Tool Reference

This page catalogs the tools the Team Lead agent can call. The list is the same tool surface the model is shown at request time, organized by category.

Tools fall into two groups. The static catalog is the Team Lead's built-in capabilities for managing tasks, files, memory, web access, messaging, triggers, and connections. The dynamic catalog is the per-org ability tools that appear once you activate them on a connection (see Abilities). This reference covers the static catalog only; activated ability tools vary by org and are documented by the underlying service.

Tasks

The agent's task list is visible in the UI; users can check and uncheck items.

list_tasks

List tasks in the agent's task list. With no arguments, returns every task. With taskIds, returns only the matching tasks. Each task carries taskId, title, description, status, and timestamps.

manage_tasks

Batch add, update, or delete tasks from the agent's task list. Each operation is processed in order.

Filesystem

The agent has its own workspace at /agent/.... Writes are blocked under /agent/blocks, /agent/uploads, /agent/skills, and /agent/toolcalls.

read_file

Read a file from the agent workspace. Output uses the line-prefix format ( 1→content) with 1-based line numbers. Optional start_line / end_line slice the output (negative values count from the end of the file, e.g. start_line=-50 reads the last 50 lines). PDF support via pdf_start_page / pdf_end_page / pdf_format is reserved on the schema but not yet wired in this build.

write_file

Write, edit, or delete a file in the agent workspace. op=write replaces the file with content. op=edit performs a single-anchor replacement of old_string with new_string; if old_string matches more than once, the call fails unless replace_all=true. An empty new_string deletes the matched span. op=delete removes the file.

Memory

Each agent has a private SQLite database for long-lived structured memory.

get_agent_db_schema

List every table in the agent's database with its row count and CREATE TABLE DDL. Returns {"tables": [{"name": ..., "rowCount": ..., "sql": ...}], "tableCount": N}. No parameters.

run_agent_memory_sql

Run one SQL statement against this agent's private SQLite database. SELECT returns a JSON array of row objects keyed by column name. INSERT / UPDATE / DELETE / CREATE / DROP / etc. return {"rowsAffected": N}. Multiple statements per call are rejected; split into separate calls.

Web

web_search_web

Search the public web. Returns a JSON array of {title, url, snippet, position} objects ordered by relevance. limit caps the number of hits (default 10). location may be a city or country string the backend uses for regional bias.

web_scrape_website

Fetch a public URL and return its readable content as Markdown. Uses a 10-second HTTP timeout. Strips boilerplate (nav, footer, scripts, styles) before conversion.

Preview

show_user_preview

Open the right-side preview pane and show a file. Phase 4 supports kind: "file" only; app, computer, and browser are reserved for Phase 6. Pass an avfs:///agent/... filepath the user can browse.

Messaging

The Team Lead can send and reply to email and SMS through verified contact methods. Contact methods belong to the org and are reused across sessions.

list_contact_methods

List every contact method registered for this org. Returns [{type, value, name, verified}]. Workspace member emails are pre-verified.

add_contact_method

Register a new contact method (email or phone). Sends a verification challenge to the address (link for email, "reply YES" for SMS) and blocks for up to 4 minutes waiting for the recipient to confirm. This is a pause-resume tool: the chat surfaces a pending UI card while it waits.

send_message

Send an outbound email or text message to one or more contacts only when the user explicitly asks you to send, contact, or notify someone outside this chat. Each entry in to must be either an email address (contains @) or a phone number (starts with +). All recipients in a single call must be the same kind. Email requires subject; SMS does not allow it.

reply_message

Reply to an existing inbound message in its original thread. Use this only when the user explicitly refers to a prior inbound message and provides its messageId. The messageId is the channel-specific identifier the inbound adapter assigned ({channel}:{recipient}[:{thread_id}]).

Triggers

Triggers fire automated work on a schedule or in response to an event. See Triggers for the user-facing model.

search_triggers

Search the trigger catalog by keyword. Returns the matching triggers, each with id, display name, description, setup schema, and edit schema.

setup_trigger

Set up a new trigger of triggerId. params must match the trigger's setupSchema (call search_triggers first to discover the schema). This is a pause-resume tool: the chat surfaces a pending UI card while waiting for confirmation.

manage_active_triggers

List or delete the agent's currently active triggers. operation is one of list or delete. delete requires triggerInstanceId. Enable and disable are not supported by the underlying machinery; delete and re-create instead.

Connections

Connections are the per-org bridges to outside services (Gmail, Slack, etc.). ClankTeam supports two connection types: integrations (a first-party ClankTeam ability) and mcp (an admin-defined per-org MCP server). direct_api and computer_use are out of scope.

search_for_integrations

Search integrations the org has access to. Returns integration ID, name, description, quality score, and builder. Searches both ClankTeam first-party abilities and admin-curated MCP servers.

get_integrations_capabilities

List capabilities for given integration IDs, including tool list, quality ratings (GREAT / GOOD / OK / LIMITED / UNKNOWN), and notes. Use after search_for_integrations to inspect a candidate before creating a connection.

list_users_connections

List all connections the user has created. Returns connectionId, serviceName, description, accountName, connectionType, and connection-specific details. No parameters.

get_details_for_connections

Return the full tool list (activated and deactivated) for the given connections, with descriptions and arguments. Use this to discover what tools a connection has before activating any of them via manage_activated_tools_for_connections.

create_new_connections

Create one or more new connections. Surfaces a UI card for the user to authorize. This is a pause-resume tool.

delete_connection

Permanently delete a connection. Surfaces a confirmation UI showing affected agents and tools that will become unavailable. Irreversible. This is a pause-resume tool.

reauthorize_connection

Re-authorize an expired connection. Surfaces a UI card with a "Sign in again" CTA. Use only on authorization errors or when the user explicitly asks to reauthorize. This is a pause-resume tool.

manage_activated_tools_for_connections

Activate or deactivate tools on existing connections. Activated tools become callable on subsequent turns prefixed by their connectionId. Surfaces a UI card for user approval. This is a pause-resume tool.