POST /notify returns 200 after re-enabling talk_to_user" "$CODE" "200"
+
+# ─────────────────────────────────────────────────────────────────────────────
+echo ""
+echo "=== Part 2: broadcast ability ==="
+
+echo ""
+echo "--- 2a: Broadcast blocked by default (broadcast_enabled=false) ---"
+CODE=$(curl -s -o /dev/null -w "%{http_code}" -X POST "$BASE/workspaces/$SENDER_ID/broadcast" \
+  -H "Content-Type: application/json" -H "$SENDER_AUTH" \
+  -d '{"message":"Should be blocked"}')
+assert "POST /broadcast returns 403 when broadcast_enabled=false (default)" "$CODE" "403"
+
+echo ""
+echo "--- 2b: Enable broadcast ---"
+CODE=$(curl -s -o /dev/null -w "%{http_code}" -X PATCH "$BASE/workspaces/$SENDER_ID/abilities" \
+  -H "Content-Type: application/json" -H "$ADMIN_AUTH" \
+  -d '{"broadcast_enabled": true}')
+assert "PATCH /abilities broadcast_enabled=true returns 200" "$CODE" "200"
+
+WS=$(curl -s "$BASE/workspaces/$SENDER_ID" -H "$SENDER_AUTH")
+FLAG=$(echo "$WS" | python3 -c 'import json,sys;print(json.load(sys.stdin).get("broadcast_enabled","MISSING"))')
+assert "GET /workspaces/:id reflects broadcast_enabled=true" "$FLAG" "True"
+
+echo ""
+echo "--- 2c: Successful broadcast fan-out ---"
+BCAST=$(curl -s -X POST "$BASE/workspaces/$SENDER_ID/broadcast" \
+  -H "Content-Type: application/json" -H "$SENDER_AUTH" \
+  -d '{"message":"Org-wide notice: scheduled maintenance in 5 minutes."}')
+BSTATUS=$(echo "$BCAST" | python3 -c 'import json,sys;print(json.load(sys.stdin).get("status",""))' 2>/dev/null || echo "")
+BDELIVERED=$(echo "$BCAST" | python3 -c 'import json,sys;print(json.load(sys.stdin).get("delivered","-1"))' 2>/dev/null || echo "-1")
+assert "POST /broadcast returns status=sent" "$BSTATUS" "sent"
+
+# delivered count must be >= 1 (the receiver workspace).
+echo "  INFO — broadcast delivered=$BDELIVERED"
+if python3 -c "import sys; sys.exit(0 if int('$BDELIVERED') >= 1 else 1)" 2>/dev/null; then
+  echo "  PASS — delivered count >= 1"
+  PASS=$((PASS+1))
+else
+  echo "  FAIL — expected delivered >= 1, got $BDELIVERED"
+  FAIL=$((FAIL+1))
+fi
+
+echo ""
+echo "--- 2d: Receiver activity log has broadcast_receive entry ---"
+RECEIVER_TOKEN=$(e2e_mint_test_token "$RECEIVER_ID")
+[ -n "$RECEIVER_TOKEN" ] || { echo "Failed to mint receiver token"; exit 1; }
+RECEIVER_AUTH="Authorization: Bearer $RECEIVER_TOKEN"
+
+ACT=$(curl -s -H "$RECEIVER_AUTH" "$BASE/workspaces/$RECEIVER_ID/activity?source=agent&limit=20")
+ROW=$(echo "$ACT" | python3 -c '
+import json, sys
+rows = json.load(sys.stdin) or []
+for r in rows:
+    if r.get("activity_type") == "broadcast_receive":
+        print(json.dumps(r))
+        break
+')
+[ -n "$ROW" ] || {
+  echo "  FAIL — could not find broadcast_receive row in receiver activity"
+  FAIL=$((FAIL+1))
+}
+
+if [ -n "$ROW" ]; then
+  # Message is stored in summary field.
+  MSG=$(echo "$ROW" | python3 -c 'import json,sys;r=json.load(sys.stdin);print(r.get("summary",""))')
+  assert_contains "broadcast_receive row summary has original message" "$MSG" "scheduled maintenance"
+  # Sender ID is stored in source_id field.
+  SRC=$(echo "$ROW" | python3 -c 'import json,sys;r=json.load(sys.stdin);print(r.get("source_id",""))')
+  assert "broadcast_receive row source_id is sender workspace" "$SRC" "$SENDER_ID"
+fi
+
+echo ""
+echo "--- 2e: Sender activity log has broadcast_sent entry ---"
+ACT_SENDER=$(curl -s -H "$SENDER_AUTH" "$BASE/workspaces/$SENDER_ID/activity?limit=20")
+SENT_ROW=$(echo "$ACT_SENDER" | python3 -c '
+import json, sys
+rows = json.load(sys.stdin) or []
+for r in rows:
+    if r.get("activity_type") == "broadcast_sent":
+        print(json.dumps(r))
+        break
+')
+[ -n "$SENT_ROW" ] || {
+  echo "  FAIL — could not find broadcast_sent row in sender activity"
+  FAIL=$((FAIL+1))
+}
+
+if [ -n "$SENT_ROW" ]; then
+  # Delivered count is baked into the summary field (no response_body for sender row).
+  SUMMARY=$(echo "$SENT_ROW" | python3 -c 'import json,sys;print(json.load(sys.stdin).get("summary",""))')
+  assert_contains "broadcast_sent summary mentions workspace count" "$SUMMARY" "workspace"
+fi
+
+echo ""
+echo "--- 2f: Sender does NOT receive a broadcast_receive entry ---"
+SELF_RECV=$(echo "$ACT_SENDER" | python3 -c '
+import json, sys
+rows = json.load(sys.stdin) or []
+for r in rows:
+    if r.get("activity_type") == "broadcast_receive":
+        print("found")
+        break
+')
+assert_not_contains "sender has no broadcast_receive in own activity log" "${SELF_RECV:-}" "found"
+
+# ─────────────────────────────────────────────────────────────────────────────
+echo ""
+echo "--- 2g: Empty message is rejected ---"
+CODE=$(curl -s -o /dev/null -w "%{http_code}" -X POST "$BASE/workspaces/$SENDER_ID/broadcast" \
+  -H "Content-Type: application/json" -H "$SENDER_AUTH" \
+  -d '{"message":""}')
+assert "POST /broadcast with empty message returns 400" "$CODE" "400"
+
+echo ""
+echo "--- 2h: Partial PATCH does not clobber other flags ---"
+# Set talk_to_user=false, then patch only broadcast — talk_to_user must stay false.
+curl -s -o /dev/null -X PATCH "$BASE/workspaces/$SENDER_ID/abilities" \
+  -H "Content-Type: application/json" -H "$ADMIN_AUTH" \
+  -d '{"talk_to_user_enabled": false}'
+curl -s -o /dev/null -X PATCH "$BASE/workspaces/$SENDER_ID/abilities" \
+  -H "Content-Type: application/json" -H "$ADMIN_AUTH" \
+  -d '{"broadcast_enabled": false}'
+WS=$(curl -s "$BASE/workspaces/$SENDER_ID" -H "$SENDER_AUTH")
+TUF=$(echo "$WS" | python3 -c 'import json,sys;print(json.load(sys.stdin).get("talk_to_user_enabled","MISSING"))')
+BEF=$(echo "$WS" | python3 -c 'import json,sys;print(json.load(sys.stdin).get("broadcast_enabled","MISSING"))')
+assert "partial PATCH preserves talk_to_user_enabled=false" "$TUF" "False"
+assert "partial PATCH sets broadcast_enabled=false" "$BEF" "False"
+
+# ─────────────────────────────────────────────────────────────────────────────
+echo ""
+echo "=== Results: $PASS passed, $FAIL failed ==="
+[ "$FAIL" -eq 0 ]
@@ -482,6 +482,13 @@ func (h *ActivityHandler) Notify(c *gin.Context) {
 			c.JSON(http.StatusNotFound, gin.H{"error": "workspace not found"})
 			return
 		}
+		if errors.Is(err, ErrTalkToUserDisabled) {
+			c.JSON(http.StatusForbidden, gin.H{
+				"error": "talk_to_user_disabled",
+				"hint":  "This workspace is not allowed to send messages directly to the user. Forward your update to a parent workspace using delegate_task — they may be able to reach the user.",
+			})
+			return
+		}
 		c.JSON(http.StatusInternalServerError, gin.H{"error": "internal error"})
 		return
 	}
@@ -464,9 +464,9 @@ func TestNotify_PersistsToActivityLogsForReloadRecovery(t *testing.T) {
 	t.Cleanup(func() { db.DB = prevDB; mockDB.Close() })

 	// Workspace existence check
-	mock.ExpectQuery(`SELECT name FROM workspaces`).
+	mock.ExpectQuery(`SELECT name, talk_to_user_enabled FROM workspaces`).
 		WithArgs("ws-notify").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("DD"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("DD", true))

 	// Persistence INSERT — verify shape
 	mock.ExpectExec(`INSERT INTO activity_logs`).
@@ -511,9 +511,9 @@ func TestNotify_WithAttachments_PersistsFilePartsForReload(t *testing.T) {
 	db.DB = mockDB
 	t.Cleanup(func() { db.DB = prevDB; mockDB.Close() })

-	mock.ExpectQuery(`SELECT name FROM workspaces`).
+	mock.ExpectQuery(`SELECT name, talk_to_user_enabled FROM workspaces`).
 		WithArgs("ws-attach").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("DD"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("DD", true))

 	// Capture the JSONB arg so we can assert on the persisted shape
 	// AFTER the call (must include parts[].kind=file so reload
@@ -640,9 +640,9 @@ func TestNotify_DBFailure_StillBroadcastsAnd200(t *testing.T) {
 	db.DB = mockDB
 	t.Cleanup(func() { db.DB = prevDB; mockDB.Close() })

-	mock.ExpectQuery(`SELECT name FROM workspaces`).
+	mock.ExpectQuery(`SELECT name, talk_to_user_enabled FROM workspaces`).
 		WithArgs("ws-x").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("DD"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("DD", true))
 	mock.ExpectExec(`INSERT INTO activity_logs`).
 		WillReturnError(fmt.Errorf("simulated db hiccup"))

@@ -54,6 +54,11 @@ import (
 // timeout) surface as wrapped errors and should be treated as 503.
 var ErrWorkspaceNotFound = errors.New("agent_message: workspace not found")

+// ErrTalkToUserDisabled is returned when the workspace has
+// talk_to_user_enabled=false. Callers surface HTTP 403 so the Python tool
+// can detect it and suggest forwarding to a parent workspace.
+var ErrTalkToUserDisabled = errors.New("agent_message: talk_to_user disabled")
+
 // AgentMessageAttachment is one file attached to an agent → user
 // message. Identical to handlers.NotifyAttachment in field set; kept
 // distinct so the writer's API doesn't import a handler type with HTTP
@@ -107,16 +112,20 @@ func (w *AgentMessageWriter) Send(
 	// notify call surfaced as "workspace not found" and masked real
 	// incidents in the alert path.
 	var wsName string
+	var talkToUserEnabled bool
 	err := w.db.QueryRowContext(ctx,
-		`SELECT name FROM workspaces WHERE id = $1 AND status != 'removed'`,
+		`SELECT name, talk_to_user_enabled FROM workspaces WHERE id = $1 AND status != 'removed'`,
 		workspaceID,
-	).Scan(&wsName)
+	).Scan(&wsName, &talkToUserEnabled)
 	if errors.Is(err, sql.ErrNoRows) {
 		return ErrWorkspaceNotFound
 	}
 	if err != nil {
 		return fmt.Errorf("agent_message: workspace lookup: %w", err)
 	}
+	if !talkToUserEnabled {
+		return ErrTalkToUserDisabled
+	}

 	// 2. Build broadcast payload + WS-emit. Same shape that ChatTab's
 	//    AGENT_MESSAGE handler in canvas/src/store/canvas-events.ts has
@@ -88,9 +88,9 @@ func TestAgentMessageWriter_Send_Success_NoAttachments(t *testing.T) {
 	mock := setupTestDB(t)
 	w := NewAgentMessageWriter(db.DB, newTestBroadcaster())

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-1").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("CEO Ryan PC"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("CEO Ryan PC", true))

 	mock.ExpectExec(`INSERT INTO activity_logs.*'a2a_receive'.*'notify'`).
 		WithArgs(
@@ -116,9 +116,9 @@ func TestAgentMessageWriter_Send_Success_WithAttachments(t *testing.T) {
 	mock := setupTestDB(t)
 	w := NewAgentMessageWriter(db.DB, newTestBroadcaster())

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-att").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("Ryan"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("Ryan", true))

 	mock.ExpectExec(`INSERT INTO activity_logs.*'a2a_receive'.*'notify'`).
 		WithArgs(
@@ -173,9 +173,9 @@ func TestAgentMessageWriter_Send_WorkspaceNotFound(t *testing.T) {
 	emitter := &capturingEmitter{}
 	w := NewAgentMessageWriter(db.DB, emitter)

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-missing").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}))

 	err := w.Send(context.Background(), "ws-missing", "lost in the void", nil)
 	if !errors.Is(err, ErrWorkspaceNotFound) {
@@ -202,9 +202,9 @@ func TestAgentMessageWriter_Send_DBInsertFailureStillReturnsNil(t *testing.T) {
 	mock := setupTestDB(t)
 	w := NewAgentMessageWriter(db.DB, newTestBroadcaster())

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-dbfail").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("CEO Ryan PC"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("CEO Ryan PC", true))

 	mock.ExpectExec(`INSERT INTO activity_logs`).
 		WillReturnError(errors.New("transient db error"))
@@ -223,9 +223,9 @@ func TestAgentMessageWriter_Send_PreviewTruncation(t *testing.T) {
 	mock := setupTestDB(t)
 	w := NewAgentMessageWriter(db.DB, newTestBroadcaster())

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-trunc").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("Ryan"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("Ryan", true))

 	longMsg := strings.Repeat("x", 200)
 	mock.ExpectExec(`INSERT INTO activity_logs`).
@@ -263,9 +263,9 @@ func TestAgentMessageWriter_Send_BroadcastsAgentMessageEvent(t *testing.T) {
 	emitter := &capturingEmitter{}
 	w := NewAgentMessageWriter(db.DB, emitter)

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-bc").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("Workspace Name"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("Workspace Name", true))
 	mock.ExpectExec(`INSERT INTO activity_logs`).
 		WillReturnResult(sqlmock.NewResult(1, 1))

@@ -315,7 +315,7 @@ func TestAgentMessageWriter_Send_DBErrorOnLookupReturnsWrapped(t *testing.T) {
 	w := NewAgentMessageWriter(db.DB, newTestBroadcaster())

 	transientErr := errors.New("connection refused")
-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-dbdown").
 		WillReturnError(transientErr)

@@ -350,9 +350,9 @@ func TestAgentMessageWriter_Send_NonASCIIMessagePersists(t *testing.T) {
 	// the byte-slice bug.
 	msg := strings.Repeat("你", 200)

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-cjk").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("CEO Ryan PC"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("CEO Ryan PC", true))

 	mock.ExpectExec(`INSERT INTO activity_logs`).
 		WithArgs(
@@ -395,9 +395,9 @@ func TestAgentMessageWriter_Send_OmitsAttachmentsKeyWhenEmpty(t *testing.T) {
 	emitter := &capturingEmitter{}
 	w := NewAgentMessageWriter(db.DB, emitter)

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-noatt").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("X"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("X", true))
 	mock.ExpectExec(`INSERT INTO activity_logs`).
 		WillReturnResult(sqlmock.NewResult(1, 1))

@@ -60,9 +60,7 @@ func (h *ApprovalsHandler) Create(c *gin.Context) {

 	// Auto-escalate to parent
 	var parentID *string
-	if err := db.DB.QueryRowContext(ctx, `SELECT parent_id FROM workspaces WHERE id = $1`, workspaceID).Scan(&parentID); err != nil {
-		log.Printf("CreateApproval parent lookup error workspace=%s: %v", workspaceID, err)
-	}
+	db.DB.QueryRowContext(ctx, `SELECT parent_id FROM workspaces WHERE id = $1`, workspaceID).Scan(&parentID)
 	if parentID != nil {
 		h.broadcaster.RecordAndBroadcast(ctx, string(events.EventApprovalEscalated), *parentID, map[string]interface{}{
 			"approval_id":       approvalID,
@@ -82,12 +80,10 @@ func (h *ApprovalsHandler) ListAll(c *gin.Context) {
 	ctx := c.Request.Context()

 	// Auto-expire stale approvals (older than 10 min)
-	if _, err := db.DB.ExecContext(ctx, `
+	db.DB.ExecContext(ctx, `
 		UPDATE approval_requests SET status = 'denied', decided_by = 'auto-expired', decided_at = now()
 		WHERE status = 'pending' AND created_at < now() - interval '10 minutes'
-	`); err != nil {
-		log.Printf("ListAll auto-expire error: %v", err)
-	}
+	`)

 	rows, err := db.DB.QueryContext(ctx, `
 		SELECT a.id, a.workspace_id, w.name, a.action, a.reason, a.status, a.created_at
@@ -646,8 +646,12 @@ const externalOpenClawTemplate = `# OpenClaw MCP config — outbound tool path.
 # external machine today, pair with the Python SDK tab.

 # 1. Install openclaw CLI + the workspace runtime wheel:
+#    The version pin (>=0.1.999) ensures the "molecule-mcp" console
+#    script is present — it is what keeps the workspace ALIVE on canvas
+#    (register-on-startup + 20s heartbeat). Older versions only ship
+#    a2a_mcp_server which does not heartbeat.
 npm install -g openclaw@latest
-pip install molecule-ai-workspace-runtime
+pip install "molecule-ai-workspace-runtime>=0.1.999"

 # 2. Onboard openclaw against your model provider (one-time setup).
 #    --non-interactive needs an explicit --provider + --model so it
@@ -230,20 +230,21 @@ func TestWorkspaceList_WithData(t *testing.T) {
 	broadcaster := newTestBroadcaster()
 	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())

-	// 21 cols — see scanWorkspaceRow for order (max_concurrent_tasks
-	// lands between active_tasks and last_error_rate).
+	// 23 cols — broadcast_enabled + talk_to_user_enabled added after monthly_spend
+	// (migration 20260514). Column order must match scanWorkspaceRow exactly.
 	columns := []string{
 		"id", "name", "role", "tier", "status", "agent_card", "url",
 		"parent_id", "active_tasks", "max_concurrent_tasks",
 		"last_error_rate", "last_sample_error",
 		"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 		"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 	}
 	rows := sqlmock.NewRows(columns).
 		AddRow("ws-1", "Agent One", "worker", 1, "online", []byte(`{"name":"agent1"}`), "http://localhost:8001",
-			nil, 3, 1, 0.02, "", 7200, "processing", "langgraph", "", 10.0, 20.0, false, nil, int64(0)).
+			nil, 3, 1, 0.02, "", 7200, "processing", "langgraph", "", 10.0, 20.0, false, nil, int64(0), false, true).
 		AddRow("ws-2", "Agent Two", "", 2, "degraded", []byte("null"), "",
-			nil, 0, 1, 0.6, "timeout", 100, "", "claude-code", "", 50.0, 60.0, true, nil, int64(0))
+			nil, 0, 1, 0.6, "timeout", 100, "", "claude-code", "", 50.0, 60.0, true, nil, int64(0), false, true)

 	mock.ExpectQuery("SELECT w.id, w.name").
 		WillReturnRows(rows)
@@ -407,21 +407,21 @@ func TestWorkspaceList(t *testing.T) {
 	broadcaster := newTestBroadcaster()
 	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", "/tmp/configs")

-	// 21 cols: `max_concurrent_tasks` added between active_tasks and
-	// last_error_rate (see scanWorkspaceRow + COALESCE(w.max_concurrent_tasks, 1)
-	// in workspace.go). Column order must match that scan exactly.
+	// 23 cols: broadcast_enabled + talk_to_user_enabled added after monthly_spend
+	// (migration 20260514). Column order must match scanWorkspaceRow exactly.
 	columns := []string{
 		"id", "name", "role", "tier", "status", "agent_card", "url",
 		"parent_id", "active_tasks", "max_concurrent_tasks",
 		"last_error_rate", "last_sample_error",
 		"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 		"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 	}
 	rows := sqlmock.NewRows(columns).
 		AddRow("ws-1", "Agent One", "worker", 1, "online", []byte("null"), "http://localhost:8001",
-			nil, 0, 1, 0.0, "", 100, "", "claude-code", "", 10.0, 20.0, false, nil, int64(0)).
+			nil, 0, 1, 0.0, "", 100, "", "claude-code", "", 10.0, 20.0, false, nil, int64(0), false, true).
 		AddRow("ws-2", "Agent Two", "manager", 2, "provisioning", []byte("null"), "",
-			nil, 0, 1, 0.0, "", 0, "", "langgraph", "", 50.0, 60.0, false, nil, int64(0))
+			nil, 0, 1, 0.0, "", 0, "", "langgraph", "", 50.0, 60.0, false, nil, int64(0), false, true)

 	mock.ExpectQuery("SELECT w.id, w.name").
 		WillReturnRows(rows)
@@ -1135,13 +1135,14 @@ func TestWorkspaceGet_CurrentTask(t *testing.T) {
 		"parent_id", "active_tasks", "max_concurrent_tasks", "last_error_rate", "last_sample_error",
 		"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 		"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 	}
 	mock.ExpectQuery("SELECT w.id, w.name").
 		WithArgs("dddddddd-0004-0000-0000-000000000000").
 		WillReturnRows(sqlmock.NewRows(columns).AddRow(
 			"dddddddd-0004-0000-0000-000000000000", "Task Worker", "worker", 1, "online", []byte("null"), "http://localhost:9000",
 			nil, 2, 1, 0.0, "", 300, "Analyzing document", "langgraph", "", 10.0, 20.0, false,
-			nil, int64(0),
+			nil, int64(0), false, true,
 		))

 	w := httptest.NewRecorder()
@@ -248,9 +248,6 @@ func (h *InstructionsHandler) Resolve(c *gin.Context) {
 		b.WriteString(content)
 		b.WriteString("\n\n")
 	}
-	if rowsErr := rows.Err(); rowsErr != nil {
-		log.Printf("ResolveInstructions rows.Err workspace=%s: %v", workspaceID, rowsErr)
-	}

 	c.JSON(http.StatusOK, gin.H{
 		"workspace_id": workspaceID,
@@ -261,7 +258,6 @@ func (h *InstructionsHandler) Resolve(c *gin.Context) {
 func scanInstructions(rows interface {
 	Next() bool
 	Scan(dest ...interface{}) error
-	Err() error
 }) []Instruction {
 	var instructions []Instruction
 	for rows.Next() {
@@ -273,9 +269,6 @@ func scanInstructions(rows interface {
 		}
 		instructions = append(instructions, inst)
 	}
-	if scanErr := rows.Err(); scanErr != nil {
-		log.Printf("scanInstructions rows.Err: %v", scanErr)
-	}
 	if instructions == nil {
 		instructions = []Instruction{}
 	}
@@ -751,9 +751,9 @@ func TestMCPHandler_SendMessageToUser_DBErrorLogsAndStill200s(t *testing.T) {
 	t.Setenv("MOLECULE_MCP_ALLOW_SEND_MESSAGE", "true")
 	h, mock := newMCPHandler(t)

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-err").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("CEO Ryan PC"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("CEO Ryan PC", true))

 	// INSERT fails — must NOT abort the tool response.
 	mock.ExpectExec(`INSERT INTO activity_logs.*'a2a_receive'.*'notify'`).
@@ -802,9 +802,9 @@ func TestMCPHandler_SendMessageToUser_ResponseBodyShape(t *testing.T) {

 	const userMessage = "Hi there from the agent"

-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-shape").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("CEO Ryan PC"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("CEO Ryan PC", true))

 	// Capture the response_body argument and assert its exact shape.
 	mock.ExpectExec(`INSERT INTO activity_logs.*'a2a_receive'.*'notify'`).
@@ -861,9 +861,9 @@ func TestMCPHandler_SendMessageToUser_PersistsToActivityLog(t *testing.T) {
 	// before it does anything else. Returning a name lets the
 	// broadcast payload populate; the test doesn't assert on the
 	// broadcast (no observable WS in this fake), only on the DB.
-	mock.ExpectQuery("SELECT name FROM workspaces").
+	mock.ExpectQuery("SELECT name, talk_to_user_enabled FROM workspaces").
 		WithArgs("ws-msg").
-		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("CEO Ryan PC"))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "talk_to_user_enabled"}).AddRow("CEO Ryan PC", true))

 	// The persistence INSERT — pin the exact shape so a future
 	// refactor that switches columns or drops `method='notify'`
@@ -177,7 +177,7 @@ func expandEnvRef(key, ref, whole string, env map[string]string) string {
 }


-// loadWorkspaceEnv reads the org root .env and the workspace-specific .env .env and the workspace-specific .env
+// loadWorkspaceEnv reads the org root .env and the workspace-specific .env
 // (workspace overrides org root). Used by both secret injection and channel
 // config expansion.
 //
@@ -109,11 +109,9 @@ func (h *TerminalHandler) HandleConnect(c *gin.Context) {
 	// provisionWorkspaceCP → migration 038). Null instance_id means the
 	// workspace runs as a local Docker container on this tenant.
 	var instanceID string
-	if err := db.DB.QueryRowContext(ctx,
+	db.DB.QueryRowContext(ctx,
 		`SELECT COALESCE(instance_id, '') FROM workspaces WHERE id = $1`,
-		workspaceID).Scan(&instanceID); err != nil {
-		log.Printf("HandleConnect instanceID lookup error workspace=%s: %v", workspaceID, err)
-	}
+		workspaceID).Scan(&instanceID)

 	if instanceID != "" {
 		h.handleRemoteConnect(c, workspaceID, instanceID)
@@ -146,9 +144,7 @@ func (h *TerminalHandler) handleLocalConnect(c *gin.Context, workspaceID string)
 	// Look up workspace name for manual container naming
 	var wsName string
 	if _, err := h.docker.Ping(ctx); err == nil {
-		if err := db.DB.QueryRowContext(ctx, `SELECT LOWER(REPLACE(name, ' ', '-')) FROM workspaces WHERE id = $1`, workspaceID).Scan(&wsName); err != nil {
-			log.Printf("handleLocalConnect wsName lookup error workspace=%s: %v", workspaceID, err)
-		}
+		db.DB.QueryRowContext(ctx, `SELECT LOWER(REPLACE(name, ' ', '-')) FROM workspaces WHERE id = $1`, workspaceID).Scan(&wsName)
 		if wsName != "" {
 			candidates = append(candidates, wsName)
 		}
@@ -591,7 +591,7 @@ func scanWorkspaceRow(rows interface {
 	var id, name, role, status, url, sampleError, currentTask, runtime, workspaceDir string
 	var tier, activeTasks, maxConcurrentTasks, uptimeSeconds int
 	var errorRate, x, y float64
-	var collapsed bool
+	var collapsed, broadcastEnabled, talkToUserEnabled bool
 	var parentID *string
 	var agentCard []byte
 	var budgetLimit sql.NullInt64
@@ -600,7 +600,7 @@ func scanWorkspaceRow(rows interface {
 	err := rows.Scan(&id, &name, &role, &tier, &status, &agentCard, &url,
 		&parentID, &activeTasks, &maxConcurrentTasks, &errorRate, &sampleError, &uptimeSeconds,
 		&currentTask, &runtime, &workspaceDir, &x, &y, &collapsed,
-		&budgetLimit, &monthlySpend)
+		&budgetLimit, &monthlySpend, &broadcastEnabled, &talkToUserEnabled)
 	if err != nil {
 		return nil, err
 	}
@@ -624,6 +624,8 @@ func scanWorkspaceRow(rows interface {
 		"x":                    x,
 		"y":                    y,
 		"collapsed":            collapsed,
+		"broadcast_enabled":    broadcastEnabled,
+		"talk_to_user_enabled": talkToUserEnabled,
 	}

 	// budget_limit: nil when no limit set, int64 otherwise
@@ -659,7 +661,8 @@ const workspaceListQuery = `
 		   COALESCE(w.current_task, ''), COALESCE(w.runtime, 'langgraph'),
 		   COALESCE(w.workspace_dir, ''),
 		   COALESCE(cl.x, 0), COALESCE(cl.y, 0), COALESCE(cl.collapsed, false),
-		   w.budget_limit, COALESCE(w.monthly_spend, 0)
+		   w.budget_limit, COALESCE(w.monthly_spend, 0),
+		   w.broadcast_enabled, w.talk_to_user_enabled
 	FROM workspaces w
 	LEFT JOIN canvas_layouts cl ON cl.workspace_id = w.id
 	WHERE w.status != 'removed'
@@ -719,7 +722,8 @@ func (h *WorkspaceHandler) Get(c *gin.Context) {
 			   COALESCE(w.current_task, ''), COALESCE(w.runtime, 'langgraph'),
 			   COALESCE(w.workspace_dir, ''),
 			   COALESCE(cl.x, 0), COALESCE(cl.y, 0), COALESCE(cl.collapsed, false),
-			   w.budget_limit, COALESCE(w.monthly_spend, 0)
+			   w.budget_limit, COALESCE(w.monthly_spend, 0),
+			   w.broadcast_enabled, w.talk_to_user_enabled
 		FROM workspaces w
 		LEFT JOIN canvas_layouts cl ON cl.workspace_id = w.id
 		WHERE w.id = $1
@@ -0,0 +1,82 @@
+package handlers
+
+// workspace_abilities.go — PATCH /workspaces/:id/abilities
+//
+// Allows users and admin agents to toggle two workspace-level ability flags:
+//
+//   broadcast_enabled   — workspace may POST /broadcast to send org-wide messages
+//   talk_to_user_enabled — workspace may deliver canvas chat messages via
+//                          send_message_to_user / POST /notify
+//
+// Gated behind AdminAuth so workspace agents cannot self-modify their own
+// ability flags (that would let any agent grant itself broadcast rights or
+// suppress its own chat-silence constraint).
+
+import (
+	"log"
+	"net/http"
+
+	"github.com/Molecule-AI/molecule-monorepo/platform/internal/db"
+	"github.com/gin-gonic/gin"
+)
+
+// AbilitiesPayload carries the subset of ability flags the caller wants to
+// update. Fields are pointers so that the handler can distinguish "caller
+// supplied false" from "caller omitted the field" (omitempty semantics).
+type AbilitiesPayload struct {
+	BroadcastEnabled   *bool `json:"broadcast_enabled"`
+	TalkToUserEnabled  *bool `json:"talk_to_user_enabled"`
+}
+
+// PatchAbilities handles PATCH /workspaces/:id/abilities (AdminAuth).
+func PatchAbilities(c *gin.Context) {
+	id := c.Param("id")
+	if err := validateWorkspaceID(id); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid workspace ID"})
+		return
+	}
+
+	var body AbilitiesPayload
+	if err := c.ShouldBindJSON(&body); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid request body"})
+		return
+	}
+	if body.BroadcastEnabled == nil && body.TalkToUserEnabled == nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "at least one ability field required"})
+		return
+	}
+
+	ctx := c.Request.Context()
+
+	var exists bool
+	if err := db.DB.QueryRowContext(ctx,
+		`SELECT EXISTS(SELECT 1 FROM workspaces WHERE id = $1 AND status != 'removed')`, id,
+	).Scan(&exists); err != nil || !exists {
+		c.JSON(http.StatusNotFound, gin.H{"error": "workspace not found"})
+		return
+	}
+
+	if body.BroadcastEnabled != nil {
+		if _, err := db.DB.ExecContext(ctx,
+			`UPDATE workspaces SET broadcast_enabled = $2, updated_at = now() WHERE id = $1`,
+			id, *body.BroadcastEnabled,
+		); err != nil {
+			log.Printf("PatchAbilities broadcast_enabled for %s: %v", id, err)
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "update failed"})
+			return
+		}
+	}
+
+	if body.TalkToUserEnabled != nil {
+		if _, err := db.DB.ExecContext(ctx,
+			`UPDATE workspaces SET talk_to_user_enabled = $2, updated_at = now() WHERE id = $1`,
+			id, *body.TalkToUserEnabled,
+		); err != nil {
+			log.Printf("PatchAbilities talk_to_user_enabled for %s: %v", id, err)
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "update failed"})
+			return
+		}
+	}
+
+	c.JSON(http.StatusOK, gin.H{"status": "updated"})
+}
@@ -33,6 +33,7 @@ var wsColumns = []string{
 	"parent_id", "active_tasks", "max_concurrent_tasks", "last_error_rate", "last_sample_error",
 	"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 	"budget_limit", "monthly_spend",
+	"broadcast_enabled", "talk_to_user_enabled",
 }

 // ==================== GET — financial fields stripped from open endpoint ====================
@@ -52,8 +53,10 @@ func TestWorkspaceBudget_Get_NilLimit(t *testing.T) {
 				[]byte(`{}`), "http://localhost:9001",
 				nil, 0, 1, 0.0, "", 0, "", "langgraph", "",
 				0.0, 0.0, false,
-				nil, // budget_limit NULL
-				0))  // monthly_spend 0
+				nil,   // budget_limit NULL
+				0,     // monthly_spend 0
+				false, // broadcast_enabled
+				true)) // talk_to_user_enabled

 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
@@ -96,7 +99,8 @@ func TestWorkspaceBudget_Get_WithLimit(t *testing.T) {
 				nil, 0, 1, 0.0, "", 0, "", "langgraph", "",
 				0.0, 0.0, false,
 				int64(500),  // budget_limit = $5.00 in DB
-				int64(123))) // monthly_spend = $1.23 in DB
+				int64(123),  // monthly_spend = $1.23 in DB
+				false, true)) // broadcast_enabled, talk_to_user_enabled

 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
@@ -29,6 +29,7 @@ func TestWorkspaceGet_Success(t *testing.T) {
 		"parent_id", "active_tasks", "max_concurrent_tasks", "last_error_rate", "last_sample_error",
 		"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 		"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 	}
 	mock.ExpectQuery("SELECT w.id, w.name").
 		WithArgs("cccccccc-0001-0000-0000-000000000000").
@@ -36,7 +37,7 @@ func TestWorkspaceGet_Success(t *testing.T) {
 			AddRow("cccccccc-0001-0000-0000-000000000000", "My Agent", "worker", 1, "online", []byte(`{"name":"test"}`),
 				"http://localhost:8001", nil, 2, 1, 0.05, "", 3600, "working", "langgraph",
 				"", 10.0, 20.0, false,
-				nil, 0))
+				nil, 0, false, true))

 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
@@ -118,6 +119,7 @@ func TestWorkspaceGet_RemovedReturns410(t *testing.T) {
 		"parent_id", "active_tasks", "max_concurrent_tasks", "last_error_rate", "last_sample_error",
 		"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 		"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 	}
 	mock.ExpectQuery("SELECT w.id, w.name").
 		WithArgs(id).
@@ -125,7 +127,7 @@ func TestWorkspaceGet_RemovedReturns410(t *testing.T) {
 			AddRow(id, "Old Agent", "worker", 1, string(models.StatusRemoved), []byte(`null`),
 				"", nil, 0, 1, 0.0, "", 0, "", "langgraph",
 				"", 0.0, 0.0, false,
-				nil, 0))
+				nil, 0, false, true))
 	mock.ExpectQuery(`SELECT updated_at FROM workspaces`).
 		WithArgs(id).
 		WillReturnRows(sqlmock.NewRows([]string{"updated_at"}).AddRow(removedAt))
@@ -181,6 +183,7 @@ func TestWorkspaceGet_RemovedReturns410WithNullRemovedAtOnTimestampFetchFailure(
 		"parent_id", "active_tasks", "max_concurrent_tasks", "last_error_rate", "last_sample_error",
 		"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 		"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 	}
 	mock.ExpectQuery("SELECT w.id, w.name").
 		WithArgs(id).
@@ -188,7 +191,7 @@ func TestWorkspaceGet_RemovedReturns410WithNullRemovedAtOnTimestampFetchFailure(
 			AddRow(id, "Vanished", "worker", 1, string(models.StatusRemoved), []byte(`null`),
 				"", nil, 0, 1, 0.0, "", 0, "", "langgraph",
 				"", 0.0, 0.0, false,
-				nil, 0))
+				nil, 0, false, true))
 	// Simulate the row vanishing between the two queries.
 	mock.ExpectQuery(`SELECT updated_at FROM workspaces`).
 		WithArgs(id).
@@ -243,6 +246,7 @@ func TestWorkspaceGet_RemovedWithIncludeQueryReturns200(t *testing.T) {
 		"parent_id", "active_tasks", "max_concurrent_tasks", "last_error_rate", "last_sample_error",
 		"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 		"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 	}
 	mock.ExpectQuery("SELECT w.id, w.name").
 		WithArgs(id).
@@ -250,7 +254,7 @@ func TestWorkspaceGet_RemovedWithIncludeQueryReturns200(t *testing.T) {
 			AddRow(id, "Audit Agent", "worker", 1, string(models.StatusRemoved), []byte(`null`),
 				"", nil, 0, 1, 0.0, "", 0, "", "langgraph",
 				"", 0.0, 0.0, false,
-				nil, 0))
+				nil, 0, false, true))
 	// last_outbound_at follow-up query (existing path)
 	mock.ExpectQuery(`SELECT last_outbound_at FROM workspaces`).
 		WithArgs(id).
@@ -714,6 +718,7 @@ func TestWorkspaceList_Empty(t *testing.T) {
 			"parent_id", "active_tasks", "last_error_rate", "last_sample_error",
 			"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 			"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 		}))

 	w := httptest.NewRecorder()
@@ -1417,6 +1422,7 @@ func TestWorkspaceGet_FinancialFieldsStripped(t *testing.T) {
 		"parent_id", "active_tasks", "max_concurrent_tasks", "last_error_rate", "last_sample_error",
 		"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 		"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 	}
 	// Populate with non-zero financial values to confirm they are stripped.
 	mock.ExpectQuery("SELECT w.id, w.name").
@@ -1425,7 +1431,7 @@ func TestWorkspaceGet_FinancialFieldsStripped(t *testing.T) {
 			AddRow("cccccccc-0010-0000-0000-000000000000", "Finance Test", "worker", 1, "online", []byte(`{}`),
 				"http://localhost:9001", nil, 0, 1, 0.0, "", 0, "", "langgraph",
 				"", 0.0, 0.0, false,
-				int64(50000), int64(12500))) // budget_limit=500 USD, spend=125 USD
+				int64(50000), int64(12500), false, true)) // budget_limit=500 USD, spend=125 USD

 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
@@ -1473,6 +1479,7 @@ func TestWorkspaceGet_SensitiveFieldsStripped(t *testing.T) {
 		"parent_id", "active_tasks", "max_concurrent_tasks", "last_error_rate", "last_sample_error",
 		"uptime_seconds", "current_task", "runtime", "workspace_dir", "x", "y", "collapsed",
 		"budget_limit", "monthly_spend",
+		"broadcast_enabled", "talk_to_user_enabled",
 	}
 	mock.ExpectQuery("SELECT w.id, w.name").
 		WithArgs("cccccccc-0955-0000-0000-000000000000").
@@ -1485,7 +1492,7 @@ func TestWorkspaceGet_SensitiveFieldsStripped(t *testing.T) {
 				"langgraph",
 				"/home/user/secret-projects/client-work",
 				0.0, 0.0, false,
-				nil, 0))
+				nil, 0, false, true))

 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
@@ -36,6 +36,15 @@ type Workspace struct {
 	// to activity_logs, agent reads via GET /activity?since_id=). See
 	// migration 045 + RFC #2339.
 	DeliveryMode       string          `json:"delivery_mode" db:"delivery_mode"`
+	// BroadcastEnabled: when true the workspace may call POST /broadcast to
+	// deliver a message to all non-removed agent workspaces in the org.
+	// Default false — only privileged orchestrators should hold this ability.
+	BroadcastEnabled   bool            `json:"broadcast_enabled" db:"broadcast_enabled"`
+	// TalkToUserEnabled: when false the workspace's send_message_to_user calls
+	// and POST /notify requests are rejected with HTTP 403 so the agent is
+	// forced to route updates through a parent workspace. Default true
+	// (preserves existing behaviour for all workspaces).
+	TalkToUserEnabled  bool            `json:"talk_to_user_enabled" db:"talk_to_user_enabled"`
 	// Canvas layout fields (from JOIN)
 	X         float64 `json:"x"`
 	Y         float64 `json:"y"`
@@ -189,6 +189,24 @@ const containerNamePrefix = "ws-"
 // (the wiped-DB case after `docker compose down -v`).
 const LabelManaged = "molecule.platform.managed"

+// AgentUID / AgentGID are the uid/gid of the unprivileged `agent` user that
+// every workspace template creates and drops to via `gosu agent` before
+// exec'ing the runtime (the a2a_mcp_server runs under this uid). The value is
+// fixed at 1000:1000 across all templates — see:
+//   - workspace-configs-templates/claude-code-default/Dockerfile (`useradd -u 1000 ... agent`)
+//   - workspace-configs-templates/hermes/Dockerfile               (`useradd -u 1000 ... agent`)
+//   - workspace/entrypoint.sh                                     (`exec gosu agent` — "uid 1000")
+//
+// Files the platform injects into /configs AFTER the entrypoint's
+// `chown -R agent:agent /configs` (the post-start #418 re-injection and the
+// pre-start #1877 volume write) must be owned by this uid/gid, otherwise the
+// agent-uid MCP server hits EACCES reading /configs/.auth_token, sends an
+// empty bearer, and the platform 401s on /registry/{id}/peers (list_peers).
+const (
+	AgentUID = 1000
+	AgentGID = 1000
+)
+
 // managedLabels is the canonical label map applied to every workspace
 // container + volume. Pulled out so a future addition (e.g. instance
 // UUID for multi-platform-shared-daemon disambiguation) is one edit.
@@ -862,8 +880,18 @@ func buildTemplateTar(templatePath string) (*bytes.Buffer, error) {
 	return &buf, nil
 }

-// WriteFilesToContainer writes in-memory files into /configs in the container.
-func (p *Provisioner) WriteFilesToContainer(ctx context.Context, containerID string, files map[string][]byte) error {
+// buildConfigFilesTar builds the tar stream that WriteFilesToContainer streams
+// into /configs via CopyToContainer. Every entry is stamped Uid/Gid = agent
+// (AgentUID/AgentGID) so the files land agent-owned after extraction. This is
+// the issue #418 post-start re-injection path: it runs AFTER the template
+// entrypoint's `chown -R agent:agent /configs`, so without explicit ownership
+// in the tar header the files extract as root:root (tar Uid/Gid default 0) and
+// the agent-uid MCP server can no longer read /configs/.auth_token (and
+// /configs/.platform_inbound_secret) → empty bearer → list_peers 401.
+//
+// Pulled out as a pure function so the ownership contract is unit-testable
+// without a live Docker daemon (mirrors buildTemplateTar).
+func buildConfigFilesTar(files map[string][]byte) (*bytes.Buffer, error) {
 	var buf bytes.Buffer
 	tw := tar.NewWriter(&buf)

@@ -876,8 +904,10 @@ func (p *Provisioner) WriteFilesToContainer(ctx context.Context, containerID str
 				Typeflag: tar.TypeDir,
 				Name:     dir + "/",
 				Mode:     0755,
+				Uid:      AgentUID,
+				Gid:      AgentGID,
 			}); err != nil {
-				return fmt.Errorf("failed to write tar dir header for %s: %w", dir, err)
+				return nil, fmt.Errorf("failed to write tar dir header for %s: %w", dir, err)
 			}
 			createdDirs[dir] = true
 		}
@@ -886,19 +916,30 @@ func (p *Provisioner) WriteFilesToContainer(ctx context.Context, containerID str
 			Name: name,
 			Mode: 0644,
 			Size: int64(len(data)),
+			Uid:  AgentUID,
+			Gid:  AgentGID,
 		}
 		if err := tw.WriteHeader(header); err != nil {
-			return fmt.Errorf("failed to write tar header for %s: %w", name, err)
+			return nil, fmt.Errorf("failed to write tar header for %s: %w", name, err)
 		}
 		if _, err := tw.Write(data); err != nil {
-			return fmt.Errorf("failed to write tar data for %s: %w", name, err)
+			return nil, fmt.Errorf("failed to write tar data for %s: %w", name, err)
 		}
 	}
 	if err := tw.Close(); err != nil {
-		return fmt.Errorf("failed to close tar writer: %w", err)
+		return nil, fmt.Errorf("failed to close tar writer: %w", err)
 	}
+	return &buf, nil
+}

-	return p.cli.CopyToContainer(ctx, containerID, "/configs", &buf, container.CopyToContainerOptions{})
+// WriteFilesToContainer writes in-memory files into /configs in the container,
+// agent-owned (see buildConfigFilesTar).
+func (p *Provisioner) WriteFilesToContainer(ctx context.Context, containerID string, files map[string][]byte) error {
+	buf, err := buildConfigFilesTar(files)
+	if err != nil {
+		return err
+	}
+	return p.cli.CopyToContainer(ctx, containerID, "/configs", buf, container.CopyToContainerOptions{})
 }

 // CopyToContainer exposes CopyToContainer from the Docker client for use by other packages.
@@ -988,13 +1029,28 @@ func (p *Provisioner) ReadFromVolume(ctx context.Context, volumeName, filePath s
 	return clean, nil
 }

+// writeAuthTokenVolumeCmd is the shell command the throwaway alpine container
+// runs to seed /vol/.auth_token. alpine runs it as root, so without the
+// explicit `chown 1000:1000` the file stays root:root after the template
+// entrypoint's `chown -R agent:agent /configs` has already run — the agent-uid
+// (AgentUID) MCP server then gets EACCES reading it → empty bearer →
+// list_peers 401. Pulled out as a pure function so the ownership contract is
+// unit-testable without a live Docker daemon. Issue #1877.
+func writeAuthTokenVolumeCmd() string {
+	return fmt.Sprintf(
+		"mkdir -p /vol && printf '%%s' $TOKEN > /vol/.auth_token && chmod 0600 /vol/.auth_token && chown %d:%d /vol/.auth_token",
+		AgentUID, AgentGID,
+	)
+}
+
 // WriteAuthTokenToVolume writes the workspace auth token into the config volume
 // BEFORE the container starts, eliminating the token-injection race window where
 // a restarted container could read a stale token from /configs/.auth_token before
 // WriteFilesToContainer writes the new one. Issue #1877.
 //
 // Uses a throwaway alpine container to write directly to the named volume,
-// bypassing the container lifecycle entirely.
+// bypassing the container lifecycle entirely. The written file is chowned to
+// the agent uid/gid (see writeAuthTokenVolumeCmd).
 func (p *Provisioner) WriteAuthTokenToVolume(ctx context.Context, workspaceID, token string) error {
 	if p == nil || p.cli == nil {
 		return ErrNoBackend
@@ -1002,7 +1058,7 @@ func (p *Provisioner) WriteAuthTokenToVolume(ctx context.Context, workspaceID, t
 	volName := ConfigVolumeName(workspaceID)
 	resp, err := p.cli.ContainerCreate(ctx, &container.Config{
 		Image: "alpine",
-		Cmd:   []string{"sh", "-c", "mkdir -p /vol && printf '%s' $TOKEN > /vol/.auth_token && chmod 0600 /vol/.auth_token"},
+		Cmd:   []string{"sh", "-c", writeAuthTokenVolumeCmd()},
 		Env:   []string{"TOKEN=" + token},
 	}, &container.HostConfig{
 		Binds: []string{volName + ":/vol"},
@@ -0,0 +1,95 @@
+package provisioner
+
+import (
+	"archive/tar"
+	"errors"
+	"io"
+	"strings"
+	"testing"
+)
+
+// These tests pin the P0 fix for the fleet-wide list_peers 401 (Hermes and
+// every other template): the workspace-server token-injection paths wrote
+// /configs/.auth_token (and /configs/.platform_inbound_secret) as root:root
+// AFTER the template entrypoint's `chown -R agent:agent /configs` ran, so the
+// agent-uid (1000) MCP server (a2a_mcp_server, running via `gosu agent`) hit
+// `[Errno 13] Permission denied` reading the bearer → empty bearer → platform
+// 401 on /registry/{id}/peers (the literal tool_list_peers path).
+//
+// The agent uid is 1000:1000, verified from the templates:
+//   - workspace-configs-templates/claude-code-default/Dockerfile: `useradd -u 1000 ... agent`
+//   - workspace-configs-templates/hermes/Dockerfile:               `useradd -u 1000 ... agent`
+//   - workspace/entrypoint.sh / claude-code-default/entrypoint.sh:  `exec gosu agent` ("uid 1000")
+//
+// Both tests assert the real artifact (the tar headers Docker's CopyToContainer
+// honours for ownership, and the literal shell command the throwaway alpine
+// container runs), not a mock that bypasses ownership. They FAIL on pre-fix
+// code (no Uid/Gid in tar headers; no chown in the alpine command → root:root)
+// and PASS post-fix (agent-owned).
+
+// TestWriteFilesToContainerTar_FilesAreAgentOwned covers the issue #418
+// post-start re-injection path (WriteFilesToContainer): the tar it streams
+// into /configs via CopyToContainer must carry Uid/Gid = agent (1000) so the
+// extracted files land agent-readable, not root:root. This is the path that
+// (re)writes BOTH .auth_token and .platform_inbound_secret on a cadence.
+func TestWriteFilesToContainerTar_FilesAreAgentOwned(t *testing.T) {
+	files := map[string][]byte{
+		".auth_token":              []byte("tok-abc123"),
+		".platform_inbound_secret": []byte("inbound-secret-xyz"),
+		"nested/dir/file.txt":      []byte("data"),
+	}
+
+	buf, err := buildConfigFilesTar(files)
+	if err != nil {
+		t.Fatalf("buildConfigFilesTar: %v", err)
+	}
+
+	tr := tar.NewReader(buf)
+	seen := map[string]bool{}
+	for {
+		hdr, err := tr.Next()
+		if errors.Is(err, io.EOF) {
+			break
+		}
+		if err != nil {
+			t.Fatalf("read tar: %v", err)
+		}
+		if _, err := io.Copy(io.Discard, tr); err != nil {
+			t.Fatalf("drain %s: %v", hdr.Name, err)
+		}
+		seen[hdr.Name] = true
+		if hdr.Uid != AgentUID {
+			t.Fatalf("tar entry %q Uid = %d, want %d (agent) — root-owned injection causes the list_peers 401",
+				hdr.Name, hdr.Uid, AgentUID)
+		}
+		if hdr.Gid != AgentGID {
+			t.Fatalf("tar entry %q Gid = %d, want %d (agent)", hdr.Name, hdr.Gid, AgentGID)
+		}
+	}
+
+	for _, want := range []string{".auth_token", ".platform_inbound_secret"} {
+		if !seen[want] {
+			t.Fatalf("tar missing %q (seen: %v)", want, seen)
+		}
+	}
+}
+
+// TestWriteAuthTokenVolumeCmd_ChownsToAgent covers the issue #1877 pre-start
+// volume-write path (WriteAuthTokenToVolume): the throwaway alpine container
+// writes /vol/.auth_token then chmod 0600 but, pre-fix, never chowns it, so it
+// stays root:root (alpine runs the command as root). The literal command must
+// chown the file to the agent uid:gid so the agent-uid MCP server can read it.
+func TestWriteAuthTokenVolumeCmd_ChownsToAgent(t *testing.T) {
+	cmd := writeAuthTokenVolumeCmd()
+
+	if !strings.Contains(cmd, "chmod 0600 /vol/.auth_token") {
+		t.Fatalf("alpine cmd lost the 0600 chmod (regression): %q", cmd)
+	}
+
+	wantChown := "chown 1000:1000 /vol/.auth_token"
+	if !strings.Contains(cmd, wantChown) {
+		t.Fatalf("alpine cmd = %q, missing %q — without it .auth_token stays root:root "+
+			"and the agent-uid MCP server gets EACCES → empty bearer → list_peers 401",
+			cmd, wantChown)
+	}
+}
@@ -146,6 +146,9 @@ func Setup(hub *ws.Hub, broadcaster *events.Broadcaster, prov *provisioner.Provi
 		wsAdmin.GET("/workspaces", wh.List)
 		wsAdmin.POST("/workspaces", wh.Create)
 		wsAdmin.DELETE("/workspaces/:id", wh.Delete)
+		// Ability toggles — admin-only so workspace agents cannot self-modify
+		// broadcast_enabled or talk_to_user_enabled.
+		wsAdmin.PATCH("/workspaces/:id/abilities", handlers.PatchAbilities)
 		// Out-of-band bootstrap signal: CP's watcher POSTs here when it
 		// detects "RUNTIME CRASHED" in a workspace EC2 console output,
 		// so the canvas flips to failed in seconds instead of waiting
@@ -201,6 +204,12 @@ func Setup(hub *ws.Hub, broadcaster *events.Broadcaster, prov *provisioner.Provi
 		// to 'hibernated'. The workspace auto-wakes on the next A2A message.
 		wsAuth.POST("/hibernate", wh.Hibernate)

+		// Broadcast — send a message to all non-removed workspaces in the org.
+		// Requires broadcast_enabled=true on the source workspace (checked
+		// inside the handler). WorkspaceAuth on wsAuth proves token ownership.
+		broadcastH := handlers.NewBroadcastHandler(broadcaster)
+		wsAuth.POST("/broadcast", broadcastH.Broadcast)
+
 		// External-workspace credential lifecycle (issue #319 follow-up to
 		// the Create flow). Both endpoints reject runtime ≠ external with
 		// 400 — see external_rotate.go for the rationale.
@@ -224,10 +233,6 @@ func Setup(hub *ws.Hub, broadcaster *events.Broadcaster, prov *provisioner.Provi
 		wsAuth.POST("/delegations/record", delh.Record)
 		wsAuth.POST("/delegations/:delegation_id/update", delh.UpdateStatus)

-		// Workspace broadcast (OFFSEC-015 org isolation)
-		bch := handlers.NewBroadcastHandler(broadcaster)
-		wsAuth.POST("/broadcast", bch.Broadcast)
-
 		// Traces (Langfuse proxy)
 		trh := handlers.NewTracesHandler()
 		wsAuth.GET("/traces", trh.List)
@@ -0,0 +1,3 @@
+ALTER TABLE workspaces
+    DROP COLUMN IF EXISTS broadcast_enabled,
+    DROP COLUMN IF EXISTS talk_to_user_enabled;
@@ -0,0 +1,16 @@
+-- Workspace abilities: opt-in flags that gate platform-level behaviours.
+--
+-- broadcast_enabled (default FALSE): when TRUE the workspace may call
+--   POST /workspaces/:id/broadcast to send a message to every non-removed
+--   agent workspace in the org. Off by default — only privileged
+--   orchestrator workspaces should hold this ability.
+--
+-- talk_to_user_enabled (default TRUE): when FALSE the workspace is not
+--   allowed to deliver messages to the canvas user via send_message_to_user /
+--   POST /notify. The platform returns HTTP 403 so the agent can forward its
+--   update to a parent workspace instead. Default TRUE preserves existing
+--   behaviour for all current workspaces.
+
+ALTER TABLE workspaces
+    ADD COLUMN IF NOT EXISTS broadcast_enabled    BOOLEAN NOT NULL DEFAULT FALSE,
+    ADD COLUMN IF NOT EXISTS talk_to_user_enabled BOOLEAN NOT NULL DEFAULT TRUE;
@@ -29,6 +29,7 @@ from typing import Callable
 import inbox

 from a2a_tools import (
+    tool_broadcast_message,
    tool_chat_history,
    tool_check_task_status,
    tool_commit_memory,
@@ -160,6 +161,11 @@ async def handle_tool_call(name: str, arguments: dict) -> str:
            arguments.get("before_ts", ""),
            source_workspace_id=arguments.get("source_workspace_id") or None,
        )
+    elif name == "broadcast_message":
+        return await tool_broadcast_message(
+            arguments.get("message", ""),
+            workspace_id=arguments.get("workspace_id") or None,
+        )
    return f"Unknown tool: {name}"


@@ -137,6 +137,7 @@ from a2a_tools_delegation import (  # noqa: E402  (import after the from-a2a_cli
 # identically.
 from a2a_tools_messaging import (  # noqa: E402  (import after the top-of-module imports)
    _upload_chat_files,
+    tool_broadcast_message,
    tool_chat_history,
    tool_get_workspace_info,
    tool_list_peers,
@@ -101,6 +101,50 @@ async def _upload_chat_files(
    return uploaded, None


+async def tool_broadcast_message(
+    message: str,
+    workspace_id: str | None = None,
+) -> str:
+    """Send a broadcast message to ALL agent workspaces in the org.
+
+    Requires the workspace to have broadcast_enabled=true (set by a user or
+    admin via PATCH /workspaces/:id/abilities). Use for urgent org-wide
+    signals — status changes, critical alerts, coordination instructions.
+    Every non-removed workspace receives the message in its activity log so
+    poll-mode agents pick it up, and push-mode canvases get a real-time
+    BROADCAST_MESSAGE WebSocket event.
+
+    Args:
+        message: The broadcast text. Keep it concise — all agents receive
+            this, so avoid lengthy prose that floods every context.
+        workspace_id: Optional. Which registered workspace to send the
+            broadcast from. Single-workspace agents omit this.
+    """
+    if not message:
+        return "Error: message is required"
+    target_workspace_id = (workspace_id or "").strip() or WORKSPACE_ID
+    try:
+        async with httpx.AsyncClient(timeout=30.0) as client:
+            resp = await client.post(
+                f"{PLATFORM_URL}/workspaces/{target_workspace_id}/broadcast",
+                json={"message": message},
+                headers=_auth_headers_for_heartbeat(target_workspace_id),
+            )
+            if resp.status_code == 200:
+                data = resp.json()
+                delivered = data.get("delivered", "?")
+                return f"Broadcast sent to {delivered} workspace(s)"
+            if resp.status_code == 403:
+                try:
+                    hint = resp.json().get("hint", "")
+                except Exception:
+                    hint = ""
+                return f"Error: broadcast ability not enabled.{(' ' + hint) if hint else ''}"
+            return f"Error: platform returned {resp.status_code}"
+    except Exception as e:
+        return f"Error sending broadcast: {e}"
+
+
 async def tool_send_message_to_user(
    message: str,
    attachments: list[str] | None = None,
@@ -151,6 +195,20 @@ async def tool_send_message_to_user(
                if uploaded:
                    return f"Message sent to user with {len(uploaded)} attachment(s)"
                return "Message sent to user"
+            if resp.status_code == 403:
+                try:
+                    body = resp.json()
+                    if body.get("error") == "talk_to_user_disabled":
+                        hint = body.get("hint", "")
+                        return (
+                            "Error: this workspace is not allowed to send messages "
+                            "directly to the user (talk_to_user is disabled). "
+                            + (hint + " " if hint else "")
+                            + "Use delegate_task to forward your update to a parent "
+                            "or supervisor workspace that can reach the user."
+                        )
+                except Exception:
+                    pass
            return f"Error: platform returned {resp.status_code}"
    except Exception as e:
        return f"Error sending message: {e}"
@@ -340,6 +340,10 @@ _CLI_A2A_COMMAND_KEYWORDS: dict[str, str | None] = {
    "delegate_task_async": "delegate --async",
    "check_task_status": "status",
    "get_workspace_info": "info",
+    # `broadcast_message` is not exposed via the CLI subprocess interface
+    # today — it's an MCP-first capability. If a2a_cli grows a `broadcast`
+    # subcommand, map it here and the alignment test will gate the change.
+    "broadcast_message": None,
    # `send_message_to_user` is not exposed via the CLI subprocess
    # interface today — it requires a structured `attachments` field
    # that wouldn't survive a positional-arg shell invocation cleanly.
@@ -51,6 +51,7 @@ from dataclasses import dataclass
 from typing import Any, Literal

 from a2a_tools import (
+    tool_broadcast_message,
    tool_chat_history,
    tool_check_task_status,
    tool_commit_memory,
@@ -288,6 +289,44 @@ _GET_WORKSPACE_INFO = ToolSpec(
    section=A2A_SECTION,
 )

+_BROADCAST_MESSAGE = ToolSpec(
+    name="broadcast_message",
+    short=(
+        "Send a message to ALL agent workspaces in the org simultaneously. "
+        "Requires broadcast_enabled=true on this workspace (set by user/admin)."
+    ),
+    when_to_use=(
+        "Use for urgent, org-wide signals: critical status changes, emergency "
+        "stop instructions, coordinated task announcements. Every non-removed "
+        "workspace receives the message in its activity log (poll-mode agents "
+        "see it on their next poll; push-mode canvases get a real-time banner). "
+        "This tool returns an error if broadcast_enabled is false — a user or "
+        "admin must enable it via the workspace abilities settings first."
+    ),
+    input_schema={
+        "type": "object",
+        "properties": {
+            "message": {
+                "type": "string",
+                "description": (
+                    "The broadcast text. Keep it concise — every agent in the "
+                    "org receives this in their activity feed."
+                ),
+            },
+            "workspace_id": {
+                "type": "string",
+                "description": (
+                    "Optional. Multi-workspace mode: the registered workspace "
+                    "to broadcast from. Single-workspace agents omit this."
+                ),
+            },
+        },
+        "required": ["message"],
+    },
+    impl=tool_broadcast_message,
+    section=A2A_SECTION,
+)
+
 _SEND_MESSAGE_TO_USER = ToolSpec(
    name="send_message_to_user",
    short=(
@@ -603,6 +642,7 @@ TOOLS: list[ToolSpec] = [
    _CHECK_TASK_STATUS,
    _LIST_PEERS,
    _GET_WORKSPACE_INFO,
+    _BROADCAST_MESSAGE,
    _SEND_MESSAGE_TO_USER,
    # Inbox (standalone-only; in-container returns informational error)
    _WAIT_FOR_MESSAGE,
@@ -5,6 +5,7 @@
 - **check_task_status**: Poll the status of a task started with delegate_task_async; returns result when done.
 - **list_peers**: List the workspaces this agent can communicate with — name, ID, status, role for each.
 - **get_workspace_info**: Get this workspace's own info — ID, name, role, tier, parent, status.
+- **broadcast_message**: Send a message to ALL agent workspaces in the org simultaneously. Requires broadcast_enabled=true on this workspace (set by user/admin).
 - **send_message_to_user**: Send a message directly to the user's canvas chat — pushed instantly via WebSocket. Use this to: (1) acknowledge a task immediately ('Got it, I'll start working on this'), (2) send interim progress updates while doing long work, (3) deliver follow-up results after delegation completes, (4) attach files (zip, pdf, csv, image) for the user to download via the `attachments` field (NEVER paste file URLs in `message`). The message appears in the user's chat as if you're proactively reaching out.
 - **wait_for_message**: Block until the next inbound message (canvas user OR peer agent) arrives, or until ``timeout_secs`` elapses.
 - **inbox_peek**: List pending inbound messages without removing them.
@@ -26,6 +27,9 @@ Call this first when you need to delegate but don't know the target's ID. Access
 ### get_workspace_info
 Use to introspect your own identity (e.g. before reporting back to the user, or to determine whether you're a tier-0 root that can write GLOBAL memory).

+### broadcast_message
+Use for urgent, org-wide signals: critical status changes, emergency stop instructions, coordinated task announcements. Every non-removed workspace receives the message in its activity log (poll-mode agents see it on their next poll; push-mode canvases get a real-time banner). This tool returns an error if broadcast_enabled is false — a user or admin must enable it via the workspace abilities settings first.
+
 ### send_message_to_user
 Use proactively across the lifecycle of a task — early to acknowledge, mid-flight to update, late to deliver. Never paste file URLs in the message body — always pass absolute paths in `attachments` so the platform serves them as download chips (works on SaaS where external file hosts are unreachable).