codex-autorunner 1.1.0__tar.gz → 1.2.0__tar.gz

{codex_autorunner-1.1.0/src/codex_autorunner.egg-info → codex_autorunner-1.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codex-autorunner
-Version: 1.1.0
+Version: 1.2.0
 Summary: Codex autorunner CLI per DESIGN-V1
 Author: Codex
 License: MIT License
@@ -85,7 +85,7 @@ The philosophy behind CAR is to let the agents do what they do best, and get out
 CAR treats tickets as the control plane and models as the execution layer. This means that we rely on agents to follow the instructions written in the tickets. If you use a sufficiently weak model, CAR may not work well for you. CAR is an amplifier for agent capabilities. Agents who like to scope creep (create too many new tickets) or reward hack (mark a ticket as done despite it being incomplete) are not a good fit for CAR.
 
 ## Interaction patterns
-CAR's core is a set of python functions surfaced as a CLI, operating on a file system. There are current 2 UIs built on top of this core.
+CAR's core is a set of python functions surfaced as a CLI, operating on a file system. There are current 3 UIs built on top of this core.
 
 ### Web UI
 The web UI is the main control plane for CAR. From here you can set up new repositories or clone existing ones, chat with agents using their TUI, and run the ticket autorunner. There are many quality-of-life features like Whisper integration, editing documents by chatting with AI (useful for mobile), viewing usage analytics, and much more. The Web UI is the most full featured user-facing interface and a good starting point for trying out CAR.
@@ -95,30 +95,23 @@ I recommend serving the web UI over Tailscale. There is an auth token option but
 ### Telegram
 Telegram is the "on-the-go" and notification hub for CAR. From here you can kick off and monitor existing tickets, set up new tickets, and chat with agents. Your primary UX here is asking the agent to do things for you rather than you doing it yourself like you would on the web UI. This is great for on-the-go work, but it doesn't have full feature parity with the web UI.
 
+### Project Manager Agent
+The project manager agent (PMA) is a way to use an AI agent to run CAR. For example instead of making/editing tickets in the UI, you can ask the PMA to do it for you. Instead of starting a ticket flow, the PMA can babysit it for you. The PMA can be accessed in the web and in Telegram, and also uses the CAR CLI directly.
+
 ## Quickstart
 
 The fastest way to get started is to pass [this setup guide](docs/AGENT_SETUP_GUIDE.md) to your favorite AI agent. The agent will walk you through installation and configuration interactively based on your environment.
 
-**TL;DR for the impatient:**
+### From source (repo checkout)
 
-# Install
-```
-pipx install codex-autorunner
-```
-# Initialize in your repo
-```
-cd /path/to/your/repo
-car init
-```
-# Verify setup
-```
-car doctor
-```
-# Create a ticket and run
-```
-car run
+If you're working from a fresh clone of this repo, you can run the repo-local CLI shim:
+
+```bash
+./car --help
 ```
 
+The shim will try `PYTHONPATH=src` first and, if dependencies are missing, will bootstrap a local `.venv` and install CAR.
+
 ## Supported models
 CAR currently supports:
 - Codex
@@ -150,5 +143,8 @@ On the go? The web UI is mobile responsive, or if you prefer you can type or voi
 ![CAR Telegram Media Voice Screenshot](docs/screenshots/telegram-media-voice.PNG)
 ![CAR Telegram Media Image Screenshot](docs/screenshots/telegram-media-image.PNG)
 
+Don't want to use the product directly? Delegate work to the PMA.
+![CAR Telegram PMA Screenshot](docs/screenshots/telegram-pma.png)
+
 ## Star history
 [![Star History Chart](https://api.star-history.com/svg?repos=Git-on-my-level/codex-autorunner&type=Date)](https://star-history.com/#Git-on-my-level/codex-autorunner&Date)

{codex_autorunner-1.1.0 → codex_autorunner-1.2.0}/README.md

@@ -21,7 +21,7 @@ The philosophy behind CAR is to let the agents do what they do best, and get out
 CAR treats tickets as the control plane and models as the execution layer. This means that we rely on agents to follow the instructions written in the tickets. If you use a sufficiently weak model, CAR may not work well for you. CAR is an amplifier for agent capabilities. Agents who like to scope creep (create too many new tickets) or reward hack (mark a ticket as done despite it being incomplete) are not a good fit for CAR.
 
 ## Interaction patterns
-CAR's core is a set of python functions surfaced as a CLI, operating on a file system. There are current 2 UIs built on top of this core.
+CAR's core is a set of python functions surfaced as a CLI, operating on a file system. There are current 3 UIs built on top of this core.
 
 ### Web UI
 The web UI is the main control plane for CAR. From here you can set up new repositories or clone existing ones, chat with agents using their TUI, and run the ticket autorunner. There are many quality-of-life features like Whisper integration, editing documents by chatting with AI (useful for mobile), viewing usage analytics, and much more. The Web UI is the most full featured user-facing interface and a good starting point for trying out CAR.
@@ -31,30 +31,23 @@ I recommend serving the web UI over Tailscale. There is an auth token option but
 ### Telegram
 Telegram is the "on-the-go" and notification hub for CAR. From here you can kick off and monitor existing tickets, set up new tickets, and chat with agents. Your primary UX here is asking the agent to do things for you rather than you doing it yourself like you would on the web UI. This is great for on-the-go work, but it doesn't have full feature parity with the web UI.
 
+### Project Manager Agent
+The project manager agent (PMA) is a way to use an AI agent to run CAR. For example instead of making/editing tickets in the UI, you can ask the PMA to do it for you. Instead of starting a ticket flow, the PMA can babysit it for you. The PMA can be accessed in the web and in Telegram, and also uses the CAR CLI directly.
+
 ## Quickstart
 
 The fastest way to get started is to pass [this setup guide](docs/AGENT_SETUP_GUIDE.md) to your favorite AI agent. The agent will walk you through installation and configuration interactively based on your environment.
 
-**TL;DR for the impatient:**
+### From source (repo checkout)
 
-# Install
-```
-pipx install codex-autorunner
-```
-# Initialize in your repo
-```
-cd /path/to/your/repo
-car init
-```
-# Verify setup
-```
-car doctor
-```
-# Create a ticket and run
-```
-car run
+If you're working from a fresh clone of this repo, you can run the repo-local CLI shim:
+
+```bash
+./car --help
 ```
 
+The shim will try `PYTHONPATH=src` first and, if dependencies are missing, will bootstrap a local `.venv` and install CAR.
+
 ## Supported models
 CAR currently supports:
 - Codex
@@ -86,5 +79,8 @@ On the go? The web UI is mobile responsive, or if you prefer you can type or voi
 ![CAR Telegram Media Voice Screenshot](docs/screenshots/telegram-media-voice.PNG)
 ![CAR Telegram Media Image Screenshot](docs/screenshots/telegram-media-image.PNG)
 
+Don't want to use the product directly? Delegate work to the PMA.
+![CAR Telegram PMA Screenshot](docs/screenshots/telegram-pma.png)
+
 ## Star history
 [![Star History Chart](https://api.star-history.com/svg?repos=Git-on-my-level/codex-autorunner&type=Date)](https://star-history.com/#Git-on-my-level/codex-autorunner&Date)

{codex_autorunner-1.1.0 → codex_autorunner-1.2.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "codex-autorunner"
-version = "1.1.0"
+version = "1.2.0"
 description = "Codex autorunner CLI per DESIGN-V1"
 readme = "README.md"
 license = {file = "LICENSE"}
@@ -108,6 +108,7 @@ ignore = ["E501"]
 "src/codex_autorunner/static_assets.py" = ["F403"]
 "src/codex_autorunner/cli.py" = ["B008"]
 "src/codex_autorunner/surfaces/cli/cli.py" = ["B008"]
+"src/codex_autorunner/surfaces/cli/pma_cli.py" = ["B008"]
 "src/codex_autorunner/routes/terminal_images.py" = ["B008"]
 "src/codex_autorunner/surfaces/web/routes/terminal_images.py" = ["B008"]
 "src/codex_autorunner/routes/voice.py" = ["B008"]

{codex_autorunner-1.1.0 → codex_autorunner-1.2.0}/src/codex_autorunner/agents/opencode/client.py

@@ -21,6 +21,7 @@ class OpenCodeApiProfile:
 
     supports_prompt_async: bool = True
     supports_global_endpoints: bool = True
+    max_text_chars: Optional[int] = None
     spec_fetched: bool = False
 
 
@@ -83,6 +84,7 @@ class OpenCodeClient:
         *,
         auth: Optional[tuple[str, str]] = None,
         timeout: Optional[float] = None,
+        max_text_chars: Optional[int] = None,
         logger: Optional[logging.Logger] = None,
     ) -> None:
         self._client = httpx.AsyncClient(
@@ -93,6 +95,10 @@ class OpenCodeClient:
         self._logger = logger or logging.getLogger(__name__)
         self._api_profile: Optional[OpenCodeApiProfile] = None
         self._api_profile_lock = asyncio.Lock()
+        self._max_text_chars_override = (
+            int(max_text_chars) if isinstance(max_text_chars, int) else None
+        )
+        self._max_text_chars_cache: Optional[int] = None
 
     async def close(self) -> None:
         await self._client.aclose()
@@ -121,6 +127,7 @@ class OpenCodeClient:
                     profile.supports_global_endpoints = self.has_endpoint(
                         spec, "get", "/global/health"
                     ) or self.has_endpoint(spec, "get", "/global/event")
+                    profile.max_text_chars = self._extract_max_text_chars(spec)
 
                 log_event(
                     self._logger,
@@ -147,6 +154,103 @@ class OpenCodeClient:
             return OpenCodeApiProfile()
         return self._api_profile
 
+    @staticmethod
+    def _extract_max_text_chars(spec: dict[str, Any]) -> Optional[int]:
+        if not isinstance(spec, dict):
+            return None
+        components = spec.get("components")
+        if not isinstance(components, dict):
+            return None
+        schemas = components.get("schemas")
+        if not isinstance(schemas, dict):
+            return None
+        candidates: list[int] = []
+        for schema in schemas.values():
+            max_len = OpenCodeClient._find_text_max_length(schema)
+            if isinstance(max_len, int) and max_len > 0:
+                candidates.append(max_len)
+        return min(candidates) if candidates else None
+
+    @staticmethod
+    def _find_text_max_length(schema: Any) -> Optional[int]:
+        if not isinstance(schema, dict):
+            return None
+        candidates: list[int] = []
+        properties = schema.get("properties")
+        if isinstance(properties, dict) and "text" in properties:
+            text_schema = properties.get("text")
+            if isinstance(text_schema, dict):
+                max_len = text_schema.get("maxLength")
+                if isinstance(max_len, int) and max_len > 0:
+                    candidates.append(max_len)
+        for key in ("allOf", "anyOf", "oneOf"):
+            seq = schema.get(key)
+            if isinstance(seq, list):
+                for item in seq:
+                    item_len = OpenCodeClient._find_text_max_length(item)
+                    if isinstance(item_len, int) and item_len > 0:
+                        candidates.append(item_len)
+        return min(candidates) if candidates else None
+
+    def set_max_text_chars(self, value: Optional[int]) -> None:
+        self._max_text_chars_override = int(value) if isinstance(value, int) else None
+        self._max_text_chars_cache = None
+
+    async def _resolve_max_text_chars(
+        self, profile: Optional[OpenCodeApiProfile] = None
+    ) -> Optional[int]:
+        if self._max_text_chars_cache is not None:
+            return self._max_text_chars_cache
+        if profile is None:
+            profile = await self.detect_api_shape()
+        detected = (
+            profile.max_text_chars
+            if isinstance(profile.max_text_chars, int) and profile.max_text_chars > 0
+            else None
+        )
+        override = (
+            self._max_text_chars_override
+            if isinstance(self._max_text_chars_override, int)
+            and self._max_text_chars_override > 0
+            else None
+        )
+        if override is None:
+            resolved = detected
+        elif detected is None:
+            resolved = override
+        else:
+            resolved = min(override, detected)
+        self._max_text_chars_cache = resolved
+        return resolved
+
+    @staticmethod
+    def _split_text(text: str, max_chars: int) -> list[str]:
+        if max_chars <= 0 or len(text) <= max_chars:
+            return [text]
+        parts: list[str] = []
+        start = 0
+        length = len(text)
+        while start < length:
+            end = min(start + max_chars, length)
+            if end < length:
+                split = text.rfind("\n", start, end)
+                if split <= start:
+                    split = text.rfind(" ", start, end)
+                if split > start:
+                    end = split + 1
+            parts.append(text[start:end])
+            start = end
+        return [part for part in parts if part]
+
+    async def _build_text_parts(
+        self, message: str, profile: Optional[OpenCodeApiProfile] = None
+    ) -> list[dict[str, str]]:
+        limit = await self._resolve_max_text_chars(profile)
+        if limit is None:
+            return [{"type": "text", "text": message}]
+        chunks = self._split_text(message, limit)
+        return [{"type": "text", "text": chunk} for chunk in chunks]
+
     def _dir_params(self, directory: Optional[str]) -> dict[str, str]:
         return {"directory": directory} if directory else {}
 
@@ -275,8 +379,10 @@ class OpenCodeClient:
         model: Optional[dict[str, str]] = None,
         variant: Optional[str] = None,
     ) -> Any:
+        profile = await self.detect_api_shape()
+        parts = await self._build_text_parts(message, profile)
         payload: dict[str, Any] = {
-            "parts": [{"type": "text", "text": message}],
+            "parts": parts,
         }
         if agent:
             payload["agent"] = agent
@@ -300,8 +406,10 @@ class OpenCodeClient:
         model: Optional[dict[str, str]] = None,
         variant: Optional[str] = None,
     ) -> Any:
+        profile = await self.detect_api_shape()
+        parts = await self._build_text_parts(message, profile)
         payload: dict[str, Any] = {
-            "parts": [{"type": "text", "text": message}],
+            "parts": parts,
         }
         if agent:
             payload["agent"] = agent
@@ -310,7 +418,6 @@ class OpenCodeClient:
         if variant:
             payload["variant"] = variant
 
-        profile = await self.detect_api_shape()
         if profile.supports_prompt_async:
             return await self._request(
                 "POST",
@@ -335,8 +442,10 @@ class OpenCodeClient:
         model: Optional[dict[str, str]] = None,
         variant: Optional[str] = None,
     ) -> Any:
+        profile = await self.detect_api_shape()
+        parts = await self._build_text_parts(message, profile)
         payload: dict[str, Any] = {
-            "parts": [{"type": "text", "text": message}],
+            "parts": parts,
         }
         if agent:
             payload["agent"] = agent

{codex_autorunner-1.1.0 → codex_autorunner-1.2.0}/src/codex_autorunner/agents/opencode/supervisor.py

@@ -55,6 +55,7 @@ class OpenCodeSupervisor:
         base_url: Optional[str] = None,
         subagent_models: Optional[Mapping[str, str]] = None,
         session_stall_timeout_seconds: Optional[float] = None,
+        max_text_chars: Optional[int] = None,
     ) -> None:
         self._command = [str(arg) for arg in command]
         self._logger = logger or logging.getLogger(__name__)
@@ -70,6 +71,7 @@ class OpenCodeSupervisor:
         self._base_env = base_env
         self._base_url = base_url
         self._subagent_models = subagent_models or {}
+        self._max_text_chars = max_text_chars
         self._handles: dict[str, OpenCodeHandle] = {}
         self._lock: Optional[asyncio.Lock] = None
 
@@ -275,6 +277,7 @@ class OpenCodeSupervisor:
                 base_url,
                 auth=self._auth,
                 timeout=self._request_timeout,
+                max_text_chars=self._max_text_chars,
                 logger=self._logger,
             )
             try:
@@ -344,6 +347,7 @@ class OpenCodeSupervisor:
                 base_url,
                 auth=self._auth,
                 timeout=self._request_timeout,
+                max_text_chars=self._max_text_chars,
                 logger=self._logger,
             )
             try:

{codex_autorunner-1.1.0 → codex_autorunner-1.2.0}/src/codex_autorunner/agents/registry.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import importlib.metadata
 import logging
+import threading
 from dataclasses import dataclass
 from typing import Any, Callable, Iterable, Literal, Optional
 
@@ -102,6 +103,9 @@ _BUILTIN_AGENTS: dict[str, AgentDescriptor] = {
 # Lazy-loaded cache of built-in + plugin agents.
 _AGENT_CACHE: Optional[dict[str, AgentDescriptor]] = None
 
+# Lock to protect cache initialization and reload from concurrent access.
+_AGENT_CACHE_LOCK = threading.Lock()
+
 
 def _select_entry_points(group: str) -> Iterable[importlib.metadata.EntryPoint]:
     """Compatibility wrapper for `importlib.metadata.entry_points()` across py versions."""
@@ -164,10 +168,13 @@ def _load_agent_plugins() -> dict[str, AgentDescriptor]:
             continue
 
         api_version_raw = getattr(descriptor, "plugin_api_version", None)
-        try:
-            api_version = int(api_version_raw)
-        except Exception:
+        if api_version_raw is None:
             api_version = None
+        else:
+            try:
+                api_version = int(api_version_raw)
+            except Exception:
+                api_version = None
         if api_version is None:
             _logger.warning(
                 "Ignoring agent plugin %s: invalid api_version %s",
@@ -215,9 +222,11 @@ def _load_agent_plugins() -> dict[str, AgentDescriptor]:
 def _all_agents() -> dict[str, AgentDescriptor]:
     global _AGENT_CACHE
     if _AGENT_CACHE is None:
-        agents = _BUILTIN_AGENTS.copy()
-        agents.update(_load_agent_plugins())
-        _AGENT_CACHE = agents
+        with _AGENT_CACHE_LOCK:
+            if _AGENT_CACHE is None:
+                agents = _BUILTIN_AGENTS.copy()
+                agents.update(_load_agent_plugins())
+                _AGENT_CACHE = agents
     return _AGENT_CACHE
 
 
@@ -228,7 +237,8 @@ def reload_agents() -> dict[str, AgentDescriptor]:
     """
 
     global _AGENT_CACHE
-    _AGENT_CACHE = None
+    with _AGENT_CACHE_LOCK:
+        _AGENT_CACHE = None
     return get_registered_agents()