cecli/cecli/coders/agent_coder.py at main · cecli-dev/cecli · GitHub

Repository navigation

History

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

468

469

470

471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504

505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

523

524

525

526

527

528

529

530

531

532

533

534

535

536

537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560

561

562

563

564

565

566

567

568

569

570

571

572

573

574

575

576

577

578

579

580

581

582

583

584

585

586

587

588

589

590

591

592

593

594

595

596

597

598

599

600

601

602

603

604

605

606

607

608

609

610

611

612

613

614

615

616

617

618

619

620

621

622

623

624

625

626

627

628

629

630

631

632

633

634

635

636

637

638

639

640

641

642

643

644

645

646

647

648

649

650

651

652

653

654

655

656

657

658

659

660

661

662

663

664

665

666

667

668

669

670

671

672

673

674

675

676

677

678

679

680

681

682

683

684

685

686

687

688

689

690

691

692

693

694

695

696

697

698

699

700

701

702

703

704

705

706

707

708

709

710

711

712

713

714

715

716

717

718

719

720

721

722

723

724

725

726

727

728

729

730

731

732

733

734

735

736

737

738

739

740

741

742

743

744

745

746

747

748

749

750

751

752

753

754

755

756

757

758

759

760

761

762

763

764

765

766

767

768

769

770

771

772

773

774

775

776

777

778

779

780

781

782

783

784

785

786

787

788

789

790

791

792

793

794

795

796

797

798

799

800

801

802

803

804

805

806

807

808

809

810

811

812

813

814

815

816

817

818

819

820

821

822

823

824

825

826

827

828

829

830

831

832

833

834

835

836

837

838

839

840

841

842

843

844

845

846

847

848

849

850

851

852

853

854

855

856

857

858

859

860

861

862

863

864

865

866

867

868

869

870

871

872

873

874

875

876

877

878

879

880

881

882

883

884

885

886

887

888

889

890

891

892

893

894

895

896

897

898

899

900

901

902

903

904

905

906

907

908

909

910

911

912

913

914

915

916

917

918

919

920

921

922

923

924

925

926

927

928

929

930

931

932

933

934

935

936

937

938

939

940

941

942

943

944

945

946

947

948

949

950

951

952

953

954

955

956

957

958

959

960

961

962

963

964

965

966

967

968

969

970

971

972

973

974

975

976

977

978

979

980

981

982

983

984

985

986

987

988

989

990

991

992

993

994

995

996

997

998

999

1000

import asyncio

import base64

import json

import locale

import logging

import os

import platform

import random

import time

import traceback

from collections import defaultdict

from datetime import datetime

from pathlib import Path

import xxhash

from cecli.args import AGENT_CONFIG_LIST_FIELDS

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 (

cosine_similarity,

create_bigram_vector,

normalize_vector,

)

from cecli.helpers.skills import SkillsManager

from cecli.hooks import HookIntegration

from cecli.mcp import LocalServer, McpServerManager

from cecli.tools.utils.base_tool import BaseTool

from cecli.tools.utils.registry import ToolRegistry

from cecli.utils import copy_tool_call, tool_call_to_dict

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."""

edit_format = "agent"

prompt_format = "agent"

context_management_enabled = True

hashlines = True

stop_on_empty = False

def __init__(self, *args, **kwargs):

if kwargs.get("uuid", None):

self.uuid = kwargs.get("uuid")

self.start_up_errors = []

self.recently_removed = {}

self.tool_usage_history = []

self.loaded_custom_tools = []

self.tool_usage_retries = 20

self.last_round_tools = []

self.tool_call_vectors = []

self.tool_similarity_threshold = 0.95

self.max_tool_vector_history = 20

self.read_tools = {

"command",

"explorecode",

"ls",

"readfile",

"grep",

"thinking",

"updatetodolist",

"searchfacts",

}

self.write_tools = {

"command",

"editfile",

"undochange",

"replacefacts",

}

self.edit_allowed = True

self.max_tool_calls = 10000

self.large_file_token_threshold = 8192

self.skills_manager = None

self.change_tracker = ChangeTracker()

self.args = kwargs.get("args")

self.files_added_in_exploration = set()

self.file_read_cache = set()

self.tool_call_count = 0

self.turn_count = 0

self.use_enhanced_context = True

self._last_edited_file = None

self._cur_message_divider = None

self.allowed_context_blocks = set()

self.context_block_tokens = {}

self.context_blocks_cache = {}

self.current_tasks = []

self.hot_reload_enabled = False

self.tokens_calculated = False

self.skip_cli_confirmations = False

self.agent_finished = False

init_metadata = kwargs.pop("init_metadata", {})

self.agent_config = self._get_agent_config(config=init_metadata)

self.max_sub_agents = self.agent_config.get("max_sub_agents", 30)

self.sub_agent_paths = self.agent_config.get("subagent_paths", [])

self._setup_agent()

# primary_root is needed early so the skills and tool registries can

# resolve relative paths against the primary workspace root rather than

# this coder's (possibly overridden) local root. base_coder re-affirms

# and keeps this value in super().__init__().

self.primary_root = kwargs.get("primary_root") or getattr(self, "primary_root", None)

# Allow a sub-agent to target a different working root (multi-project workspace).

if not kwargs.get("root"):

configured_root = self.agent_config.get("root")

if configured_root:

kwargs["root"] = configured_root

self._initialize_skills_manager(

self.agent_config, root=kwargs.get("root", self.primary_root)

)

AgentService.build_registry(

self.sub_agent_paths, root=kwargs.get("root", self.primary_root)

)

ToolRegistry.build_registry(

agent_config=self.agent_config, root=kwargs.get("root", self.primary_root)

)

self.loaded_custom_tools = ToolRegistry.loaded_custom_tools

super().__init__(*args, **kwargs)

def post_init(self):

super().post_init()

if not self._inherited_tools:

# 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", []))

)

for err in self.start_up_errors:

self.io.tool_warning(err)

self.start_up_errors = []

# Preserve the original 10000 setting now that max tool calls is configurable

self.max_tool_calls = self.max_tool_calls * 400

def _setup_agent(self):

os.makedirs(".cecli/temp", exist_ok=True)

def _get_agent_config(self, config=None):

"""

Parse and return agent configuration from args.agent_config.

Args:

config: Optional dict of metadata (e.g. sub-agent front matter)

that is merged on top of args.agent_config. Parameter

takes priority over args.agent_config values.

Returns:

dict: Agent configuration with defaults for missing values

"""

config = config or {}

# init_metadata provides agent-config directly, replacing args

if "agent_config" in config and config["agent_config"]:

config = config["agent_config"]

# Primary agent: load from args.agent_config

elif (

hasattr(self, "args")

and self.args

and hasattr(self.args, "agent_config")

and self.args.agent_config

try:

args_config = json.loads(self.args.agent_config)

# Validate that array fields are lists, wrap scalars in lists

list_fields = AGENT_CONFIG_LIST_FIELDS

for field in list_fields:

if field in args_config:

if args_config[field] is None:

# YAML null / JSON null -> treat as empty list

args_config[field] = []

elif not isinstance(args_config[field], list):

self.start_up_errors.append(

f"agent-config field '{field}' should be a list but got "

f"{type(args_config[field]).__name__}, wrapping in list"

)

args_config[field] = [args_config[field]]

# Merge metadata (caller-provided config) on top of

# args.agent_config so sub-agent front matter takes priority.

args_config.update(config)

config = args_config

except (json.JSONDecodeError, TypeError) as e:

self.start_up_errors.append(f"Failed to parse agent-config JSON: {e}")

return {}

config["large_file_token_threshold"] = nested.getter(

config, "large_file_token_threshold", 8192

)

config["show_lint_errors"] = nested.getter(config, "show_lint_errors", False)

config["skip_cli_confirmations"] = nested.getter(

config, "skip_cli_confirmations", nested.getter(config, "yolo", [])

)

config["command_timeout"] = nested.getter(config, "command_timeout", 30)

config["allowed_commands"] = nested.getter(config, "allowed_commands", [])

config["hot_reload"] = nested.getter(config, "hot_reload", False)

config["diff_colors"] = nested.getter(config, "diff_colors", True)

config["allow_nested_delegation"] = nested.getter(config, "allow_nested_delegation", False)

config["allow_orchestration"] = nested.getter(config, "allow_orchestration", True)

config["max_sub_agents"] = nested.getter(config, "max_sub_agents", 30)

if config["max_sub_agents"] == -1:

config["max_sub_agents"] = 2 ^ 31 - 1

config["tools_paths"] = nested.getter(config, ["tools_paths", "tool_paths"], [])

config["tools_includelist"] = nested.getter(

config, ["tools_includelist", "tools_whitelist"], []

)

config["tools_excludelist"] = nested.getter(

config,

["tools_excludelist", "tools_blacklist"],

[

"gitbranch",

"gitdiff",

"gitlog",

"gitremote",

"gitshow",

"gitstatus",

"searchfacts",

"replacefacts",

)

if config["tools_includelist"]:

config["tools_includelist"].append("yield")

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,

"include_context_blocks",

{

# "context_summary",

# "directory_structure",

"environment_info",

# "git_status",

# "symbol_outline",

"todo_list",

"sub_agents",

"skills",

"servers",

)

config["exclude_context_blocks"] = set(nested.getter(config, "exclude_context_blocks", []))

if config["allow_orchestration"]:

config["include_context_blocks"].add("orchestration")

config["exclude_context_blocks"].discard("orchestration")

else:

config["include_context_blocks"].discard("orchestration")

config["exclude_context_blocks"].add("orchestration")

config["tools_excludelist"].append("orchestrate")

self.large_file_token_threshold = config["large_file_token_threshold"]

self.skip_cli_confirmations = config["skip_cli_confirmations"]

self.hot_reload_enabled = config["hot_reload"]

self.allowed_context_blocks = config["include_context_blocks"]

for context_block in config["exclude_context_blocks"]:

try:

self.allowed_context_blocks.remove(context_block)

except KeyError:

pass

if "skills" in self.allowed_context_blocks:

config["skills_paths"] = nested.getter(config, "skills_paths", [])

config["skills_includelist"] = nested.getter(

config, ["skills_includelist", "skills_whitelist"], []

)

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", []

config["tools_excludelist"].append("loadskill")

config["tools_excludelist"].append("removeskill")

return config

def _initialize_skills_manager(self, config, root=None):

"""

Initialize the skills manager with the configured directory paths and filters.

"""

try:

self.skills_manager = SkillsManager(

directory_paths=config.get("skills_paths", []),

include_list=config.get("skills_includelist", []),

exclude_list=config.get("skills_excludelist", []),

initialize_list=config.get("skills_init", []),

root=root or self.root or self.primary_root,

coder=self,

)

except Exception as e:

self.io.tool_warning(f"Failed to initialize skills manager: {str(e)}")

def show_announcements(self):

super().show_announcements()

if self.loaded_custom_tools:

self.io.tool_output(f"Loaded custom tools: {', '.join(self.loaded_custom_tools)}")

def get_local_tool_schemas(self):

"""Returns the JSON schemas for all local tools using the tool registry."""

schemas = []

for tool_name in ToolRegistry.get_registered_tools():

tool_module = ToolRegistry.get_tool(tool_name)

if hasattr(tool_module, "SCHEMA"):

schemas.append(tool_module.SCHEMA)

return schemas

async def initialize_mcp_tools(self):

if not self.mcp_manager:

verbose = getattr(self.args, "verbose", False) if self.args else False

self.mcp_manager = McpServerManager([], self.io, verbose)

server_name = "Local"

server = self.mcp_manager.get_server(server_name)

# We have already initialized local server and its connected

# then no need to duplicate work

if server is not None and server.is_connected:

return

# If we dont have any tools for local server to use, no point in creating it then

local_tools = self.get_local_tool_schemas()

if not local_tools:

return

if server is None:

local_server_config = {"name": server_name}

local_server = LocalServer(local_server_config)

await self.mcp_manager.add_server(local_server, connect=True)

else:

await self.mcp_manager.connect_server(server_name)

async def _execute_mcp_tool(self, server, tool_name, params):

"""Helper to execute a single MCP tool call, created from legacy format."""

from cecli.tools.utils.responses import ToolResponse

async def _exec_async():

function_dict = {"name": tool_name, "arguments": json.dumps(params)}

tool_call_dict = {

"id": f"mcp-tool-call-{time.time()}",

"function": function_dict,

"type": "function",

}

try:

session = await server.connect()

call_result = await self.call_mcp_tool_from_session(session, tool_call_dict)

except Exception as e:

if server.is_session_expired_error(e):

try:

session = await server.reconnect()

call_result = await self.call_mcp_tool_from_session(session, tool_call_dict)

except Exception as retry_exc:

self.io.tool_warning(

f"Executing {tool_name} on {server.name} failed after reconnect:\n"

f"Error: {retry_exc}"

)

return f"Error executing tool call {tool_name}: {retry_exc}"

else:

self.io.tool_warning(

f"Executing {tool_name} on {server.name} failed:\nError: {e}"

)

return f"Error executing tool call {tool_name}: {e}"

try:

content_parts = []

if call_result.content:

for item in call_result.content:

if hasattr(item, "resource"):

resource = item.resource

if hasattr(resource, "text"):

content_parts.append(resource.text)

elif hasattr(resource, "blob"):

try:

decoded_blob = base64.b64decode(resource.blob).decode("utf-8")

content_parts.append(decoded_blob)

except (UnicodeDecodeError, TypeError):

name = getattr(resource, "name", "unnamed")

mime_type = getattr(resource, "mimeType", "unknown mime type")

content_parts.append(

f"[embedded binary resource: {name} ({mime_type})]"

)

elif hasattr(item, "text"):

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}")

return f"Error executing tool call {tool_name}: {e}"

result, interrupted = await interruptible(_exec_async(), self.interrupt_event)

response = ToolResponse(tool_name)

if interrupted:

response.append_error("Tool execution interrupted by user.")

elif isinstance(result, str) and result.startswith("Error executing tool call"):

response.append_error(result)

else:

response.append_result(result)

return response

def _calculate_context_block_tokens(self, force=False):

"""

Calculate token counts for all enhanced context blocks.

This is the central method for calculating token counts,

ensuring they're consistent across all parts of the code.

This method populates the cache for context blocks and calculates tokens.

Args:

force: If True, recalculate tokens even if already calculated

"""

if hasattr(self, "tokens_calculated") and self.tokens_calculated and not force:

return

self.context_block_tokens = {}

if not hasattr(self, "context_blocks_cache"):

self.context_blocks_cache = {}

if not self.use_enhanced_context:

return

try:

self.context_blocks_cache = {}

block_types = [

"environment_info",

"directory_structure",

"git_status",

"symbol_outline",

"skills",

"servers",

"sub_agents",

"orchestration",

]

for block_type in block_types:

if block_type in self.allowed_context_blocks:

block_content = self._generate_context_block(block_type)

if block_content:

self.context_block_tokens[block_type] = self.get_active_model().token_count(

block_content

)

self.tokens_calculated = True

except Exception:

pass

def _generate_context_block(self, block_name):

"""

Generate a specific context block and cache it.

This is a helper method for get_cached_context_block.

"""

content = None

if block_name == "environment_info":

content = self.get_environment_info()

elif block_name == "directory_structure":

content = self.get_directory_structure()

elif block_name == "git_status":

content = self.get_git_status()

elif block_name == "symbol_outline":

content = self.get_context_symbol_outline()

elif block_name == "context_summary":

content = self.get_context_summary()

elif block_name == "todo_list":

content = self.get_todo_list()

elif block_name == "skills":

content = self.get_skills_context()

elif block_name == "servers":

content = self.get_servers_context()

elif block_name == "orchestration":

content = self.get_orchestration_context()

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

def get_cached_context_block(self, block_name):

"""

Get a context block from the cache, or generate it if not available.

This should be used by format_chat_chunks to avoid regenerating blocks.

This will ensure tokens are calculated if they haven't been yet.

"""

if not hasattr(self, "tokens_calculated") or not self.tokens_calculated:

self._calculate_context_block_tokens()

if hasattr(self, "context_blocks_cache") and block_name in self.context_blocks_cache:

return self.context_blocks_cache[block_name]

return self._generate_context_block(block_name)

def get_context_symbol_outline(self):

"""

Generate a symbol outline for files currently in context using Tree-sitter,

bypassing the cache for freshness.

"""

if not self.use_enhanced_context or not self.repo_map:

return None

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 += "\n\n"

files_to_outline = list(self.abs_fnames) + list(self.abs_read_only_fnames)

if not files_to_outline:

result += "No files currently in context.\n"

result += "</context>"

return result

all_tags_by_file = defaultdict(list)

has_symbols = False

if not self.repo_map:

self.io.tool_warning("RepoMap not initialized, cannot generate symbol outline.")

return None

for abs_fname in sorted(files_to_outline):

rel_fname = self.get_rel_fname(abs_fname)

try:

tags = list(self.repo_map.get_tags_raw(abs_fname, rel_fname))

if tags:

all_tags_by_file[rel_fname].extend(tags)

has_symbols = True

except Exception as e:

self.io.tool_warning(f"Could not get symbols for {rel_fname}: {e}")

if not has_symbols:

result += "No symbols found in the current context files.\n"

else:

for rel_fname in sorted(all_tags_by_file.keys()):

tags = sorted(all_tags_by_file[rel_fname], key=lambda t: (t.line, t.name))

definition_tags = []

for tag in tags:

kind_to_check = tag.specific_kind or tag.kind

if (

kind_to_check

and kind_to_check.lower() in self.repo_map.definition_kinds

definition_tags.append(tag)

if definition_tags:

result += f"### {rel_fname}\n"

for tag in definition_tags:

line_info = (

f", lines {tag.start_line + 1} - {tag.end_line + 1}"

if tag.line >= 0

else ""

)

kind_to_check = tag.specific_kind or tag.kind

result += f"- {tag.name} ({kind_to_check}{line_info})\n"

result += "\n"

result += "</context>"

return result.strip()

except Exception as e:

self.io.tool_error(f"Error generating symbol outline: {str(e)}")

return None

def format_chat_chunks(self):

"""

Override parent's format_chat_chunks to include enhanced context blocks with a

cleaner, more hierarchical structure for better organization.

Optimized for prompt caching by placing context blocks strategically:

1. Relatively static blocks (directory structure, environment info) before done_messages

2. Dynamic blocks (context summary, symbol outline, git status) after chat_files

This approach preserves prefix caching while providing fresh context information.

"""

if not self.use_enhanced_context:

# Use parent's implementation which may use conversation system if flag is enabled

return super().format_chat_chunks()

# Choose appropriate fence based on file content

self.choose_fence()

ConversationService.get_chunks(self).initialize_conversation_system()

# Clean up ConversationFiles and remove corresponding messages

ConversationService.get_chunks(self).cleanup_files()

# Add reminder message with list of readonly and editable files

ConversationService.get_chunks(self).add_file_list_reminder()

# Add system messages (including examples and reminder)

# 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()

# Add rules messages

ConversationService.get_chunks(self).add_rules_messages()

# Handle file messages using conversation module helper methods

# These methods will add messages to ConversationManager

ConversationService.get_chunks(self).add_repo_map_messages()

# Add pre-message context blocks (priority 125 - between REPO and READONLY_FILES)

ConversationService.get_chunks(self).add_pre_message_context_blocks()

ConversationService.get_chunks(self).add_active_skills_messages(

getattr(self, "skills_manager", None)

)

ConversationService.get_chunks(self).add_readonly_files_messages()

ConversationService.get_chunks(self).add_chat_files_messages()

ConversationService.get_manager(self).flush_queue()

# Add post-message context blocks (priority 250 - between CUR and REMINDER)

ConversationService.get_chunks(self).add_post_message_context_blocks()

# Background command output is debounced independently so it is not

# re-dumped every turn as its contents mutate slightly

ConversationService.get_chunks(self).add_background_command_output()

# Add sub-agent states context block (same priority as post-message blocks)

ConversationService.get_chunks(self).add_sub_agent_states()

ConversationService.get_chunks(self).add_randomized_cta()

return ConversationService.get_manager(self).get_messages_dict()

def get_context_summary(self):

"""

Generate a summary of the current context, including file content tokens and additional context blocks,

with an accurate total token count.

"""

if not self.use_enhanced_context:

return None

if hasattr(self, "context_blocks_cache") and "context_summary" in self.context_blocks_cache:

return self.context_blocks_cache["context_summary"]

try:

if not hasattr(self, "context_block_tokens") or not self.context_block_tokens:

self._calculate_context_block_tokens()

result = '<context name="context_summary" from="agent">\n'

result += "## Current Context Overview\n\n"

max_input_tokens = self.get_active_model().info.get("max_input_tokens") or 0

if max_input_tokens:

result += f"Model context limit: {max_input_tokens:,} tokens\n\n"

total_file_tokens = 0

editable_tokens = 0

readonly_tokens = 0

editable_files = []

readonly_files = []

if self.abs_fnames:

result += "### Editable Files\n\n"

for fname in sorted(self.abs_fnames):

rel_fname = self.get_rel_fname(fname)

content = self.io.read_text(fname)

if content is not None:

tokens = self.get_active_model().token_count(content)

total_file_tokens += tokens

editable_tokens += tokens

size_indicator = (

"Large" if tokens > 5000 else "Medium" if tokens > 1000 else "Small"

)

editable_files.append(

f"- {rel_fname}: {tokens:,} tokens ({size_indicator})"

)

if editable_files:

result += "\n".join(editable_files) + "\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:

result += "### Read-Only Files\n\n"

for fname in sorted(self.abs_read_only_fnames):

rel_fname = self.get_rel_fname(fname)

content = self.io.read_text(fname)

if content is not None:

tokens = self.get_active_model().token_count(content)

total_file_tokens += tokens

readonly_tokens += tokens

size_indicator = (

"Large" if tokens > 5000 else "Medium" if tokens > 1000 else "Small"

)

readonly_files.append(

f"- {rel_fname}: {tokens:,} tokens ({size_indicator})"

)

if readonly_files:

result += "\n".join(readonly_files) + "\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())

total_tokens = total_file_tokens + extra_tokens

result += f"**Total files usage: {total_file_tokens:,} tokens**\n\n"

result += f"**Additional context usage: {extra_tokens:,} tokens**\n\n"

result += f"**Total context usage: {total_tokens:,} tokens**"

if max_input_tokens:

percentage = total_tokens / max_input_tokens * 100

result += f" ({percentage:.1f}% of limit)"

if percentage > 80:

result += "\n\n⚠ **Context is getting full!**\n"

result += "- Remove non-essential files, skills and tool servers via the `ResourceManager` tool.\n"

result += "- Keep only essential files, skills, and MCP servers in context for best performance"

result += "\n</context>"

if not hasattr(self, "context_blocks_cache"):

self.context_blocks_cache = {}

self.context_blocks_cache["context_summary"] = result

return result

except Exception as e:

self.io.tool_error(f"Error generating context summary: {str(e)}")

return None

def get_environment_info(self):

"""

Generate an environment information context block with key system details.

Returns formatted string with working directory, platform, date, and other relevant environment details,

including the agent's own identity and the primary agent's UUID.

"""

if not self.use_enhanced_context:

return None

try:

current_date = datetime.now().strftime("%Y-%m-%d")

platform_info = platform.platform()

language = self.chat_language or locale.getlocale()[0] or "en-US"

result = '<context name="environment_info" from="agent">\n'

result += "## Environment Information\n\n"

result += f"- Working directory: {self.root}\n"

result += f"- Current date: {current_date}\n"

result += f"- Platform: {platform_info}\n"

result += f"- Language preference: {language}\n"

# Agent identity so the model can recognise itself and the primary

# agent. Kept here (a static context block) rather than in the

# sub-agent states block to avoid re-injecting an agent's own

# identity whenever sub-agent state changes mid-conversation.

service = AgentService.get_instance(self)

self_uuid = str(self.uuid)

primary_uuid = str(service.coder.uuid)

self_name = service.get_agent_name(self) or "primary"

result += f"- Agent: {self_name} ({self_uuid}) [self]\n"

result += f"- Primary agent: primary ({primary_uuid})\n"

if self.repo:

try:

rel_repo_dir = self.repo.get_rel_repo_dir()

num_files = len(self.repo.get_tracked_files())

result += f"- Git repository: {rel_repo_dir} with {num_files:,} files\n"

except Exception:

result += "- Git repository: active but details unavailable\n"

if self.mcp_manager and self.mcp_manager.connected_servers:

num_mcp_servers = len(self.mcp_manager.connected_servers)

result += f"- Connected MCP servers: {num_mcp_servers}\n"

result += "- Git repository: none\n"

result += "</context>"

return result

except Exception as e:

self.io.tool_error(f"Error generating environment info: {str(e)}")

return None

async def process_tool_calls(self, tool_call_response):

"""

Track tool usage before calling the base implementation.

"""

await self.auto_save_session()

self.last_round_tools = []

self.turn_count += 1

if self.partial_response_tool_calls:

for tool_call in self.partial_response_tool_calls:

tool_name = getattr(tool_call.function, "name", None)

tool_call_copy = tool_call_to_dict(copy_tool_call(tool_call))

if "id" in tool_call_copy:

del tool_call_copy["id"]

if tool_name:

self.last_round_tools.append(tool_name)

content = (

str(self.partial_response_content) if self.partial_response_content else ""

)

tool_call_str = str(tool_call_copy)

tool_vector = create_bigram_vector((tool_call_str, content))

tool_vector_norm = normalize_vector(tool_vector)

self.tool_call_vectors.append(tool_vector_norm)

if self.last_round_tools:

self.tool_usage_history += self.last_round_tools

self.tool_usage_history = list(filter(None, self.tool_usage_history))

if len(self.tool_usage_history) > self.tool_usage_retries:

self.tool_usage_history.pop(0)

if len(self.tool_call_vectors) > self.max_tool_vector_history:

self.tool_call_vectors.pop(0)

# Ensure we call base implementation to trigger execution of all tools (native + extracted)

return await super().process_tool_calls(tool_call_response)

async def _execute_local_tools(self, tool_calls_list):

"""Execute local tools via ToolRegistry."""

tool_responses = []

used_write_tool = False

for tool_call in tool_calls_list:

tool_name = tool_call.function.name

result_message = ""

self.io.tool_call_event(tool_call.id, "in_progress")

try:

if responses.unprefix_tool_name(tool_name)[1].lower() in self.write_tools:

used_write_tool = True

args_string = tool_call.function.arguments.strip()

parsed_args_list = []

if not await HookIntegration.call_pre_tool_hooks(self, tool_name, args_string):

tool_responses.append(

{

"role": "tool",

"tool_call_id": tool_call.id,

"content": "Tool Request Aborted.",

}

)

continue

if args_string:

parsed = responses.parse_tool_arguments(args_string)

if isinstance(parsed, dict) and "@error" in parsed:

self.io.tool_warning(

f"Malformed JSON arguments in tool {tool_name}: {parsed['@error']}"

)

tool_responses.append(

{

"role": "tool",

"tool_call_id": tool_call.id,

"content": (

f"Malformed JSON arguments in tool {tool_name}: {parsed['@error']}"

}

)

continue

parsed_args_list = [parsed]

if not parsed_args_list and not args_string:

parsed_args_list.append({})

all_results_content = []

norm_tool_name = tool_name.lower()

tasks = []

if norm_tool_name in ToolRegistry.get_registered_tools():

tool_module = ToolRegistry.get_tool(norm_tool_name)

for params in parsed_args_list:

result = tool_module.process_response(self, params)

if asyncio.iscoroutine(result):

tasks.append(result)

else:

tasks.append(asyncio.to_thread(lambda: result))

else:

all_results_content.append(f"Error: Unknown tool name '{tool_name}'")

if tasks:

async def gather_and_await():

return await asyncio.gather(*tasks, return_exceptions=True)

task_results, interrupted = await interruptible(

gather_and_await(), self.interrupt_event

)

if interrupted:

self.io.tool_warning("Tool execution interrupted.")

all_results_content.append("Tool execution interrupted by user.")

elif task_results:

for res in task_results:

if isinstance(res, Exception):

all_results_content.append(f"Error in tool execution: {res}")

else:

all_results_content.append(str(res))

if not await HookIntegration.call_post_tool_hooks(

self, tool_name, args_string, "\n\n".join(all_results_content)

tool_responses.append(

{

"role": "tool",

"tool_call_id": tool_call.id,

"content": "Tool Response Redacted.",

}

)

continue

result_message = "\n\n".join(all_results_content)

except Exception as e:

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()}"

)

tool_responses.append(

{"role": "tool", "tool_call_id": tool_call.id, "content": result_message}

)

if self.auto_lint and used_write_tool:

edited = list(self.files_edited_by_tools)

lint_coro = self.lint_edited(edited, show_output=False)

lint_errors, interrupted = await interruptible(lint_coro, self.interrupt_event)

if interrupted:

# Abort the turn with CancelledError, which the linear run loop

# handles by re-prompting for input. KeyboardInterrupt is a

# BaseException that would escape the worker's event loop and

# leave it unable to accept further prompts.

raise asyncio.CancelledError("Interrupted during linting")

has_errors = False

if self.lint_outcome is False:

has_errors = True

self.lint_outcome = not lint_errors

if lint_errors:

lint_errors = lint_errors.replace(

"# Fix any linting errors below, if possible.",

"# Fix the linting errors below, and then continue with your task.",

)

lint_hash = xxhash.xxh3_128_hexdigest(lint_errors.encode("utf-8", errors="replace"))

ConversationService.get_manager(self).add_message(

message_dict=dict(role="user", content=lint_errors),

tag=MessageTag.LINT,

hash_key=("lint_errors", "agent", lint_hash),

)

ConversationService.get_manager(self).queue_message(

message_dict=dict(

role="user", content="Please address the latest linting errors."

tag=MessageTag.LINT,

hash_key=("lint_errors", "agent", lint_hash, "cta"),

promotion=ConversationService.get_manager(self).DEFAULT_TAG_PROMOTION_VALUE,

mark_for_demotion=1,

)

else:

if has_errors:

ConversationService.get_manager(self).add_message(

message_dict=dict(

role="user",

content=(

'<context name="linting_confirmation" from="agent">'

"All linting errors resolved."

"</context>"

tag=MessageTag.LINT,

hash_key=("lint_errors", "agent", str(time.monotonic_ns())),

)

return tool_responses

async def _execute_mcp_tools(self, server, tool_calls):

"""Execute MCP tools via LiteLLM."""

tool_responses = []

for tool_call in tool_calls:

# Use existing _execute_mcp_tool logic

result = await self._execute_mcp_tool(

server,

View remainder of file in raw view

cecli/cecli/coders/agent_coder.py at main · cecli-dev/cecli · GitHub

Repository navigation

Expand file tree

Breadcrumbs

agent_coder.py

Latest commit

History

Breadcrumbs

agent_coder.py

File metadata and controls