Handle input that starts with /.

Check if more searches are allowed

Check if CLI optional dependencies are installed.

Check if we have received input from all expected sources

Check if any stop condition is met # Arguments `message` - The latest assistant message `iteration` - Current iteration count `state_json` - Serializ

Generate the pattern for listing checkpoints

Check if checkpointing is enabled

ACP 서버 실행용 동기 CLI 엔트리포인트입니다.

Collect SKILL.md files to validate

모든 미들웨어의 도구 수집

Compute all active vertices in parallel Returns (updates, outboxes, newly_halted_vertex_ids)

Add conditional edges from a node.

Convert a slice of DeepAgents messages to Rig format Filters out messages that cannot be converted (e.g., system messages that should be handled via

Copy selected text from app widgets to clipboard. This queries all widgets for their text_selection and copies any selected text to the syste

Count tokens for a single message

Count tool usage across all steps in a trajectory. Args: trajectory_path: Path to the trajectory.json file in ATIF format Returns:

ActionRequest 생성

Create a checkpoint at the current state # Design Decision (Gemini/Qwen Review) Uses `Checkpoint::with_retry_counts()` to store retry state as a fir

Create a LangSmith dataset from Harbor tasks. Args: dataset_name: Dataset name (used for both Harbor download and LangSmith dataset)

Synchronous wrapper for create_experiment_async.

Create a LangSmith experiment session for the given dataset. Args: dataset_name: Name of the LangSmith dataset to create experiment for

Create a minimal ToolRuntime for tool execution

Get current frame without advancing.

모든 기본 도구 반환 (SubAgent task 제외)

Delete thread checkpoints. Returns True if deleted.

CLI handler for: deepagents threads delete.

Deliver pending messages to vertex inboxes

Determine the next phase based on current state.

Discover all skills in a directory. Args: skills_dir: Path to skills directory. Returns: List of skill metadata dictionaries

파일 내 문자열을 치환하여 편집합니다. `EditResult`를 반환합니다. 외부 스토리지 백엔드는 `files_update=None`을 사용합니다.

Ensure the global agent directory exists and return its path. Args: agent_name: Name of the agent Returns: P

Ensure the checkpoint directory exists

Ensure project-level skills directory exists and return its path. Returns: Path to {project_root}/.deepagents/skills/, or None if

Estimate how many messages (from the end) fit within a token budget.

Evaluate a branch condition against a value

Python 코드를 WASM 샌드박스에서 실행합니다. 실제 구현에서는 Node.js subprocess나 WebWorker를 통해 Pyodide를 실행합니다.

Execute a single HTTP request

Execute skills subcommands based on parsed arguments. Args: args: Parsed command line arguments with skills_command attribute

Execute a spec-based subagent

Execute a task with output directed to Textual UI. This is the Textual-compatible version of execute_task() that uses the TextualUIAdapter fo

Execute a tool and return the result

도구 호출 실행

Execute HTTP request with retry and backoff

Anthropic 응답에서 캐시 메트릭을 추출합니다.

DeepSeek 응답에서 캐시 메트릭을 추출합니다.

발견 항목들에서 AI 관련 키워드를 추출한다. 사전 정의된 AI 키워드 목록을 기반으로 각 키워드의 출현 빈도를 계산합니다. Args: findings: 분석할 Finding 객체 목록. Returns: 키

OpenAI 응답에서 캐시 메트릭을 추출합니다.

Extract the task instructions from the trajectory file. Looks for the user message in the trajectory steps.

Extract task metadata from config.json and other files. Args: trial_dir: Path to the trial directory Returns: Dictionary con

웹페이지를 가져와서 HTML을 Markdown으로 변환한다. 이 헬퍼 함수는 HTTP GET 요청을 수행하고(브라우저와 유사한 User-Agent 사용), 응답 상태 코드를 검증한 후, `markdownify`를 사용하여 반환된 HTML을 Mar

Filter tools based on allowed list

Find the cutoff index for message partitioning. Ensures we don't split in the middle of an AI/Tool message pair.

Find a safe cutoff point that doesn't split AI/Tool pairs. If the initial cutoff lands on a Tool message, advance past all consecutive Tool messages

Find the task directory for a given trial. Args: trial_dir: Path to the trial directory task_name: Name of the task (from config.

Find vertices with no outgoing edges (terminal vertices).

Focus the input field.

Format skills as JSON.

Format skills as markdown table.

메시지 내용을 화면에 표시 가능한 문자열로 변환한다.

Rich 포매팅을 사용해 메시지 목록을 출력한다.

Format the search results based on the specified output format. Parameters ---------- papers : list[dict[str, Any]] List of paper

Format skills as plain text.

Convert ToolMessage content into a printable string.

Generate tool description with available agents

Generate JSON schema for parameters

연구 결과 보고서를 Markdown 형식으로 생성한다. Args: session: 연구 세션 객체. keywords: 키워드 -> 빈도 매핑. output_path: 보고서 저장 경로.

Generate a new 8-char hex thread ID.

Get a subagent by name

모든 SubAgent 정의를 목록으로 반환한다. Returns: SubAgent 명세 딕셔너리 목록. Note: 도구는 비어 있으며 런타임에서 사용 가능한 도구를 기반으로 에이전트 생성 시 채워져야 한다.

이 tool에 대한 승인 위젯 클래스와 데이터를 반환합니다. Args: tool_args: The tool arguments from action_request Returns: Tuple of

Get list of available sandbox provider types. Returns: List of sandbox type names (e.g., ["modal", "runloop", "daytona"])

Return toolbar function that shows auto-approve status and BASH MODE.

Get cached summaries synchronously (for modify_system_prompt)

Get the recommended chars_per_token for a model name. Returns a model-specific ratio for more accurate token counting. # Arguments `model` - Model

Get AsyncSqliteSaver for the global database.