core/agent+config: soft context budget in estimated tokens

Replace max_context_chars with max_context_tokens (default 200k est.
tokens via char/4); compact and /status use the token budget.

tests/test_agent/test_compact.py

@@ -6,7 +6,7 @@ from msgspec import UNSET
 
 from plyngent.agent.budget import (
     compact_messages_for_request,
-    estimate_messages_chars,
+    estimate_messages_tokens,
     truncate_tool_result,
 )
 from plyngent.agent.loop import run_chat_loop
@@ -75,7 +75,7 @@ def test_compact_shrinks_old_tool_results() -> None:
 
     compacted = compact_messages_for_request(
         messages,
-        max_chars=estimate_messages_chars(messages) - 1,
+        max_tokens=max(1, estimate_messages_tokens(messages) - 1),
         old_tool_result_chars=40,
         keep_recent_tool_results=1,
     )
@@ -90,11 +90,11 @@ def test_compact_shrinks_old_tool_results() -> None:
     assert compacted[5].content == "NEW" * 50
 
 
-def test_compact_disabled_when_max_chars_zero() -> None:
+def test_compact_disabled_when_max_tokens_zero() -> None:
     messages: list[AnyChatMessage] = [
         ToolChatMessage(content="x" * 500, tool_call_id="1"),
     ]
-    out = compact_messages_for_request(messages, max_chars=0)
+    out = compact_messages_for_request(messages, max_tokens=0)
     assert out[0] is messages[0] or (
         isinstance(out[0], ToolChatMessage) and out[0].content == "x" * 500
     )
@@ -211,7 +211,7 @@ async def test_loop_sends_compacted_request_not_history() -> None:
             history,
             model="m",
             stream=False,
-            max_context_chars=200,
+            max_context_tokens=50,
             max_tool_result_chars=50,
         )
     ]

tests/test_agent/test_compact_summarize.py

@@ -43,7 +43,7 @@ def test_soft_compact_transcript_shrinks_tools() -> None:
         ToolChatMessage(content=big, tool_call_id="1"),
         ToolChatMessage(content="recent", tool_call_id="2"),
     ]
-    out = soft_compact_transcript(messages, max_chars=500)
+    out = soft_compact_transcript(messages, max_tokens=100)
     assert "truncated" in out or len(out) < len(big) + 50
     assert "recent" in out
 

tests/test_cli/test_repl_commands.py

@@ -142,7 +142,7 @@ async def test_rounds(state: ReplState) -> None:
     assert state.agent.max_rounds == 40
 
 
-async def test_status_shows_context_chars(
+async def test_status_shows_context_tokens(
     state: ReplState, capsys: pytest.CaptureFixture[str]
 ) -> None:
     from plyngent.lmproto.openai_compatible.model import UserChatMessage
@@ -150,6 +150,7 @@ async def test_status_shows_context_chars(
     state.agent.messages = [UserChatMessage(content="hello")]
     assert await handle_slash(state, "/status") is True
     out = capsys.readouterr().out
+    assert "context_tokens~=" in out
     assert "context_chars=" in out
     assert "tool_result_max=" in out
     assert str(state.workspace) in out

tests/test_config/test_agent_section.py

@@ -17,7 +17,7 @@ def test_agent_section_defaults(tmp_path: Path) -> None:
     assert store.agent_config.parallel_tools is True
     assert store.agent_config.confirm_destructive is True
     assert store.agent_config.path_denylist == []
-    assert store.agent_config.max_context_chars == 200_000
+    assert store.agent_config.max_context_tokens == 200_000
 
 
 def test_agent_section_parse(tmp_path: Path) -> None:
@@ -30,7 +30,7 @@ max_tool_result_chars = 100
 parallel_tools = false
 confirm_destructive = false
 path_denylist = ["/secrets/", ".ssh/"]
-max_context_chars = 5000
+max_context_tokens = 5000
 """,
         encoding="utf-8",
     )
@@ -40,4 +40,5 @@ max_context_chars = 5000
     assert store.agent_config.parallel_tools is False
     assert store.agent_config.confirm_destructive is False
     assert store.agent_config.path_denylist == ["/secrets/", ".ssh/"]
-    assert store.agent_config.max_context_chars == 5000
+    assert store.agent_config.max_context_tokens == 5000
+