feat: initial agent-reviewer meta-agent

A2A agent that audits another deployed agent's source pre-deploy.
Reads the target Gitea repo via a short-lived read-only token minted
through ctx.mint_gitea_token(), runs an inner DeepAgents graph backed
by GiteaBackend, and emits a typed ReviewReport.

Skill bundles:
- security-anti-patterns (credentials, eval, sandbox bypass, egress)
- a2apack-best-practices (class shape, decorators, types, timeouts)
- grant-scope-evaluation (declared vs actual workspace scope)

Tools:
- sandbox_a2a_card  : round-trips the project through microsandbox
- sandbox_ruff_check: objective static checks
- submit_review_report: typed final report

7 smoke tests pass; agent card loads cleanly via `a2a card`.

tests/test_agent.py

@@ -0,0 +1,139 @@
+"""Smoke tests for agent-reviewer: card, schema, validation behaviors."""
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+import pytest
+
+ROOT = Path(__file__).resolve().parent.parent
+if str(ROOT) not in sys.path:
+    sys.path.insert(0, str(ROOT))
+
+from agent import AgentReviewer  # noqa: E402
+from agent_reviewer.tools import Finding, ReviewReport, ToolContext, build_tools  # noqa: E402
+from agent_reviewer.config import load_settings  # noqa: E402
+
+
+def test_class_metadata() -> None:
+    assert AgentReviewer.name == "agent-reviewer"
+    assert AgentReviewer.version
+    assert AgentReviewer.wants_cp_jwt is True
+
+
+def test_review_skill_registered() -> None:
+    spec = AgentReviewer._skills["review"]  # type: ignore[attr-defined]
+    assert spec.stream is True
+    assert spec.policy.timeout_seconds == 900
+    # Reviewer is read-only — no write grants declared.
+    assert not spec.policy.grant_write_prefixes
+    assert spec.policy.grant_outputs_prefix is None
+
+
+def test_review_report_round_trips() -> None:
+    report = ReviewReport(
+        ok=False,
+        agent_name="hello",
+        ref="main",
+        summary="one critical finding",
+        findings=[
+            Finding(
+                severity="critical",
+                category="security",
+                message="hardcoded API key",
+                file="agent.py",
+                line=42,
+                suggestion="read ctx.llm.api_key instead",
+            )
+        ],
+    )
+    dumped = report.model_dump()
+    assert dumped["ok"] is False
+    again = ReviewReport.model_validate(dumped)
+    assert again.findings[0].severity == "critical"
+
+
+def test_invalid_severity_rejected() -> None:
+    with pytest.raises(Exception):
+        Finding(
+            severity="catastrophic",  # type: ignore[arg-type]
+            category="security",
+            message="x",
+        )
+
+
+@pytest.mark.asyncio
+async def test_submit_review_report_writes_completion_box() -> None:
+    box: dict = {}
+    tools = build_tools(
+        ToolContext(
+            settings=load_settings(),
+            agent_name="hello",
+            ref="main",
+            fetch_source=lambda: {},  # unused for submit
+            completion_box=box,
+        )
+    )
+    submit = next(t for t in tools if t.name == "submit_review_report")
+    result = await submit.ainvoke(
+        {
+            "report_json": json.dumps(
+                {
+                    "ok": True,
+                    "agent_name": "ignored",
+                    "ref": "ignored",
+                    "summary": "clean",
+                    "findings": [],
+                }
+            )
+        }
+    )
+    assert result == {"accepted": True}
+    assert "report" in box
+    # Reviewer overrides agent_name / ref from ctx, not from the LLM.
+    assert box["report"].agent_name == "hello"
+    assert box["report"].ref == "main"
+
+
+@pytest.mark.asyncio
+async def test_submit_review_report_rejects_bad_json() -> None:
+    box: dict = {}
+    tools = build_tools(
+        ToolContext(
+            settings=load_settings(),
+            agent_name="hello",
+            ref="main",
+            fetch_source=lambda: {},
+            completion_box=box,
+        )
+    )
+    submit = next(t for t in tools if t.name == "submit_review_report")
+    result = await submit.ainvoke({"report_json": "{not json"})
+    assert result["accepted"] is False
+    assert "error" in result
+    assert "report" not in box
+
+
+@pytest.mark.asyncio
+async def test_submit_review_report_rejects_bad_schema() -> None:
+    box: dict = {}
+    tools = build_tools(
+        ToolContext(
+            settings=load_settings(),
+            agent_name="hello",
+            ref="main",
+            fetch_source=lambda: {},
+            completion_box=box,
+        )
+    )
+    submit = next(t for t in tools if t.name == "submit_review_report")
+    result = await submit.ainvoke(
+        {
+            "report_json": json.dumps(
+                {"ok": True, "summary": "missing agent_name"}
+            )
+        }
+    )
+    assert result["accepted"] is False
+    assert "report" not in box