feat(autonomous): simply team access and separate admin control to a tab

ai_platform_engineering/autonomous_agents/pyproject.toml

@@ -7,7 +7,9 @@ dependencies = [
     "fastapi==0.135.3",
     "uvicorn[standard]==0.34.2",
     "apscheduler==3.11.0",
+    "boto3==1.43.16",
     "cnoe-agent-utils==0.4.1",
+    "cryptography==50.0.0",
     "httpx==0.28.1",
     "motor==3.7.1",
     "pydantic==2.13.4",

ai_platform_engineering/autonomous_agents/src/autonomous_agents/config.py

@@ -48,6 +48,10 @@ class Settings(BaseSettings):
     # dynamic agent's own model config governs execution).
     llm_provider: str = "anthropic-claude"
 
+    # Cron and interval tasks may not fire more frequently than this. Webhook
+    # tasks are event-driven and deliberately exempt.
+    minimum_schedule_interval_seconds: int = Field(default=1800, ge=1)
+
     # Dynamic-agents runtime — the single execution backend for autonomous
     # tasks. Every task targets a dynamic_agent_id and runs through this
     # service (its tools / system prompt / model / middleware).
@@ -85,6 +89,14 @@ def _reject_nonfinite(cls, v: float) -> float:
     # Per-task secrets always win when both are configured.
     webhook_secret: str | None = None
 
+    # Per-task webhook secrets use the same envelope-encryption scheme as the
+    # UI credential store (including UI-managed Webex OAuth secrets): a fresh
+    # AES-256-GCM data key per write, wrapped by this AWS KMS CMK. When the CMK
+    # is unset, tasks without per-task secrets still work, but persisting or
+    # reading a per-task secret fails closed instead of writing plaintext.
+    credential_kms_cmk_id: str | None = None
+    credential_kms_region: str | None = None
+
     # IMP-07 — webhook replay protection.
     #
     # When > 0, signed webhooks must additionally carry an
@@ -100,6 +112,20 @@ def _reject_nonfinite(cls, v: float) -> float:
     # timestamp header. See README.md for the signing contract.
     webhook_replay_window_seconds: int = Field(default=0, ge=0)
 
+    # Application-level webhook overload protection. Each webhook task is a
+    # FIFO with exactly one active run. Queue item/byte ceilings bound memory;
+    # separate owner/global execution limits allow safe parallelism across
+    # different webhooks. Edge/WAF limiting remains the first DDoS boundary.
+    webhook_max_payload_bytes: int = Field(default=1_048_576, ge=1)
+    webhook_max_pending_per_task: int = Field(default=100, ge=1)
+    webhook_max_pending_per_owner: int = Field(default=500, ge=1)
+    webhook_max_pending_global: int = Field(default=5_000, ge=1)
+    webhook_max_pending_payload_bytes_global: int = Field(
+        default=67_108_864, ge=1
+    )
+    webhook_max_concurrent_per_owner: int = Field(default=20, ge=1)
+    webhook_max_concurrent_global: int = Field(default=100, ge=1)
+
     # Path to the YAML file describing webhook provider adapters
     # (signature header, scheme, algorithm, payload template, etc.).
     # ``None`` (the default) means use the bundled

ai_platform_engineering/autonomous_agents/src/autonomous_agents/error_handlers.py

@@ -0,0 +1,34 @@
+"""Application-level exception handlers."""
+
+import logging
+
+from fastapi import Request
+from fastapi.responses import JSONResponse
+
+from autonomous_agents.services.secret_encryption import WebhookSecretEncryptionError
+
+logger = logging.getLogger("autonomous_agents")
+
+
+async def webhook_secret_encryption_error_handler(
+    request: Request,
+    exc: WebhookSecretEncryptionError,
+) -> JSONResponse:
+    """Return a stable JSON error without exposing encryption internals."""
+    logger.error(
+        "Webhook secret encryption failed for %s %s: %s",
+        request.method,
+        request.url.path,
+        exc,
+        exc_info=(type(exc), exc, exc.__traceback__),
+    )
+    return JSONResponse(
+        status_code=503,
+        content={
+            "detail": (
+                "Webhook secret encryption is unavailable. Check "
+                "CREDENTIAL_KMS_CMK_ID and the Autonomous Agents service's "
+                "AWS KMS permissions."
+            )
+        },
+    )

ai_platform_engineering/autonomous_agents/src/autonomous_agents/main.py

@@ -13,6 +13,7 @@
 from fastapi.middleware.cors import CORSMiddleware
 
 from autonomous_agents.config import get_settings
+from autonomous_agents.error_handlers import webhook_secret_encryption_error_handler
 from autonomous_agents.routes import health, tasks, webex, webhooks
 from autonomous_agents.routes.webex import set_bot_person_id, set_webex_client
 from autonomous_agents.services.chat_history import NoopChatHistoryPublisher
@@ -28,6 +29,7 @@
     get_scheduler,
     register_scheduler_tasks,
 )
+from autonomous_agents.services.secret_encryption import WebhookSecretEncryptionError
 from autonomous_agents.services.task_lifecycle import set_task_store
 from autonomous_agents.services.task_runner import (
     set_chat_history_publisher,
@@ -275,6 +277,11 @@ def create_app() -> FastAPI:
         lifespan=lifespan,
     )
 
+    app.add_exception_handler(
+        WebhookSecretEncryptionError,
+        webhook_secret_encryption_error_handler,
+    )
+
     # Add CORS middleware
     app.add_middleware(
         CORSMiddleware,

ai_platform_engineering/autonomous_agents/src/autonomous_agents/models.py

@@ -2,10 +2,9 @@
 
 from datetime import datetime, timezone
 from enum import Enum
-from math import isfinite
 from typing import Any, Literal, Optional
 
-from pydantic import BaseModel, Field, field_validator, model_validator
+from pydantic import BaseModel, Field, model_validator
 
 
 class TriggerType(str, Enum):
@@ -53,7 +52,14 @@ def require_positive_interval(self) -> "IntervalTrigger":
 class WebhookTrigger(BaseModel):
     """Trigger for webhook-scheduled tasks"""
     type: Literal[TriggerType.WEBHOOK] = TriggerType.WEBHOOK
-    secret: str | None = Field(None, description="Optional HMAC secret for payload validation")
+    secret: str | None = Field(
+        None,
+        description=(
+            "HMAC secret for payload validation. The create route always generates "
+            "this value; None is accepted only so redacted update payloads can "
+            "preserve the already-stored secret."
+        ),
+    )
     provider: str = Field(
         default="generic_hmac",
         description=(
@@ -194,14 +200,6 @@ class TaskDefinition(BaseModel):
     )
     enabled: bool = True
     metadata: dict[str, Any] = Field(default_factory=dict)
-    timeout_seconds: float | None = Field(
-        default=None,
-        gt=0,
-        description=(
-            "Override the dynamic-agents call timeout for this task "
-            "(seconds, > 0). Defaults to DYNAMIC_AGENTS_TIMEOUT_SECONDS."
-        ),
-    )
     owner_id: str | None = Field(
         default=None,
         description=(
@@ -225,16 +223,6 @@ class TaskDefinition(BaseModel):
         ),
     )
 
-    @field_validator("timeout_seconds")
-    @classmethod
-    def _timeout_must_be_finite(cls, v: float | None) -> float | None:
-        """Reject non-finite values that would break httpx timeouts at runtime."""
-        if v is None:
-            return v
-        if not isfinite(v):
-            raise ValueError("timeout_seconds must be a finite number")
-        return v
-
     @model_validator(mode="after")
     def _drop_deprecated_agent_hint(self) -> "TaskDefinition":
         """Clear the deprecated ``agent`` hint when a dynamic agent is set.
@@ -303,6 +291,17 @@ class FollowUpContext(BaseModel):
     )
 
 
+class TaskRunFollowUpCreate(BaseModel):
+    """Authenticated UI request to continue one webhook run."""
+
+    user_text: str = Field(
+        ...,
+        min_length=1,
+        max_length=10_000,
+        description="Operator message to send in the selected run's context.",
+    )
+
+
 
 class TaskCreate(TaskDefinition):
     """Request body for ``POST /tasks``.
@@ -333,6 +332,18 @@ class TaskRun(BaseModel):
     # threaded timeline instead of unrelated rows. ``None`` for the
     # original webhook fire and for cron / interval / manual runs.
     parent_run_id: str | None = None
+    # Root delivery for a webhook conversation branch. Initial deliveries set
+    # this to their own run_id; follow-ups inherit it from the selected parent.
+    # None for cron/interval runs and legacy webhook records.
+    root_run_id: str | None = None
+    # Dynamic Agents conversation/checkpointer id used for execution. This is
+    # intentionally distinct from ``conversation_id`` below: the latter is a
+    # UI chat-history link, while this field protects execution isolation.
+    execution_context_id: str | None = None
+    # Exact operator message for a follow-up run. Stored separately from the
+    # augmented request_prompt so timeline clients can render a clean user turn.
+    follow_up_text: str | None = None
+    follow_up_transport: str | None = None
     # Prompt materialised for this specific run. For normal scheduled
     # runs this is the task prompt; for inbound follow-ups it includes
     # the operator reply appended by task_runner. The UI's autonomous