Built-in Tools Reference

DesireCore built-in tools evolve with the client. This page groups common tools by capability area; the exact tools available to an agent depend on the agent, enabled skills, permissions, and platform.

Files and Search

Tool	Function	Risk	Confirmation
Read	Read text, images, PDFs, and large file ranges	Low	No
Write	Create or overwrite files	Medium	Usually
Edit	Replace precise text ranges	Medium	Usually
NotebookEdit	Edit Jupyter Notebook cells	Medium	Usually
Ls	List directory contents	Low	No
Glob	Search by filename pattern	Low	No
Grep	Search file contents	Low	No
ToolSearch / DiscoverTools	Inspect authorized tools / load declarations on demand	Low	No

Read chooses a strategy by file type. PDFs may be read as text, rendered page by page for vision models, or loaded by range for large files.

System and Workspace

Tool	Function	Risk	Confirmation
Bash	Run shell commands	High	Yes
PowerShell	Run PowerShell commands on Windows	High	Yes
Which	Check command availability	Low	No
ManageWorkDirs	Manage working directories	Medium	Usually
GenerateUUID	Generate UUIDs	Low	No
JsonRepair	Repair common JSON formatting problems	Low	No

Under the default approval mode, command execution asks you to review the full command before it runs (modes such as External Tools Only or Allow All follow their own rules — see the approval modes in the security settings). Work directories provide the agent's default operating context; if you enable "restrict to work directories" in an agent's configuration, file reads and writes are strictly confined to the work-directory whitelist (with it disabled — the default — the agent may access regular local files while sensitive paths stay protected).

Web and Browser

Tool	Function	Risk	Confirmation
WebFetch	Fetch a webpage as Markdown	Low	No
WebSearch	Search the web	Low	No
HttpRequest	Send HTTP requests, primarily for Windows environments	Medium	Depends
BrowserAct	Navigation, input, waiting, screenshots, and browsing interactions	Low-Medium	Depends
BrowserDevtools	Network, site storage, console, performance, and raw CDP	Low-High	Depends
BrowserResume	Resume browser operations after the user asks to continue	Low	Verifies a new user turn
SitePattern*	Read or write site usage patterns	Low-Medium	May be needed
LocalBookmarks	Search local browser bookmarks/history hints	Low	No

Browser interaction and developer tools, SitePattern, and LocalBookmarks are Web Access skill-scoped capabilities. BrowserResume is a base entry for resuming a user-paused session. See Web Access.

Collaboration and Tasks

Tool	Function	Risk	Confirmation
Delegate	Delegate work to another agent, team, or preset flow	Medium	Depends
SpawnAgent	Create a temporary sub-agent	Medium	Usually
Handoff	Transfer work to another agent	Medium	Usually
RequestHelp	Ask a persistent agent for help	Medium	Usually
SendMessage / SendUserMessage	Communicate between agents or proactively message the user	Low	No
AskUserQuestion	Ask the user a structured question	Low	No
ManageTeam	Create or adjust an agent team	Medium	Depends
TaskBoardCreate	Create a team task board in the current trusted team context and choose its review/responsibility policy	Low	No
TaskCreate / TaskUpdate	Register and maintain tasks; state changes require run binding and authority	Low	No
TaskResponsibility	Assignment, claim, handoff, and staged review, subject to server-authorized actions	Medium	Authority checks
TaskList / TaskGet	Query task board state	Low	No

TaskBoardCreate creates a team task board only when trusted team context is available. The agent uses the returned board ID when it creates tasks. An ordinary TaskCreate without a board ID still uses the basic policy board.

Task tool sequence:

1. Register the task with TaskCreate and obtain its taskRef. This does not claim or start the task.
2. Assign the executor, then start execution through Delegate with explicit taskRef and contextMode.
3. After server admission checks pass, the executor can claim the task, advance its state, or submit it for review.

Use TaskGet to inspect currently available actions and rejection reasons.

Context, Memory, and Skills

Tool	Function	Risk	Confirmation
RecallConversation	Retrieve conversation history	Low	No
ManageContext	Pin/discard conversation content, clear history, adjust context budget	Medium	Depends
CompactSession	Compact long conversation context	Low	No
Skill	Load or invoke a skill	Depends	Depends

ManageContext lets an agent actively curate its own conversation context: pin key content it wants to keep around, discard tool results it no longer needs, collapse history down to just the system layer plus whatever is pinned, or temporarily adjust how much context budget the current session is allowed to use. None of this physically deletes the original record — the full history remains available to look back on (actions with information loss, such as clearing history or discarding a tool result, require confirmation; other operations do not).

Automation

Tool	Function	Risk	Confirmation
CreateSchedule	Create delay, at, interval, or cron schedules	Medium	Yes
HeartbeatRespond	Submit heartbeat outcome, notifications, and file links	Low	No
WorkflowCreate	Create workflow definitions	Medium	Usually
WorkflowValidate / WorkflowTest	Validate or test workflow definitions	Low	No
WorkflowRun	Run workflows	Medium	Usually
WorkflowView	Open the workflow canvas for inspection	Low	No

Documents and Media

Tool	Function	Risk	Confirmation
ExportDocument	Export PDF, DOCX, or related formats	Low-Medium	Depends
GenerateImage	Generate or edit images	Medium	Depends on provider/cost
GenerateVideo	Generate videos	Medium	Depends on provider/cost
BeautifyImage	Improve, crop, center, or optimize images	Medium	Depends
UnderstandImage	Use a vision provider	Low-Medium	Depends
MathCalc	Deterministic high-precision calculation	Low	No

Media generation tools require a configured image_gen, video_gen, or image processing provider. See Media Generation.

MCP Resources

Tool	Function	Risk	Confirmation
McpListResources / McpReadResource	List and read MCP resources	Low	No
McpListPrompts / McpGetPrompt	List and expand MCP prompts	Low	No

MCP tools come from connected external services. Their permissions, risks, and approval behavior depend on the service declaration and DesireCore local permission policy.