Introduce Devii agents subsystem with supervisor, policy, and evals

- Add agents package: controller, runner, store, tools, actions, evals, policy
- Wire agents into Devii service, session, registry, CLI, and dispatcher
- Extend database schema and config for agent persistence
- Add news and telegram bridge updates plus landing UI and docs
- Cover agents, news, and telegram flows with unit tests

tests/unit/services/devii/agents/runner.py

@@ -0,0 +1,133 @@
+# retoor <retoor@molodetz.nl>
+
+import json
+
+from devplacepy.services.devii.agents.runner import AgentRunner
+from devplacepy.services.devii.cost.tracker import CostTracker, Pricing
+from tests.conftest import run_async
+
+
+class _Ledger:
+    def __init__(self):
+        self.spent = 0.0
+
+    def spent_24h(self, owner_kind, owner_id):
+        return self.spent
+
+
+class _Settings:
+    daily_limit_usd = 1.0
+    plan_required = True
+    verify_required = True
+
+
+class _StubSession:
+    def __init__(self):
+        self.cost = CostTracker(pricing=Pricing())
+        self.settings = _Settings()
+        self._ledger = _Ledger()
+        self._llm = None
+        self.dispatcher = None
+        self.owner_kind = "user"
+        self.owner_id = "owner-1"
+
+
+def _runner(**overrides):
+    args = {
+        "session": _StubSession(),
+        "uid": "run-1",
+        "label": "run",
+        "category_key": "research",
+        "effort": "standard",
+        "task": "Compare static site generators.",
+        "output_format": "A markdown table.",
+        "tools": [],
+        "task_scope": False,
+    }
+    args.update(overrides)
+    return AgentRunner(**args)
+
+
+def _ask_call(call_id, question):
+    return {
+        "id": call_id,
+        "function": {"name": "ask_user", "arguments": json.dumps({"question": question})},
+    }
+
+
+def test_system_prompt_carries_contract_and_behavior():
+    runner = _runner(behavior_text="Be terse.")
+    prompt = runner.messages[0]["content"]
+    assert "no browser" in prompt
+    assert "ask_user" in prompt
+    assert "Be terse." in prompt
+    assert "Compare static site generators." in prompt
+    assert "A markdown table." in prompt
+
+
+def test_pending_ask_found_and_answered():
+    runner = _runner()
+    runner.messages.append(
+        {"role": "assistant", "content": "", "tool_calls": [_ask_call("c1", "Budget?")]}
+    )
+    runner.messages.append(
+        {
+            "role": "tool",
+            "tool_call_id": "c1",
+            "name": "ask_user",
+            "content": json.dumps({"status": "paused_for_user"}),
+        }
+    )
+    assert runner._pending_ask() == ("c1", "Budget?")
+    runner.apply_answer("zero")
+    assert runner._pending_ask() is None
+    assert runner.messages[-1]["content"] == "The user answered: zero"
+
+
+def test_quota_spent_short_circuits_without_llm():
+    session = _StubSession()
+    session._ledger.spent = 5.0
+    runner = _runner(session=session)
+    outcome, text = run_async(runner.run_once())
+    assert outcome == "error"
+    assert "quota" in text
+
+
+def test_trace_tracks_progress_and_streak():
+    runner = _runner()
+    runner._trace("call", "fetch_url", "example.com")
+    assert runner.tool_calls == 1
+    runner._trace("err", "fetch_url", "boom")
+    assert runner.error_streak == 1
+    runner._trace("ok", "fetch_url", "")
+    assert runner.error_streak == 0
+    assert "fetch_url" in runner.progress_note
+
+
+def test_task_scope_wraps_the_loop(monkeypatch):
+    from devplacepy.services.devii.tasks.context import in_task_run
+
+    seen = {}
+
+    async def fake_loop(**kwargs):
+        seen["scoped"] = in_task_run()
+        return "done"
+
+    monkeypatch.setattr(
+        "devplacepy.services.devii.agents.runner.react_loop", fake_loop
+    )
+    run_async(_runner(task_scope=True).run_once())
+    assert seen == {"scoped": True}
+    run_async(_runner(task_scope=False).run_once())
+    assert seen == {"scoped": False}
+
+
+def test_counts_and_usage_start_at_zero():
+    runner = _runner()
+    assert runner.counts() == (0, 0, 0.0)
+    assert runner.usage() == {
+        "prompt_tokens": 0,
+        "completion_tokens": 0,
+        "cache_hit_tokens": 0,
+        "cache_miss_tokens": 0,
+    }