Built-in tools

Mogick assembles the available toolset based on the execution mode. Plan mode supports exploration, user questions, and plan creation. Agent mode adds file changes, command execution, MCP, and change tracking. Experimental features and build options may further adjust tool availability.

File and code search

Tool	Action	Plan	Agent
file.read	Read file	✓	✓
file.list_dir	List directory	✓	✓
file.glob	Match files by glob	✓	✓
file.grep	Search file contents	✓	✓
code.search	Use mwcs for semantic code search; depends on whether the build includes mwcs	✓	✓
file.tree_find	Locate by file tree using mwcs; depends on whether the build includes mwcs	✓	✓
file.write	Create or overwrite a file	—	✓
file.edit	Edit files precisely	—	✓
file.apply_patch	Apply multi-file patches; may partially succeed, not transactional	—	✓
file.delete	Delete a file	—	✓

File writes, edits, and deletions are subject to the active permission policy. Long tool results may be truncated in the model context, while the complete results remain in the session record.

Shell and Git

Tool	Action	Plan	Agent
shell.exec	Run a shell command and return its output	—	✓
git.status	View workspace status	✓	✓
git.diff	View differences	✓	✓
git.log	View commit history	✓	✓
git.show	View commit content	✓	✓

The built-in Git tools are read-only evidence tools. For a Git command with write effects, the agent uses shell.exec, subject to the corresponding approval.

mogick run "Run go test ./... and summarize the failures"

Planning, tasks and collaboration

Tool	Action	Plan	Agent
plan.read	Read the current plan	✓	✓
plan.write	Save or incrementally update a plan	✓	—
todo.write	Write task list	✓	✓
todo.update	Update task status	—	✓
user.question	Post a structured single-choice or multiple-choice question to the user	✓	✓
tool.usage	Query tool usage	✓	✓
agent.task	Delegate a task to a subagent; controlled by the experimental feature setting	✓	✓

After plan.write completes, the model ends the turn. The TUI or ACP host presents the plan for approval and handles the mode switch. Plan mode does not use plan.exit.

Skills

Tool	Action	Plan	Agent
skill.find	Find Skills	✓	✓
skill.load	Load a Skill's instructions	✓	✓
skill.create	Create Skill	—	✓

Finding and loading Skills are read-only operations. Creating a Skill writes files and requires Agent mode.

MCP meta-tools

MCP meta-tools are only available in Agent mode.

Tool	Action
mcp.list	List servers
mcp.status	View server status
mcp.search	Search for tools across servers
mcp.help	Get tool help
mcp.read	Read server resources
mcp.wait_ready	Wait for the server to be ready
mcp.connect	Connect to server
mcp.disconnect	Disconnect the server
mcp.reconnect	Reconnect to the server
mcp.call	Invoke the specified MCP tool

mcp.call can wrap remote tools with side effects, so it requires approval by default. If the target tool declares readOnlyHint, the permission policy may treat the call as read-only.

Change tracking and execution summaries

Agent mode also provides tools for the Changes panel, /review, and execution summaries:

Tool	Action	Plan	Agent
changeset.list_groups	List Changeset groups	—	✓
changeset.read_group_patchset	Read the patch details of the group	—	✓
changeset.commit_context	Generate commit context	—	✓
execution.summary	Summarize execution results	—	✓
tracing.group	Tracing group information	—	✓
julia.interpreter	Julia code interpretation and execution; depends on the build configuration	—	✓

Read-only tools can run in parallel. Tools with side effects are ordered and approved according to the runtime policy. See Tool approvals and permission policies.

Images

image.understand and image.ocr can be available in Plan and Agent modes when the session has the required official service and scene model. They are not enabled by logging in alone for an arbitrary custom provider. See Files and images.