Tools Reference

Overview

Tools are capabilities agents use to interact with your system, files, browser, and other agents. This is the complete reference of all available tools.

Each agent can be configured with specific tool permissions based on the execution mode. Use this list to:

• Understand what tools your agents can access
• Create custom permission rules to allow/deny specific tools
• Plan what agents need for their tasks

Quick Tool Index

System & Web

• Bash — Execute terminal commands
• Browser — Control browser automation
• Screenshot — Capture screen
• WebFetch — Fetch & convert URLs to markdown
• WebSearch — Search the web
• Feedback — Send feedback to developers

Agent Workspace Files

• List — List files and directories in agent workspace
• Read — Read files from agent workspace
• Write — Write/create files in agent workspace
• Delete — Delete files from agent workspace
• Grep — Search for patterns in files

Secrets & Security

• vault_list — List secret names
• vault_save — Prompt for secret input

Agents

• agent_spawn — Create a sub-agent
• agent_message — Send message to agent
• agent_list — List all agents
• agent_delete — Delete an agent
• agent_rename — Rename an agent
• agent_context_compact — Compress history

Scheduling & Triggers

• schedule_once — Schedule one-time task
• schedule_list — List scheduled tasks
• schedule_cancel — Cancel scheduled task
• create_script_trigger — Create a script trigger
• list_script_triggers — List script triggers
• delete_script_trigger — Delete a script trigger
• create_webhook — Create a webhook trigger
• list_webhooks — List webhooks
• delete_webhook — Delete a webhook
• trigger_list — List available triggers
• trigger_subscribe — Subscribe to trigger
• trigger_unsubscribe — Unsubscribe from trigger
• trigger_subscriptions — List subscriptions

Planning & User Interaction

• update_planning_document — Update plan in plans directory
• finalize_plan — Finalize plan and request approval
• user_question — Ask user questions
• search_history — Search conversation history
• check_provider_status — Check AI provider status

---

System Tools

Shell (Bash)

Execute terminal commands on your machine.

Parameter	Type	Description
command	string	Single command to execute
commands	array	Multiple commands (parallel by default)
sequential	boolean	Run commands sequentially

Supports streaming output for long-running commands. Secrets from the vault are automatically masked in output.

Browser (Browser)

Control browser automation for web interactions.

Parameter	Type	Description
action	string	Single action to perform
actions	array	Sequential actions

Available Actions:

• navigate — Go to a URL
• click — Click an element
• type — Type text into an element
• scroll — Scroll the page
• get_content — Extract page content
• wait — Wait for element or condition

Screenshot (Screenshot)

Capture a screenshot of the display.

Requires explicit user approval. No parameters needed.

Web Fetch (WebFetch)

Fetch content from a URL and convert to markdown.

Parameter	Type	Description
url	string	URL to fetch

Returns the page content converted to markdown format.

Web Search (WebSearch)

Search the web using DuckDuckGo.

Parameter	Type	Description
query	string	Search query
max_results	number	Maximum results (default 10, max 25)

Returns search results with titles, URLs, and snippets.

Feedback (Feedback)

Send feedback to Fleet developers.

Parameter	Type	Description
message	string	Feedback message
screenshot_path	string	Optional screenshot path

---

Agent Workspace Tools

Each agent has a dedicated workspace directory for files. These tools only operate on files within the agent's workspace — they cannot access files outside the workspace directory.

List Files (List)

List files and directories in the agent's workspace only.

Parameter	Type	Description
path	string	Relative path (default: workspace root)

Cannot access files outside the workspace directory.

Read Files (Read)

Read files from the agent's workspace only.

Parameter	Type	Description
path	string	Single file path
paths	array	Multiple file paths (batch read)

For batch reads, each item should have id and path fields. Cannot access files outside the workspace directory.

Write Files (Write)

Write or overwrite files in the agent's workspace only.

Parameter	Type	Description
path	string	File path to write
content	string	Content to write

Creates parent directories automatically. Cannot write files outside the workspace directory.

Delete Files (Delete)

Delete files or directories from the agent's workspace only.

Parameter	Type	Description
path	string	Path to delete
recursive	boolean	Delete directories recursively

Cannot delete files outside the workspace directory.

---

Vault Tools

Agents can securely manage secrets stored in the Secure Vault. Secrets are encrypted, never exposed to the LLM provider, and never appear in conversation history.

See Secure Vault for details on storing and managing secrets manually.

List Secrets (vault_list)

List all available secret names in the vault.

Returns only the names of secrets, never their values. The agent can use this to discover what credentials are available.

Save Secret (vault_save)

Request the user to securely input a new secret.

Parameter	Type	Description
keys	array	List of secret names to request
key	string	Single secret name (deprecated)

When called, the user receives a secure prompt to input the secret value. The value:

• Is encrypted using your system keychain
• Never appears in chat history
• Is never sent to the LLM provider
• Can be used immediately in subsequent tool calls via {{vault:SECRET_NAME}}

Example: An agent deploying an application might call vault_save to securely request AWS credentials if they're not already stored.

---

Agent Management Tools

Spawn Agent (agent_spawn)

Create a new sub-agent to delegate work.

Parameter	Type	Description
name	string	Name for the sub-agent
prompt	string	Initial instructions
objective	string	What the sub-agent should accomplish

Returns the sub-agent ID. The parent waits for the sub-agent to complete.

Message Agent (agent_message)

Send a message to any agent in the system.

Parameter	Type	Description
recipient_id	string	Target agent ID
content	string	Message content

Messages wake up idle agents. Use agent_list to discover agent IDs.

List Agents (agent_list)

List all agents in the system with their hierarchy.

Returns a tree view showing:

• Agent IDs and names
• Current state (running, idle, etc.)
• Parent/child relationships

Delete Agent (agent_delete)

Permanently delete an agent.

Parameter	Type	Description
sub_agent_id	string	Agent to delete

Removes the agent, its logs, and workspace. Irreversible.

Rename Agent (agent_rename)

Change an agent's name.

Parameter	Type	Description
name	string	New name

Compact Context (agent_context_compact)

Compress conversation history to free token space.

Parameter	Type	Description
summary	string	Optional custom summary

Useful for long-running agents approaching token limits.

---

Scheduling Tools

Schedule Once (schedule_once)

Schedule a one-time task for future execution.

Parameter	Type	Description
name	string	Name for the task
prompt	string	Message to send when triggered
trigger_at	string	ISO 8601 timestamp
target_agent_id	string	Target agent (optional, defaults to self)

List Schedule (schedule_list)

List pending scheduled tasks.

Cancel Schedule (schedule_cancel)

Cancel a scheduled task.

Parameter	Type	Description
task_id	string	ID of the task to cancel

---

Trigger Tools

Create Script Trigger (create_script_trigger)

Create a local script trigger that runs on poll or daemon mode.

Parameter	Type	Description
name	string	Unique trigger name (required)
mode	string	poll or daemon (required)
script	string	Bash script content (required)
description	string	What the trigger monitors
interval	string	Poll interval, e.g., "5m" (poll mode only)
cooldown	string	Throttle time, e.g., "30s" (daemon mode only)
env	object	Environment variables. Use {{vault:SECRET}} for secrets

List Script Triggers (list_script_triggers)

List all script triggers with their status (enabled/disabled).

Delete Script Trigger (delete_script_trigger)

Delete a script trigger.

Parameter	Type	Description
name	string	Trigger name to delete

Create Webhook (create_webhook)

Create a webhook URL for external services. Requires Fleet Gateway connection.

Parameter	Type	Description
name	string	Descriptive name for the webhook

Returns the webhook URL to give to external services.

List Webhooks (list_webhooks)

List all webhooks created by the agent.

Delete Webhook (delete_webhook)

Delete a webhook.

Parameter	Type	Description
webhook_id	string	Webhook ID to delete

List Triggers (trigger_list)

List all available triggers in the system.

Subscribe (trigger_subscribe)

Subscribe the agent to a trigger.

Parameter	Type	Description
trigger_name	string	Trigger name (use webhook:<name> for webhooks)
prompt_template	string	Template with {{output}} placeholder

Unsubscribe (trigger_unsubscribe)

Remove a trigger subscription.

Parameter	Type	Description
subscription_id	string	Subscription to remove

List Subscriptions (trigger_subscriptions)

List current trigger subscriptions for the agent.

---

Planning Tools

Update Planning Document (update_planning_document)

Update the planning document during plan mode.

Parameter	Type	Description
content	string	New document content
explanation	string	Explanation of changes

Finalize Plan (finalize_plan)

Mark the planning session as complete and ready for user approval.

Parameter	Type	Description
summary	string	Brief 1-2 sentence description of the plan

Signals that the plan document is complete and ready for the user to review and approve before execution begins.

---

User Interaction Tools

User Question (user_question)

Ask the user one or more questions.

Parameter	Type	Description
questions	array	List of questions

Each question has:

• id — Unique identifier
• text — Question text
• type — Question type (single_choice, multiple_choice, text)
• options — Available choices (for choice types)

Returns the user's answers.

---

Provider Tools

Check Provider Status (check_provider_status)

Check the status of AI providers.

Returns information about which providers are configured and available.

---

Workspace Search

Grep (Grep)

Search for patterns in workspace files.

Parameter	Type	Description
pattern	string	Pattern to search for
path	string	File or directory path (optional)
context_lines	number	Lines of context before/after matches
max_matches	number	Maximum matches to return

---

History Tools

Search History (search_history)

Search the agent's conversation history.

Parameter	Type	Description
query	string	Search query

Returns matching messages from the conversation history.

---

Tool Results Storage

When tools produce large outputs, the results are automatically saved to disk in the agent's workspace under a tool_results folder. This includes:

• Large file reads
• Web page fetches
• Search results
• Screenshot captures
• Command outputs

This keeps the chat history lean while preserving full access to tool outputs in the workspace.

---

Batch Operations

Many tools support batch operations for efficiency:

Read multiple workspace files:

Read({ paths: [
  { id: "1", path: "file1.txt" },
  { id: "2", path: "file2.txt" }
]})

Execute commands in parallel:

Bash({ commands: ["npm test", "npm run lint"] })

Execute commands sequentially:

Bash({
  commands: ["npm install", "npm build"],
  sequential: true
})

---

Tool Permissions & Access Control

Tools are controlled by the execution mode (see Security & Permissions):

Mode	Tools Available
Restricted	Bash (safe commands), Read, List, Screenshot, Browser (view only)
Read-Only	Above + Grep, WebFetch, WebSearch
Read/Write	Above + Write, Delete, agent tools, vault, schedule, triggers

Custom Permission Rules

Beyond execution modes, you can create custom allow lists to fine-tune tool access. Use custom rules to:

• Expand permissions — Allow specific tools in restricted modes (e.g., allow npm commands in Restricted)
• Restrict operations — Deny specific tools even in Open mode (e.g., prevent file deletion)
• Pattern matching — Allow tools only for specific paths or commands

See Custom Permission Rules for details on creating rules.

Tool Approval

Tools requiring approval will pause the agent and prompt you before executing. You can approve once, always allow (which creates a rule), or deny the operation.

---

MCP Tools

When MCP (Model Context Protocol) servers are connected, their tools become available to agents. MCP tools are prefixed with the server name:

mcp_<server>_<tool>

Examples:

• mcp_github_create_issue — Create a GitHub issue
• mcp_filesystem_read_file — Read a file via filesystem server
• mcp_postgres_query — Query a PostgreSQL database

Configuring MCP Access

Control which MCP servers and tools agents can access in their AGENT.md:

mcp_servers:
  - github
  - filesystem

mcp_permissions:
  github:
    allowed_tools: ["*"]
    denied_tools: ["delete_repo"]

mcp_approval_required:
  - github_merge_pr

Common MCP Servers

Server	Tools	Description
github	search_code, create_issue, get_pull_request, etc.	GitHub API access
filesystem	read_file, write_file, list_directory	File system access
postgres	query, list_tables, describe_table	PostgreSQL database
slack	send_message, list_channels	Slack messaging

See MCP Servers for full configuration details.

---

Git Hook Tools

When agents are triggered by git hooks, they receive context about the git operation:

Context Field	Description
changed_files	Files changed in the commit/push
commit_msg	Commit message (for commit hooks)
branch	Current branch name
remote	Remote name (for push hooks)

Agents can use standard Bash commands with git:

git diff --cached
git log --oneline -5
git status

See Trigger Types - Git Hook for configuration.

---

File Watcher Context

When agents are triggered by file watchers, the trigger output contains:

Field	Description
path	Path to the changed file
event	Type of change (created, modified, deleted)

Access via the {{output}} template variable in trigger bindings.

See Trigger Types - File Watcher for configuration.