diff --git a/strix/tools/todo/tools.py b/strix/tools/todo/tools.py
index 07761f5f8..2014fdd1c 100644
--- a/strix/tools/todo/tools.py
+++ b/strix/tools/todo/tools.py
@@ -131,13 +131,30 @@ def _normalize_todo_ids(raw_ids: Any) -> list[str]:
         stripped = raw_ids.strip()
         if not stripped:
             return []
+        # json.loads is used only to unpack a JSON *array* ('["a", "b"]')
+        # or a JSON *string* scalar ('"a3f9c2"', which unwraps to the id).
+        # Any other bare token is a literal id (optionally comma-separated)
+        # and must NOT be routed through a parsed *numeric* scalar: ids are
+        # 6-char uuid slugs, and ones like "1e5230" or "2363e0" are valid
+        # JSON numbers that json.loads would mangle (-> inf / "2363.0"),
+        # silently targeting a non-existent todo.
         try:
-            data = json.loads(stripped)
+            parsed = json.loads(stripped)
         except json.JSONDecodeError:
-            data = stripped.split(",") if "," in stripped else [stripped]
-        if isinstance(data, list):
-            return [str(item).strip() for item in data if str(item).strip()]
-        return [str(data).strip()]
+            parsed = None
+        if isinstance(parsed, list):
+            return [str(item).strip() for item in parsed if str(item).strip()]
+        # Otherwise resolve to literal tokens: a JSON string scalar unwraps
+        # to its value, a comma string splits, and anything else is a single
+        # token kept verbatim (so numeric-looking ids aren't mangled by the
+        # parse above).
+        if isinstance(parsed, str):
+            tokens = [parsed]
+        elif "," in stripped:
+            tokens = stripped.split(",")
+        else:
+            tokens = [stripped]
+        return [token.strip() for token in tokens if token.strip()]
     if isinstance(raw_ids, list):
         return [str(item).strip() for item in raw_ids if str(item).strip()]
     return [str(raw_ids).strip()]
diff --git a/tests/test_todo.py b/tests/test_todo.py
new file mode 100644
index 000000000..58fc24d47
--- /dev/null
+++ b/tests/test_todo.py
@@ -0,0 +1,91 @@
+"""Tests for per-agent todo id normalization and resolution."""
+
+from __future__ import annotations
+
+import json
+from typing import TYPE_CHECKING
+
+import pytest
+
+import strix.tools.todo.tools as todo_tools
+from strix.tools.todo.tools import _normalize_todo_ids
+
+
+if TYPE_CHECKING:
+    from collections.abc import Iterator
+
+
+@pytest.fixture(autouse=True)
+def _reset_todos_storage(monkeypatch: pytest.MonkeyPatch) -> Iterator[None]:
+    monkeypatch.setattr(todo_tools, "_todos_path", None)
+    with todo_tools._todos_io_lock:
+        todo_tools._todos_storage.clear()
+    yield
+    with todo_tools._todos_io_lock:
+        todo_tools._todos_storage.clear()
+
+
+def test_bare_numeric_looking_id_is_preserved() -> None:
+    # ids are `str(uuid.uuid4())[:6]` hex slugs; slugs like these are valid
+    # JSON numbers, so json.loads would turn "1e5230" into inf and "2363e0"
+    # into 2363.0. They must be kept verbatim as literal ids.
+    assert _normalize_todo_ids("1e5230") == ["1e5230"]
+    assert _normalize_todo_ids("2363e0") == ["2363e0"]
+    assert _normalize_todo_ids("0e4440") == ["0e4440"]
+
+
+def test_plain_hex_and_digit_ids_are_preserved() -> None:
+    assert _normalize_todo_ids("a3f9c2") == ["a3f9c2"]
+    assert _normalize_todo_ids("123456") == ["123456"]
+
+
+def test_json_array_of_ids_is_unpacked() -> None:
+    assert _normalize_todo_ids('["1e5230", "a3f9c2"]') == ["1e5230", "a3f9c2"]
+
+
+def test_json_string_scalar_id_is_unwrapped() -> None:
+    # A caller may JSON-encode a single id ('"a3f9c2"'); the surrounding
+    # quotes must be stripped, not treated as part of the id. A quoted
+    # numeric-looking slug stays a literal string (not a parsed number),
+    # and an empty JSON string yields no ids.
+    assert _normalize_todo_ids('"a3f9c2"') == ["a3f9c2"]
+    assert _normalize_todo_ids('"1e5230"') == ["1e5230"]
+    assert _normalize_todo_ids('""') == []
+
+
+def test_comma_separated_ids_are_split() -> None:
+    assert _normalize_todo_ids("1e5230, a3f9c2") == ["1e5230", "a3f9c2"]
+
+
+def test_list_input_is_stringified_and_stripped() -> None:
+    assert _normalize_todo_ids([" 1e5230 ", "a3f9c2"]) == ["1e5230", "a3f9c2"]
+
+
+def test_empty_and_none_inputs_yield_no_ids() -> None:
+    assert _normalize_todo_ids("") == []
+    assert _normalize_todo_ids("   ") == []
+    assert _normalize_todo_ids(None) == []
+
+
+def test_mark_resolves_a_bare_numeric_looking_id() -> None:
+    # End-to-end: marking a bare id whose slug is a valid JSON number must
+    # find and update the real todo, not fail with "Todo with ID 'inf' not
+    # found".
+    agent_id = "agent-1"
+    todos = todo_tools._get_agent_todos(agent_id)
+    todos["1e5230"] = {
+        "title": "probe /admin",
+        "description": None,
+        "priority": "normal",
+        "status": "pending",
+        "created_at": "2026-07-03T00:00:00+00:00",
+        "updated_at": "2026-07-03T00:00:00+00:00",
+        "completed_at": None,
+    }
+
+    result = json.loads(todo_tools._mark(agent_id=agent_id, todo_ids="1e5230", new_status="done"))
+
+    assert result["success"] is True
+    assert result["marked"] == ["1e5230"]
+    assert "errors" not in result
+    assert todos["1e5230"]["status"] == "done"