cecli-dev 0.99.11__tar.gz → 0.100.0__tar.gz

{cecli_dev-0.99.11 → cecli_dev-0.100.0}/.github/workflows/release.yml

@@ -4,7 +4,7 @@ on:
   workflow_dispatch:
   push:
     tags:
-      - 'v[0-9]+.[0-9]+.[0-9]+'
+      - 'v[0-9]*.[0-9]*.[0-9]*'
 
 jobs:
   publish_cecli:

{cecli_dev-0.99.11 → cecli_dev-0.100.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cecli-dev
-Version: 0.99.11
+Version: 0.100.0
 Summary: we also can't pronounce cecli
 Project-URL: Homepage, https://github.com/dwash96/cecli
 Project-URL: Documentation, https://github.com/dwash96/cecli
@@ -76,16 +76,16 @@ Requires-Dist: cogapp; extra == "dev"
 Requires-Dist: semver; extra == "dev"
 Requires-Dist: codespell; extra == "dev"
 Requires-Dist: uv; extra == "dev"
-Requires-Dist: memray; extra == "dev"
+Requires-Dist: memray; sys_platform != "win32" and extra == "dev"
 Requires-Dist: objgraph; extra == "dev"
 Requires-Dist: pympler; extra == "dev"
 Requires-Dist: guppy3; extra == "dev"
+Requires-Dist: psutil; extra == "dev"
 Provides-Extra: help
 Requires-Dist: llama-index-embeddings-huggingface; extra == "help"
 Requires-Dist: numpy>=1.26.4; extra == "help"
 Provides-Extra: playwright
 Requires-Dist: playwright; extra == "playwright"
-Provides-Extra: tui
 Dynamic: license-file
 
 ## Why `cecli`?
@@ -110,6 +110,7 @@ LLMs are a part of our lives from here on out so join us in learning about and c
 * [MCP Configuration](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/mcp.md)
 * [TUI Configuration](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/tui.md)
 * [Skills](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/skills.md)
+* [Subagents](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/subagents.md)
 * [Session Management](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/sessions.md)
 * [Hooks](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/hooks.md)
 * [Workspaces](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/workspaces.md)
@@ -232,7 +233,7 @@ The current priorities are to improve core capabilities and user experience of t
   * [ ] Build an explicit workflow and local tooling for internal discovery mechanisms
 
 4. **Context Delivery** - [Discussion](https://github.com/dwash96/cecli/issues/47)
-  * [ ] Use workflow for internal discovery to better target file snippets needed for specific tasks
+  * [x] Use workflow for internal discovery to better target file snippets needed for specific tasks (ExploreCode and ReadRange)
   * [x] Add support for partial files and code snippets in model completion messages
   * [x] Update message request structure for optimal caching
 
@@ -251,12 +252,12 @@ The current priorities are to improve core capabilities and user experience of t
   * [x] Add a dynamic tool discovery tool to allow the system to have only the tools it needs in context
 
 7. **Sub Agents**
-  * [ ] Add `/fork` and `/rejoin` commands to manually manage parts of the conversation history
+  * [x] Add `/invoke-agent` command to manually branch a sub agent and return a summary to the main context
   * [x] Add an instance-able view of the conversation system so sub agents get their own context and workspaces
   * [x] Modify coder classes to have discrete identifiers for themselves/management utilities for them to have their own slices of the world
   * [x] Refactor global files like todo lists to live inside instance folders to avoid state conflicts
-  * [ ] Add a `spawn` tool that launches a sub agent as a background command that the parent model waits for to finish
-  * [ ] Add visibility into active sub agent calls in TUI
+  * [x] Add a `Delegate` tool that launches a sub agent as a background command that the parent model waits for to finish
+  * [x] Add visibility into active sub agent calls in TUI
 
 8. **Hooks**
   * [x] Add hooks base class for user defined python hooks with an execute method with type and priority settings
@@ -270,7 +271,7 @@ The current priorities are to improve core capabilities and user experience of t
   * [x] Update internal file diff representation to support hashline propagation
 
 10. **Dynamic Context Management** 
-  * [ ] Update compaction to use observational memory sub agent calls to generate decision records that are used as the compaction basis
+  * [x] Update compaction to use observational memory sub agent calls to generate decision records that are used as the compaction basis
   * [ ] Persist decision records to disk for sessions with some settings for managing lifetimes of such persistence
   * [ ] Integrate RLM to extract information from decision records on disk and other definable notes
   * [ ] Add a "describe" tool that launches a sub agent workflow that populates an RLM call's context with:

{cecli_dev-0.99.11 → cecli_dev-0.100.0}/README.md

@@ -20,6 +20,7 @@ LLMs are a part of our lives from here on out so join us in learning about and c
 * [MCP Configuration](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/mcp.md)
 * [TUI Configuration](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/tui.md)
 * [Skills](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/skills.md)
+* [Subagents](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/subagents.md)
 * [Session Management](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/sessions.md)
 * [Hooks](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/hooks.md)
 * [Workspaces](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/workspaces.md)
@@ -142,7 +143,7 @@ The current priorities are to improve core capabilities and user experience of t
   * [ ] Build an explicit workflow and local tooling for internal discovery mechanisms
 
 4. **Context Delivery** - [Discussion](https://github.com/dwash96/cecli/issues/47)
-  * [ ] Use workflow for internal discovery to better target file snippets needed for specific tasks
+  * [x] Use workflow for internal discovery to better target file snippets needed for specific tasks (ExploreCode and ReadRange)
   * [x] Add support for partial files and code snippets in model completion messages
   * [x] Update message request structure for optimal caching
 
@@ -161,12 +162,12 @@ The current priorities are to improve core capabilities and user experience of t
   * [x] Add a dynamic tool discovery tool to allow the system to have only the tools it needs in context
 
 7. **Sub Agents**
-  * [ ] Add `/fork` and `/rejoin` commands to manually manage parts of the conversation history
+  * [x] Add `/invoke-agent` command to manually branch a sub agent and return a summary to the main context
   * [x] Add an instance-able view of the conversation system so sub agents get their own context and workspaces
   * [x] Modify coder classes to have discrete identifiers for themselves/management utilities for them to have their own slices of the world
   * [x] Refactor global files like todo lists to live inside instance folders to avoid state conflicts
-  * [ ] Add a `spawn` tool that launches a sub agent as a background command that the parent model waits for to finish
-  * [ ] Add visibility into active sub agent calls in TUI
+  * [x] Add a `Delegate` tool that launches a sub agent as a background command that the parent model waits for to finish
+  * [x] Add visibility into active sub agent calls in TUI
 
 8. **Hooks**
   * [x] Add hooks base class for user defined python hooks with an execute method with type and priority settings
@@ -180,7 +181,7 @@ The current priorities are to improve core capabilities and user experience of t
   * [x] Update internal file diff representation to support hashline propagation
 
 10. **Dynamic Context Management** 
-  * [ ] Update compaction to use observational memory sub agent calls to generate decision records that are used as the compaction basis
+  * [x] Update compaction to use observational memory sub agent calls to generate decision records that are used as the compaction basis
   * [ ] Persist decision records to disk for sessions with some settings for managing lifetimes of such persistence
   * [ ] Integrate RLM to extract information from decision records on disk and other definable notes
   * [ ] Add a "describe" tool that launches a sub agent workflow that populates an RLM call's context with:

{cecli_dev-0.99.11 → cecli_dev-0.100.0}/cecli/__init__.py

@@ -1,6 +1,6 @@
 from packaging import version
 
-__version__ = "0.99.10.dev"
+__version__ = "0.99.12.dev"
 safe_version = __version__
 
 try:

{cecli_dev-0.99.11 → cecli_dev-0.100.0}/cecli/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.99.11'
-__version_tuple__ = version_tuple = (0, 99, 11)
+__version__ = version = '0.100.0'
+__version_tuple__ = version_tuple = (0, 100, 0)
 
-__commit_id__ = commit_id = 'g2e5e3b75c'
+__commit_id__ = commit_id = 'g8b6086f7f'

{cecli_dev-0.99.11 → cecli_dev-0.100.0}/cecli/args.py

@@ -403,7 +403,7 @@ def get_parser(default_config_files, git_root):
         "--use-enhanced-map",
         action="store_true",
         help="Use enhanced Repo Map that takes into account imports (default: False)",
-        default=False,
+        default=True,
     )
 
     ##########

{cecli_dev-0.99.11 → cecli_dev-0.100.0}/cecli/coders/__init__.py

@@ -12,6 +12,7 @@ from .editor_whole_coder import EditorWholeFileCoder
 from .hashline_coder import HashLineCoder
 from .help_coder import HelpCoder
 from .patch_coder import PatchCoder
+from .sub_agent_coder import SubAgentCoder
 from .udiff_coder import UnifiedDiffCoder
 from .udiff_simple import UnifiedDiffSimpleCoder
 from .wholefile_coder import WholeFileCoder
@@ -37,4 +38,5 @@ __all__ = [
     AgentCoder,
     CopyPasteCoder,
     HashLineCoder,
+    SubAgentCoder,
 ]

{cecli_dev-0.99.11 → cecli_dev-0.100.0}/cecli/coders/agent_coder.py

@@ -2,6 +2,7 @@ import asyncio
 import base64
 import json
 import locale
+import logging
 import os
 import platform
 import random
@@ -14,6 +15,7 @@ from pathlib import Path
 from cecli import utils
 from cecli.change_tracker import ChangeTracker
 from cecli.helpers import nested, responses
+from cecli.helpers.agents.service import AgentService
 from cecli.helpers.background_commands import BackgroundCommandManager
 from cecli.helpers.conversation import ConversationService, MessageTag
 from cecli.helpers.similarity import (
@@ -33,6 +35,8 @@ from .base_coder import Coder
 
 from cecli.helpers.coroutines import interruptible  # isort:skip
 
+logger = logging.getLogger(__name__)
+
 
 class AgentCoder(Coder):
     """Mode where the LLM autonomously manages which files are in context."""
@@ -40,6 +44,7 @@ class AgentCoder(Coder):
     edit_format = "agent"
     prompt_format = "agent"
     context_management_enabled = True
+
     hashlines = True
     stop_on_empty = False
 
@@ -71,7 +76,7 @@ class AgentCoder(Coder):
             "edittext",
             "undochange",
         }
-        self.edit_allowed = False
+        self.edit_allowed = True
         self.max_tool_calls = 10000
         self.large_file_token_threshold = 8192
         self.skills_manager = None
@@ -92,11 +97,32 @@ class AgentCoder(Coder):
         self.skip_cli_confirmations = False
         self.agent_finished = False
         self.agent_config = self._get_agent_config()
+        self.max_sub_agents = self.agent_config.get("max_sub_agents", 3)
+        self.sub_agent_paths = self.agent_config.get("subagent_paths", [])
         self._setup_agent()
+
+        AgentService.build_registry(self.sub_agent_paths)
         ToolRegistry.build_registry(agent_config=self.agent_config)
+
         self.loaded_custom_tools = ToolRegistry.loaded_custom_tools
         super().__init__(*args, **kwargs)
 
+    def post_init(self):
+        super().post_init()
+        # Populate per-instance tool and server filters from config
+        self.registered_tools["included"] = set(
+            map(str.lower, self.agent_config.get("tools_includelist", []))
+        )
+        self.registered_tools["excluded"] = set(
+            map(str.lower, self.agent_config.get("tools_excludelist", []))
+        )
+        self.registered_servers["included"] = set(
+            map(str.lower, self.agent_config.get("servers_includelist", []))
+        )
+        self.registered_servers["excluded"] = set(
+            map(str.lower, self.agent_config.get("servers_excludelist", []))
+        )
+
     def _setup_agent(self):
         os.makedirs(".cecli/temp", exist_ok=True)
 
@@ -128,6 +154,7 @@ class AgentCoder(Coder):
         )
         config["command_timeout"] = nested.getter(config, "command_timeout", 30)
         config["hot_reload"] = nested.getter(config, "hot_reload", False)
+        config["allow_nested_delegation"] = nested.getter(config, "allow_nested_delegation", False)
 
         config["tools_paths"] = nested.getter(config, ["tools_paths", "tool_paths"], [])
         config["tools_includelist"] = nested.getter(
@@ -137,6 +164,12 @@ class AgentCoder(Coder):
             config, ["tools_excludelist", "tools_blacklist"], []
         )
 
+        config["servers_includelist"] = nested.getter(
+            config, ["servers_includelist", "servers_whitelist"], []
+        )
+        config["servers_excludelist"] = nested.getter(
+            config, ["servers_excludelist", "servers_blacklist"], []
+        )
         config["include_context_blocks"] = set(
             nested.getter(
                 config,
@@ -148,6 +181,7 @@ class AgentCoder(Coder):
                     # "git_status",
                     # "symbol_outline",
                     "todo_list",
+                    "sub_agents",
                     "skills",
                 },
             )
@@ -173,6 +207,7 @@ class AgentCoder(Coder):
             config["skills_excludelist"] = nested.getter(
                 config, ["skills_excludelist", "skills_blacklist"], []
             )
+            config["skills_init"] = nested.getter(config, ["skills_init", "skills_startup"], [])
 
         if "skills" not in self.allowed_context_blocks or not nested.getter(
             config, "skills_paths", []
@@ -193,6 +228,7 @@ class AgentCoder(Coder):
                 directory_paths=config.get("skills_paths", []),
                 include_list=config.get("skills_includelist", []),
                 exclude_list=config.get("skills_excludelist", []),
+                initialize_list=config.get("skills_init", []),
                 git_root=git_root,
                 coder=self,
             )
@@ -212,6 +248,12 @@ class AgentCoder(Coder):
             joined_skills = ", ".join(skills_list)
             self.io.tool_output(f"Available Skills: {joined_skills}")
 
+        registry = AgentService.get_registry()
+        if registry:
+            names = sorted(registry.keys())
+            joined_names = ", ".join(names)
+            self.io.tool_output(f"Available Subagents: {joined_names}")
+
     def get_local_tool_schemas(self):
         """Returns the JSON schemas for all local tools using the tool registry."""
         schemas = []
@@ -282,9 +324,7 @@ class AgentCoder(Coder):
                             content_parts.append(item.text)
                 return "".join(content_parts)
             except Exception as e:
-                self.io.tool_warning(
-                    (f"Executing {tool_name} on {server.name} failed:\nError: {e}")
-                )
+                self.io.tool_warning(f"Executing {tool_name} on {server.name} failed:\nError: {e}")
                 return f"Error executing tool call {tool_name}: {e}"
 
         result, interrupted = await interruptible(_exec_async(), self.interrupt_event)
@@ -319,6 +359,7 @@ class AgentCoder(Coder):
                 "git_status",
                 "symbol_outline",
                 "skills",
+                "sub_agents",
                 "loaded_skills",
             ]
             for block_type in block_types:
@@ -354,6 +395,10 @@ class AgentCoder(Coder):
             content = self.get_skills_context()
         elif block_name == "loaded_skills":
             content = self.get_skills_content()
+        elif block_name == "sub_agents" and (
+            not self.parent_uuid or self.agent_config.get("allow_nested_delegation", False)
+        ):
+            content = self.get_sub_agents_context()
         if content is not None:
             self.context_blocks_cache[block_name] = content
         return content
@@ -381,7 +426,10 @@ class AgentCoder(Coder):
         try:
             result = '<context name="symbol_outline" from="agent">\n'
             result += "## Symbol Outline (Current Context)\n\n"
-            result += "Code definitions (classes, functions, methods, etc.) found in files currently in chat context."
+            result += (
+                "Code definitions (classes, functions, methods, etc.) found in files currently in"
+                " chat context."
+            )
             result += "\n\n"
             files_to_outline = list(self.abs_fnames) + list(self.abs_read_only_fnames)
             if not files_to_outline:
@@ -459,7 +507,20 @@ class AgentCoder(Coder):
         ConversationService.get_chunks(self).add_file_list_reminder()
 
         # Add system messages (including examples and reminder)
-        ConversationService.get_chunks(self).add_system_messages()
+        # For sub-agents, use their specific system prompt via AgentService lookup
+        # For primary agents, use the default system messages flow
+        needs_system_prompts = True
+        if hasattr(self, "parent_uuid") and self.parent_uuid:
+            service = AgentService.get_instance(self)
+            info = service.sub_agents.get(self.uuid)
+            if info:
+                config = AgentService.get_registry().get(info.name)
+                if config and config.prompt and config.prompt.strip():
+                    ConversationService.get_chunks(self).add_system_message(config.prompt)
+                    needs_system_prompts = False
+
+        if needs_system_prompts:
+            ConversationService.get_chunks(self).add_system_messages()
 
         # Add static context blocks (priority 50 - between SYSTEM and EXAMPLES)
         ConversationService.get_chunks(self).add_static_context_blocks()
@@ -522,7 +583,10 @@ class AgentCoder(Coder):
                         )
                 if editable_files:
                     result += "\n".join(editable_files) + "\n\n"
-                    result += f"**Total editable: {len(editable_files)} files, {editable_tokens:,} tokens**\n\n"
+                    result += (
+                        f"**Total editable: {len(editable_files)} files,"
+                        f" {editable_tokens:,} tokens**\n\n"
+                    )
                 else:
                     result += "No editable files in context\n\n"
             if self.abs_read_only_fnames:
@@ -542,7 +606,10 @@ class AgentCoder(Coder):
                         )
                 if readonly_files:
                     result += "\n".join(readonly_files) + "\n\n"
-                    result += f"**Total read-only: {len(readonly_files)} files, {readonly_tokens:,} tokens**\n\n"
+                    result += (
+                        f"**Total read-only: {len(readonly_files)} files,"
+                        f" {readonly_tokens:,} tokens**\n\n"
+                    )
                 else:
                     result += "No read-only files in context\n\n"
             extra_tokens = sum(self.context_block_tokens.values())
@@ -730,7 +797,7 @@ class AgentCoder(Coder):
                 self.model_kwargs = {}
                 result_message = f"Error executing {tool_name}: {e}"
                 self.io.tool_error(
-                    (f"Error during {tool_name} execution: {e}\n{traceback.format_exc()}")
+                    f"Error during {tool_name} execution: {e}\n{traceback.format_exc()}"
                 )
             tool_responses.append(
                 {"role": "tool", "tool_call_id": tool_call.id, "content": result_message}
@@ -738,7 +805,13 @@ class AgentCoder(Coder):
 
         if self.auto_lint and used_write_tool:
             edited = list(self.files_edited_by_tools)
-            lint_errors = self.lint_edited(edited, show_output=False)
+            lint_coro = self.lint_edited(edited, show_output=False)
+            lint_errors, interrupted = await self.coroutines.interruptible(
+                lint_coro, self.interrupt_event
+            )
+            if interrupted:
+                raise KeyboardInterrupt("Interrupted during linting")
+
             self.lint_outcome = not lint_errors
 
             if lint_errors:
@@ -801,6 +874,7 @@ class AgentCoder(Coder):
         # 1. Handle Tool Execution Follow-up (Reflection)
         if self.agent_finished:
             self.tool_usage_history = []
+            self.tool_call_vectors = []
             self.reflected_message = None
             if self.files_edited_by_tools:
                 _ = await self.auto_commit(self.files_edited_by_tools)
@@ -840,7 +914,12 @@ class AgentCoder(Coder):
                 " its outputs are no longer necessary"
             )
             self.io.tool_output(waiting_msg)
-            await asyncio.sleep(command_timeout / 2)
+            sleep_coro = asyncio.sleep(command_timeout / 2)
+            _res, interrupted = await self.coroutines.interruptible(
+                sleep_coro, self.interrupt_event
+            )
+            if interrupted:
+                raise KeyboardInterrupt("Interrupted while waiting for background commands")
             return True
 
         # Check for recently finished commands that need reflection
@@ -853,11 +932,15 @@ class AgentCoder(Coder):
                 self.tool_usage_history = []
             return True
 
-        if content and not tool_calls_found and self.num_reflections < self.max_reflections:
-            self.reflected_message = (
-                "Continue with your task. If you have completed it, call the `Finished` tool."
-            )
-            return True
+        # 4. If we have called no tools (e.g. the first message)
+        # Allow early exiting
+        # If a model forgets a tool call, replay the request instead of stopping early
+        if self.tool_call_vectors:
+            if content and not tool_calls_found and self.num_reflections < self.max_reflections:
+                self.reflected_message = (
+                    "Continue with your task. If you have completed it, call the `Finished` tool."
+                )
+                return True
 
         if tool_calls_found and self.num_reflections < self.max_reflections:
             self.tool_call_count = 0
@@ -1020,8 +1103,8 @@ class AgentCoder(Coder):
             context_parts.append("## File Editing Tools Disabled")
             context_parts.append(
                 "File editing tools are currently disabled. Use `ReadRange` to determine the"
-                " current content hash prefixes needed to perform an edit and activate them when you"
-                " are ready to edit a file."
+                " current content hash prefixes needed to perform an edit and activate them when"
+                " you are ready to edit a file."
             )
 
         context_parts.append("\n\n")
@@ -1377,6 +1460,42 @@ Todo list does not exist. Please update it with the `UpdateTodoList` tool.</cont
             self.io.tool_error(f"Error generating skills content context: {str(e)}")
             return None
 
+    def get_sub_agents_context(self):
+        """
+        Generate a context block for registered sub-agents.
+        Only shown for primary coders (no parent_uuid).
+
+        Returns:
+            Formatted context block string with sub-agent names and descriptions,
+            or None if no sub-agents are registered or if called from a sub-agent.
+        """
+        if not self.use_enhanced_context:
+            return None
+        if hasattr(self, "parent_uuid") and self.parent_uuid:
+            return None
+        try:
+            registry = AgentService.get_registry()
+            if not registry:
+                return None
+
+            result = '<context name="sub_agents" from="agent">\n'
+            result += "## Available Sub-Agents\n\n"
+            result += f"Found {len(registry)} registered sub-agent(s):\n\n"
+
+            for name, config in sorted(registry.items()):
+                result += f"**{name}**:\n"
+                desc = config.metadata.get("description", "")
+                if desc:
+                    result += f"{desc}\n"
+                result += "\n"
+
+            result += "Use the `Delegate` tool with the sub-agent name to delegate tasks.\n"
+            result += "</context>"
+            return result
+        except Exception as e:
+            self.io.tool_error(f"Error generating sub-agents context: {str(e)}")
+            return None
+
     def get_background_command_output(self):
         """
         Get background command output to append after the main message.