feat(runners): add runner_bridge context window display, enhance claude/codex/opencode runners and telegram session handling

src/takopi/markdown.py

@@ -240,6 +240,8 @@ class MarkdownFormatter:
 
     def _format_footer(self, state: ProgressState) -> str | None:
         lines: list[str] = []
+        if state.usage_line:
+            lines.append(state.usage_line)
         if state.context_line:
             lines.append(state.context_line)
         if state.resume_line:

src/takopi/progress.py

@@ -25,6 +25,7 @@ class ProgressState:
     resume: ResumeToken | None
     resume_line: str | None
     context_line: str | None
+    usage_line: str | None = None
 
 
 class ProgressTracker:
@@ -82,6 +83,7 @@ class ProgressTracker:
         *,
         resume_formatter: Callable[[ResumeToken], str] | None = None,
         context_line: str | None = None,
+        usage_line: str | None = None,
     ) -> ProgressState:
         resume_line: str | None = None
         if self.resume is not None and resume_formatter is not None:
@@ -96,4 +98,5 @@ class ProgressTracker:
             resume=self.resume,
             resume_line=resume_line,
             context_line=context_line,
+            usage_line=usage_line,
         )

src/takopi/runner_bridge.py

@@ -25,6 +25,61 @@ from .transport import (
 
 logger = get_logger(__name__)
 
+_CONTEXT_WINDOWS: dict[str, int] = {
+    "claude": 200_000,
+    "pi": 200_000,
+}
+
+
+def _fuzzy_tokens(n: int) -> str:
+    if n < 1000:
+        return str(n)
+    k = n / 1000
+    formatted = f"{k:.2f}".rstrip("0").rstrip(".")
+    return f"{formatted}k"
+
+
+def _format_usage_line(usage: dict | None, *, engine: str = "") -> str:
+    if not usage:
+        return "ctx: n/a"
+    inner: dict = usage.get("usage") or {}
+    is_claude_nested = bool(inner)
+    if is_claude_nested:
+        input_tokens: int | None = inner.get("input_tokens")
+        _out: int | None = inner.get("output_tokens")
+        ctx_tokens: int | None = (
+            (input_tokens + _out)
+            if input_tokens is not None and _out is not None
+            else (input_tokens or _out)
+        )
+        output_tokens: int | None = None  # folded into ctx_tokens
+    else:
+        # codex: use last_usage (per-request) for accurate context window display.
+        # cumulative usage.input_tokens grows across API calls and exceeds window size.
+        last: dict = usage.get("last_usage") or {}
+        last_input: int | None = last.get("input_tokens")
+        last_cached: int | None = last.get("cached_input_tokens")
+        if last_input is not None:
+            ctx_tokens = last_input + (last_cached or 0)
+        else:
+            ctx_tokens = None
+        output_tokens = None
+    if ctx_tokens is None:
+        return "ctx: n/a"
+    ctx_str = _fuzzy_tokens(ctx_tokens)
+    context_window = _CONTEXT_WINDOWS.get(engine)
+    if context_window:
+        max_str = _fuzzy_tokens(context_window)
+        parts = [f"ctx: {ctx_str}/{max_str}"]
+    else:
+        parts = [f"ctx: {ctx_str}"]
+    if output_tokens is not None:
+        parts.append(f"out: {_fuzzy_tokens(output_tokens)}")
+    cost = usage.get("total_cost_usd")
+    if cost is not None:
+        parts.append(f"${cost:.4f}")
+    return " · ".join(parts)
+
 
 def _log_runner_event(evt: TakopiEvent) -> None:
     for line in render_event_cli(evt):
@@ -499,6 +554,7 @@ async def handle_message(
         state = progress_tracker.snapshot(
             resume_formatter=runner.format_resume,
             context_line=context_line,
+            usage_line=_format_usage_line(None, engine=runner.engine),
         )
         final_rendered = cfg.presenter.render_final(
             state,
@@ -535,6 +591,7 @@ async def handle_message(
         state = progress_tracker.snapshot(
             resume_formatter=runner.format_resume,
             context_line=context_line,
+            usage_line=_format_usage_line(None, engine=runner.engine),
         )
         final_rendered = cfg.presenter.render_progress(
             state,
@@ -589,6 +646,7 @@ async def handle_message(
     state = progress_tracker.snapshot(
         resume_formatter=runner.format_resume,
         context_line=context_line,
+        usage_line=_format_usage_line(completed.usage, engine=completed.engine),
     )
     final_rendered = cfg.presenter.render_final(
         state,

src/takopi/runners/claude.py

@@ -10,6 +10,7 @@ from typing import Any
 import msgspec
 
 from ..backends import EngineBackend, EngineConfig
+from ..config import ConfigError
 from ..events import EventFactory
 from ..logging import get_logger
 from ..model import Action, ActionKind, EngineId, ResumeToken, TakopiEvent
@@ -288,6 +289,7 @@ class ClaudeRunner(ResumeTokenMixin, JsonlSubprocessRunner):
     allowed_tools: list[str] | None = None
     dangerously_skip_permissions: bool = False
     use_api_billing: bool = False
+    extra_args: list[str] | None = None
     session_title: str = "claude"
     logger = logger
 
@@ -311,6 +313,8 @@ class ClaudeRunner(ResumeTokenMixin, JsonlSubprocessRunner):
             args.extend(["--allowedTools", allowed_tools])
         if self.dangerously_skip_permissions is True:
             args.append("--dangerously-skip-permissions")
+        if self.extra_args:
+            args.extend(self.extra_args)
         args.append("--")
         args.append(prompt)
         return args
@@ -455,7 +459,11 @@ class ClaudeRunner(ResumeTokenMixin, JsonlSubprocessRunner):
 
 
 def build_runner(config: EngineConfig, _config_path: Path) -> Runner:
-    claude_cmd = shutil.which("claude") or "claude"
+    cmd_override = config.get("claude_cmd")
+    if isinstance(cmd_override, str) and cmd_override:
+        claude_cmd = shutil.which(cmd_override) or cmd_override
+    else:
+        claude_cmd = shutil.which("claude") or "claude"
 
     model = config.get("model")
     if "allowed_tools" in config:
@@ -464,6 +472,17 @@ def build_runner(config: EngineConfig, _config_path: Path) -> Runner:
         allowed_tools = DEFAULT_ALLOWED_TOOLS
     dangerously_skip_permissions = config.get("dangerously_skip_permissions") is True
     use_api_billing = config.get("use_api_billing") is True
+    extra_args_raw = config.get("extra_args")
+    if extra_args_raw is None:
+        extra_args = None
+    elif isinstance(extra_args_raw, list) and all(
+        isinstance(x, str) for x in extra_args_raw
+    ):
+        extra_args = list(extra_args_raw)
+    else:
+        raise ConfigError(
+            f"Invalid `claude.extra_args` in {_config_path}; expected a list of strings."
+        )
     title = str(model) if model is not None else "claude"
 
     return ClaudeRunner(
@@ -472,6 +491,7 @@ def build_runner(config: EngineConfig, _config_path: Path) -> Runner:
         allowed_tools=allowed_tools,
         dangerously_skip_permissions=dangerously_skip_permissions,
         use_api_billing=use_api_billing,
+        extra_args=extra_args,
         session_title=title,
     )
 

src/takopi/runners/codex.py

@@ -585,13 +585,16 @@ class CodexRunner(ResumeTokenMixin, JsonlSubprocessRunner):
                         title="turn started",
                     )
                 ]
-            case codex_schema.TurnCompleted(usage=usage):
+            case codex_schema.TurnCompleted(usage=usage, last_usage=last_usage):
                 resume_for_completed = found_session or resume
+                usage_dict = msgspec.to_builtins(usage)
+                if last_usage is not None:
+                    usage_dict["last_usage"] = msgspec.to_builtins(last_usage)
                 return [
                     factory.completed_ok(
                         answer=state.final_answer or "",
                         resume=resume_for_completed,
-                        usage=msgspec.to_builtins(usage),
+                        usage=usage_dict,
                     )
                 ]
             case codex_schema.ItemCompleted(
@@ -664,7 +667,12 @@ class CodexRunner(ResumeTokenMixin, JsonlSubprocessRunner):
 
 
 def build_runner(config: EngineConfig, config_path: Path) -> Runner:
-    codex_cmd = "codex"
+    cmd_value = config.get("cmd")
+    if cmd_value is not None and not isinstance(cmd_value, str):
+        raise ConfigError(
+            f"Invalid `codex.cmd` in {config_path}; expected a string."
+        )
+    codex_cmd = cmd_value if isinstance(cmd_value, str) else "codex"
 
     extra_args_value = config.get("extra_args")
     if extra_args_value is None:

src/takopi/runners/opencode.py

@@ -308,6 +308,7 @@ class OpenCodeRunner(ResumeTokenMixin, JsonlSubprocessRunner):
 
     opencode_cmd: str = "opencode"
     model: str | None = None
+    extra_args: list[str] | None = None
     session_title: str = "opencode"
     logger = logger
 
@@ -335,6 +336,8 @@ class OpenCodeRunner(ResumeTokenMixin, JsonlSubprocessRunner):
             model = run_options.model
         if model is not None:
             args.extend(["--model", str(model)])
+        if self.extra_args:
+            args.extend(self.extra_args)
         args.extend(["--", prompt])
         return args
 
@@ -486,11 +489,24 @@ def build_runner(config: EngineConfig, config_path: Path) -> Runner:
             f"Invalid `opencode.model` in {config_path}; expected a string."
         )
 
+    extra_args_raw = config.get("extra_args")
+    if extra_args_raw is None:
+        extra_args = None
+    elif isinstance(extra_args_raw, list) and all(
+        isinstance(x, str) for x in extra_args_raw
+    ):
+        extra_args = list(extra_args_raw)
+    else:
+        raise ConfigError(
+            f"Invalid `opencode.extra_args` in {config_path}; expected a list of strings."
+        )
+
     title = str(model) if model is not None else "opencode"
 
     return OpenCodeRunner(
         opencode_cmd=opencode_cmd,
         model=model,
+        extra_args=extra_args,
         session_title=title,
     )
 

src/takopi/schemas/codex.py

@@ -54,6 +54,7 @@ class TurnStarted(msgspec.Struct, tag="turn.started", kw_only=True):
 
 class TurnCompleted(msgspec.Struct, tag="turn.completed", kw_only=True):
     usage: Usage
+    last_usage: Usage | None = None
 
 
 class TurnFailed(msgspec.Struct, tag="turn.failed", kw_only=True):

src/takopi/telegram/chat_sessions.py

@@ -90,6 +90,19 @@ class ChatSessionStore(JsonStateStore[_ChatSessionsState]):
             chat.sessions[token.engine] = _SessionState(resume=token.value)
             self._save_locked()
 
+    async def get_any_session_resume(
+        self, chat_id: int, owner_id: int | None
+    ) -> ResumeToken | None:
+        async with self._lock:
+            self._reload_locked_if_needed()
+            chat = self._get_chat_locked(chat_id, owner_id)
+            if chat is None:
+                return None
+            for engine, entry in chat.sessions.items():
+                if entry.resume:
+                    return ResumeToken(engine=engine, value=entry.resume)
+            return None
+
     async def clear_sessions(self, chat_id: int, owner_id: int | None) -> None:
         async with self._lock:
             self._reload_locked_if_needed()

src/takopi/telegram/loop.py

@@ -717,6 +717,11 @@ class ResumeResolver:
                 topic_key[1],
                 engine_for_session,
             )
+            if stored is None:
+                stored = await self._topic_store.get_any_session_resume(
+                    topic_key[0],
+                    topic_key[1],
+                )
             if stored is not None:
                 resume_token = stored
         if (
@@ -729,6 +734,11 @@ class ResumeResolver:
                 chat_session_key[1],
                 engine_for_session,
             )
+            if stored is None:
+                stored = await self._chat_session_store.get_any_session_resume(
+                    chat_session_key[0],
+                    chat_session_key[1],
+                )
             if stored is not None:
                 resume_token = stored
         return ResumeDecision(resume_token=resume_token, handled_by_running_task=False)

src/takopi/telegram/topic_state.py

@@ -174,6 +174,19 @@ class TopicStateStore(JsonStateStore[_TopicState]):
                 return None
             return ResumeToken(engine=engine, value=entry.resume)
 
+    async def get_any_session_resume(
+        self, chat_id: int, thread_id: int
+    ) -> ResumeToken | None:
+        async with self._lock:
+            self._reload_locked_if_needed()
+            thread = self._get_thread_locked(chat_id, thread_id)
+            if thread is None:
+                return None
+            for engine, entry in thread.sessions.items():
+                if entry.resume:
+                    return ResumeToken(engine=engine, value=entry.resume)
+            return None
+
     async def get_default_engine(self, chat_id: int, thread_id: int) -> str | None:
         async with self._lock:
             self._reload_locked_if_needed()