fix(security): block read-only cookbook tools for non-admin owners (#6353)

* fix(security): block read-only cookbook tools for non-admin owners

list_cached_models, list_served_models, tail_serve_output, list_downloads,
search_hf_models, list_serve_presets and list_cookbook_servers were in the
plan-mode read-only allowlist but in neither NON_ADMIN_BLOCKED_TOOLS nor the
dispatcher's _ADMIN_TOOLS. Each implementation calls a require_admin route
over the internal loopback with the process-wide INTERNAL_TOOL_TOKEN and
never checks the owner, so a non-admin user's agent inherited admin reach
through them. list_cached_models is the worst case: a caller-chosen `host`
is forwarded to /api/model/cached, which SSHes to that host with the app
user's keys and walks caller-chosen model_dirs. THREAT_MODEL.md lists model
serving as admin-only, so gate the read side the same way as serve/download.

Adding the names to NON_ADMIN_BLOCKED_TOOLS also hides them from non-admin
tool schemas and caps bearer-token agent runs; _ADMIN_TOOLS blocks them at
dispatch. Regression test covers both lists, schema hiding, and that the
loopback is never reached for a non-admin owner.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TwcwfY2uQMxNRsK3y6xBAp

* fix(cookbook): bind request authority in admin gate regressions

Keep all seven read-only Cookbook tools in the admin prompt inventory. Grant request authority in dispatch tests so the admin gate itself is exercised, and verify administrators still reach each handler. Validation: 359 focused authorization, tool-policy and schema tests passed. Against the running app with real auth accounts, all seven tools reject a non-admin and an admin reaches the authenticated Cookbook HTTP API.

---------

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: Afonso Junior <chatgpt@yack.com.br>

src/agent_loop.py

@@ -15204,7 +15204,7 @@ _ADMIN_TOOLS = {
     "manage_documents", "manage_settings", "create_session", "list_sessions",
     "send_to_session", "search_chats", "pipeline", "ask_teacher", "list_models",
     "list_cached_models", "list_downloads", "list_cookbook_servers",
-    "list_serve_presets", "list_served_models",
+    "list_serve_presets", "list_served_models", "tail_serve_output", "search_hf_models",
 }
 
 def _build_base_prompt(

src/tool_execution.py

@@ -1288,6 +1288,16 @@ _ADMIN_TOOLS = {
     "serve_preset",
     "stop_served_model",
     "cancel_download",
+    # Read-only Cookbook tools reach require_admin routes over the internal
+    # loopback without an owner check (see NON_ADMIN_BLOCKED_TOOLS in
+    # src/tool_security.py), so they are admin-only at dispatch as well.
+    "list_served_models",
+    "tail_serve_output",
+    "list_downloads",
+    "search_hf_models",
+    "list_cached_models",
+    "list_serve_presets",
+    "list_cookbook_servers",
 }
 
 

src/tool_security.py

@@ -79,6 +79,22 @@ NON_ADMIN_BLOCKED_TOOLS = BUILTIN_EMAIL_TOOLS | {
     "stop_served_model",
     "cancel_download",
     "adopt_served_model",
+    # Read-only Cookbook inspection tools. They look harmless, but every one
+    # of them calls a require_admin route over the internal loopback with the
+    # process-wide INTERNAL_TOOL_TOKEN and never checks the session owner, so
+    # a non-admin agent would inherit admin reach through them:
+    # list_cached_models forwards a caller-chosen `host` into
+    # /api/model/cached, which SSHes to that host with the app user's keys and
+    # walks caller-chosen `model_dir`s; list_served_models / tail_serve_output
+    # expose serve/download logs and running process command lines. Model
+    # serving is admin-only in THREAT_MODEL.md, so block the read side too.
+    "list_served_models",
+    "tail_serve_output",
+    "list_downloads",
+    "search_hf_models",
+    "list_cached_models",
+    "list_serve_presets",
+    "list_cookbook_servers",
 }