System tools for Optimizely Mark

18 min

This page lists the built-in Mark (Optimizely Agent Platform) tools provided and maintained by Optimizely. You and Mark can call these tools if they are enabled. Review the tools so you can see what actions are available for you and your workflows.

Click a tool's name to expand it and learn when to use it, its required and optional parameters, and example prompts for calling the tool. If you do not provide a required parameter, Mark prompts you for it.

Agents

When to use

Build an agent for a task you repeat often.

Change the instructions, inputs, tools, or output format of an existing agent.

Rename an agent or shorten its handle.

Add quality criteria and a passing score to an agent.

Turn a single-run agent into a multi-turn conversation agent.

Parameters

(Optional) agent_id – The handle of the agent to edit. Omit this value to create an agent.

(Optional) name – The display name of the agent, such as *Support Ticket Triager*. Required when you create an agent.

(Optional) description – A short summary of what the agent does and when to use it.

(Optional) prompt_template – The instructions the agent follows on every run. Write the instructions in second person and reference inputs with the [[parameter_name]] format. Required when you create an agent that does not use conversation mode.

(Optional) parameters – The inputs the agent accepts. Omit this value for an agent that needs no input.

(Optional) output – The format the agent returns, such as a report or a structured response.

(Optional) enabled_tools – The tools the agent calls during a run. Each tool must exist in your Mark instance.

(Optional) inference_type – The model tier. Use simple for most tasks, complex or complex_with_thinking for multi-step reasoning, and code for code-heavy work. Defaults to simple.

(Optional) conversation_mode – Set to true for a multi-turn chat agent and false for a single-run task agent. Defaults to false.

(Optional) system_prompt_template – The system instructions for a conversation agent. Mark applies this value only when conversation_mode is true.

(Optional) initial_message – The greeting a conversation agent displays before you type.

(Optional) enable_auto_tool_prediction – Lets the agent select relevant tools during a run, in addition to the tools in enabled_tools. Defaults to false.

(Optional) enable_personal_context – Adds your saved profile and preferences to the agent instructions. Defaults to false.

(Optional) enable_auto_skill_prediction – Applies organization skills that match the agent instructions. Defaults to false.

(Optional) enable_auto_skill_prediction_by_tools – Applies organization skills that match the tools of the agent. Defaults to false.

(Optional) baseline_evaluation_score – The score from 0 through 100 that a run must reach to pass. Mark displays this value with the evaluation criteria on the Quality tab.

(Optional) evaluation_criteria – The quality standards Mark scores each run against, such as Response answers the question directly. Mark displays these standards on the Quality tab. Pass an empty list to clear every standard.

(Optional) input_guardrails – The trained guardrail version that screens input before a run.

(Optional) is_enabled_in_chat – Makes the agent available in Mark chat. Defaults to true.

(Optional) cwd – The working directory the agent uses for files and personal context. The path must start with /workspace or /context. A /context path requires enable_personal_context set to true.

(Optional) file_urls – The reference files the agent uses as knowledge, such as templates or brand guidelines. Upload each file first. Pass an empty list to remove every attached file.

(Optional) new_agent_id – The replacement handle for an existing agent. Use letters, digits, hyphens, and underscores. To change only the display name, use name instead.

(Optional) visibility – The audience for an agent you create. Use private to restrict the agent to yourself, or shared to release it to everyone in your instance.

Example prompts

Create an agent that sorts support tickets by urgency and drafts a first reply.

Edit my release notes reviewer so it also flags passive voice.

Build a conversation agent that coaches product managers on user stories.

Add quality criteria to my changelog agent with a passing score of 80.

Rename the handle for my LinkedIn post writer to li-post-writer.

Skills

When to use

Save a standing instruction from a conversation so Mark applies it later.

Change the content, title, or trigger settings of an existing skill.

Add a shortcut so you start a skill by name in chat.

Attach reference files, such as templates or brand guidelines, to a skill.

Share a personal skill with everyone in your organization.

Parameters

skill_type – The audience for the skill. Use personal for your own use, or org to share the skill across your organization. A change of audience requires a skill.

(Optional) skill_id – The ID of the skill to edit. Omit this value to create a skill.

(Optional) title – The display name of the skill, such as *Translate to Dothraki*. Required when you create a skill.

(Optional) skill – The instructions Mark follows when the skill applies. Write complete instructions, including which tools to use. Required when you create a skill.

(Optional) mention_handle – The shortcut that starts the skill in chat, entered without the slash. Use lowercase letters, numbers, and hyphens.

(Optional) trigger_type – How Mark activates the skill. Use keyword to activate on matching phrases, or tool to activate when specific tools are present.

(Optional) trigger_conditions – A plain-language description of when Mark applies the skill.

(Optional) trigger_tools – The tools that activate the skill when they are present in a session.

(Optional) active – Makes the skill available in chat. Skills you create are active by default.

(Optional) files – The files attached to the skill, such as a template or a script. Upload each file to the conversation first. Pass an empty list to remove every attached file.

(Optional) instance_id – The instance that the skill applies to. Include this value only when you limit the skill to one instance.

(Optional) product_sku – The product that the skill applies to. Include this value only when you limit the skill to one product.

(Optional) update_summary – A short summary of what changed, such as *Add shortcut*.

Example prompts

Save this as a skill.

Create a skill that formats my release notes as a table.

Add a waterways shortcut to my AmaWaterways skill.

Change the title of my newsletter skill to "Newsletter copywriter".

Share this skill with my whole organization.

Canvas and artifacts

Your active canvas provides a dynamic and collaborative environment where you and Mark can build, customize, and deliver interactive digital assets, applications, and complex content. This workspace lets you create and manage digital assets in real-time.

Search and context

Note

If you use Opti ID, administrators can turn off generative AI in the Opti ID Admin Center. See Turn generative AI off across Optimizely applications.

System tools for Optimizely Mark

Agents

ask_about_agent_execution – Answers a plain-language question about a single agent run and detects whether the run belongs to a workflow agent or a specialized agent.

ask_agent_input – Displays a form in chat that gathers the input a specialized agent needs before the agent runs.

create_or_edit_agent – Drafts a specialized agent or edits an existing one, then displays an approval card that saves the agent after you confirm it.

execute_specialized_agent – Runs a specialized agent with the input values you supply.

execute_workflow_agent – Runs a workflow agent from a plain-language description of the work.

get_agent – Returns the full definition of one specialized agent.

get_agent_execution – Returns the details of one run and detects whether the run belongs to a workflow agent or a specialized agent.

get_workflow_agent_step_execution – Returns the full detail of one workflow step, including its input values, status, and any error.

list_agent_executions – Lists workflow runs and specialized agent runs together, starting with the most recent.

list_workflow_agent_step_executions – Lists every run of a workflow step, starting with the oldest.

lookup_approved_plans – Returns approved workflow plans that match the work a goal-driven workflow received, so the workflow reuses a proven approach.

read_agent_execution_memory – Returns the stored messages from a run, including the tool calls and reasoning behind the result.

search_agents – Finds specialized agents and workflow agents in your Mark instance by handle, name, or description.

Skills

create_or_edit_skill – Saves a reusable instruction as a personal skill or an organization skill, and edits or renames an existing skill.

find_relevant_skills – Returns the skills that best match a described intent, ranked by how well each skill fits.

find_skill – Finds a skill by matching text against the title, content, and shortcut of each skill.

get_skill – Returns the full definition of one skill.

Canvas and artifacts

canvas_to_file – Exports a canvas to a file, including PDF for markdown canvases and HTML canvases.

create_canvas – Creates a canvas, an editable document that you and Mark work on together in real time.

create_csv_canvas – Creates a canvas that holds comma-separated values (CSV) data, either from files you supply or from data Mark already has.

create_powerpoint_canvas – Creates a canvas from a PowerPoint file so you view and change the deck inside Mark.

edit_canvas – Applies changes to a canvas, either as a targeted edit or as a full replacement of the content.

edit_csv_canvas – Changes the data in a CSV canvas by adding, updating, deleting, or reordering rows and columns.

get_canvas – Returns the content and details of one canvas.

get_canvas_by_workspace – Lists every canvas in a space, with the name, format, and creation date of each one.

get_canvas_workspace – Returns the details of the canvas space attached to the current conversation.

present_widget – Displays a chart, a table, or a set of metric cards directly in the chat.

update_powerpoint_canvas – Changes a PowerPoint canvas by deleting slides, replacing text, copying a slide from another deck, or adding an image.

Search and context

search_input_guardrails – Lists the input guardrails in your Mark instance and the trained versions of each one. An input guardrail is a rubric that screens a prompt before an agent acts on it. Mark returns only the guardrails that have at least one trained version.

search_optimizely_docs – Searches official Optimizely product documentation and returns an answer with links to the source articles.

search_past_context – Searches earlier parts of a long conversation for a decision, a fact, a file, or a topic. Mark keeps recent messages close at hand, so use this tool to reach detail from further back in the thread.

System tools for Optimizely Mark

Agents

ask_about_agent_execution – Answers a plain-language question about a single agent run and detects whether the run belongs to a workflow agent or a specialized agent.

ask_agent_input – Displays a form in chat that gathers the input a specialized agent needs before the agent runs.

create_or_edit_agent – Drafts a specialized agent or edits an existing one, then displays an approval card that saves the agent after you confirm it.

execute_specialized_agent – Runs a specialized agent with the input values you supply.

execute_workflow_agent – Runs a workflow agent from a plain-language description of the work.

get_agent – Returns the full definition of one specialized agent.

get_agent_execution – Returns the details of one run and detects whether the run belongs to a workflow agent or a specialized agent.

get_workflow_agent_step_execution – Returns the full detail of one workflow step, including its input values, status, and any error.

list_agent_executions – Lists workflow runs and specialized agent runs together, starting with the most recent.

list_workflow_agent_step_executions – Lists every run of a workflow step, starting with the oldest.

lookup_approved_plans – Returns approved workflow plans that match the work a goal-driven workflow received, so the workflow reuses a proven approach.

read_agent_execution_memory – Returns the stored messages from a run, including the tool calls and reasoning behind the result.

search_agents – Finds specialized agents and workflow agents in your Mark﻿ instance by handle, name, or description.

Skills

create_or_edit_skill – Saves a reusable instruction as a personal skill or an organization skill, and edits or renames an existing skill.

find_relevant_skills – Returns the skills that best match a described intent, ranked by how well each skill fits.

find_skill – Finds a skill by matching text against the title, content, and shortcut of each skill.

get_skill – Returns the full definition of one skill.

Canvas and artifacts

canvas_to_file – Exports a canvas to a file, including PDF for markdown canvases and HTML canvases.

create_canvas – Creates a canvas, an editable document that you and Mark﻿ work on together in real time.

create_csv_canvas – Creates a canvas that holds comma-separated values (CSV) data, either from files you supply or from data Mark﻿ already has.

create_powerpoint_canvas – Creates a canvas from a PowerPoint file so you view and change the deck inside Mark﻿.

edit_canvas – Applies changes to a canvas, either as a targeted edit or as a full replacement of the content.

edit_csv_canvas – Changes the data in a CSV canvas by adding, updating, deleting, or reordering rows and columns.

get_canvas – Returns the content and details of one canvas.

get_canvas_by_workspace – Lists every canvas in a space, with the name, format, and creation date of each one.

get_canvas_workspace – Returns the details of the canvas space attached to the current conversation.

present_widget – Displays a chart, a table, or a set of metric cards directly in the chat.

update_powerpoint_canvas – Changes a PowerPoint canvas by deleting slides, replacing text, copying a slide from another deck, or adding an image.

Search and context

search_input_guardrails – Lists the input guardrails in your Mark﻿ instance and the trained versions of each one. An input guardrail is a rubric that screens a prompt before an agent acts on it. Mark﻿ returns only the guardrails that have at least one trained version.

search_optimizely_docs – Searches official Optimizely product documentation and returns an answer with links to the source articles.

search_past_context – Searches earlier parts of a long conversation for a decision, a fact, a file, or a topic. Mark﻿ keeps recent messages close at hand, so use this tool to reach detail from further back in the thread.

search_agents – Finds specialized agents and workflow agents in your Mark instance by handle, name, or description.

create_canvas – Creates a canvas, an editable document that you and Mark work on together in real time.

create_csv_canvas – Creates a canvas that holds comma-separated values (CSV) data, either from files you supply or from data Mark already has.

create_powerpoint_canvas – Creates a canvas from a PowerPoint file so you view and change the deck inside Mark.

search_input_guardrails – Lists the input guardrails in your Mark instance and the trained versions of each one. An input guardrail is a rubric that screens a prompt before an agent acts on it. Mark returns only the guardrails that have at least one trained version.

search_past_context – Searches earlier parts of a long conversation for a decision, a fact, a file, or a topic. Mark keeps recent messages close at hand, so use this tool to reach detail from further back in the thread.