All iDialogue Skills and Tools

This page is primarily for agents that need to decide which iDialogue Skills to add for a task.

In iDialogue, a Skill is also known as a tool, function, or primitive tool. In Salesforce, each configured skill is stored as a rooms__AISkill__c record. At runtime, active skills are manifested as OpenAI Responses API tools that an agent may call.

Last reviewed: May 18, 2026.

Selection Rules For Agents

1. Choose the minimum useful tool set. Add only the skills required for the agent objective.
2. Prefer read-only tools before write tools. Use schema, query, search, file-read, or knowledge-search tools before any DML, email, SMS, metadata, or background action.
3. Describe Salesforce objects before querying or writing. Use describe_sobjects before query_sobjects, dml_sobjects, or any field-sensitive Salesforce operation unless the valid object and field set is already explicit in trusted instructions.
4. Do not add externally visible tools unless the objective requires them. Email, SMS, PDF generation, metadata deployment, background execution, and Apollo enrichment can create durable side effects or external costs.
5. Treat credit usage as an estimate. The credit values below come from the configured pricing catalog where available. Model input/output tokens are billed separately from explicit skill usage.
6. Ask for clarification or confirmation before destructive writes. DML delete, metadata deploys, SMS, email, and broad enrichment should be deliberate actions.
7. Avoid declared-but-unmapped primitives for production agents. If a primitive is marked as a runtime gap or internal/testing, do not add it unless the runtime has been updated and verified.

Credit Usage Notes

Credit estimates are listed as configured skill credits:

Apollo tools may also consume Apollo.io credits depending on Apollo account policy and requested data. In particular, person enrichment that reveals personal emails may consume Apollo credits even when the iDialogue tool credit estimate is low.

Quick Chooser

Salesforce Data And Automation Skills

Global Describe

• Primitive API name: global_describe
• Also known as: Global Describe, Salesforce object catalog lookup.
• Description: Retrieves the list of Salesforce objects available to the org, optionally filtered by a search query.
• When to use: Use first when the agent must discover available objects or determine the right object API name. Skip it when the required object names are already specified in instructions or record context.
• Inputs: Optional query string to filter by object name or label.
• Output: XML-formatted object metadata with object API names, labels, and custom-object indicators.
• Estimated credit usage: 2 credits per event.
• Selection cautions: This can return a large object catalog. Prefer a filtered query when the user intent is narrow.

Describe SObjects

• Primitive API name: describe_sobjects
• Also known as: Describe SObjects, Salesforce metadata describe.
• Description: Retrieves field-level metadata for one or more Salesforce object API names.
• When to use: Use before query_sobjects, dml_sobjects, metadata planning, or any field-sensitive Salesforce action.
• Inputs: objectNames, a list of object API names such as Account, Contact, or custom object names.
• Output: Object metadata and available fields for each requested object.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Do not call repeatedly for the same object in one conversation. Reuse prior describe output and fetch only missing objects.

Query SObjects

• Primitive API name: query_sobjects
• Also known as: Query SObjects, SOQL query tool, Salesforce query.
• Description: Runs valid Salesforce SOQL and returns matching records.
• When to use: Use when the agent needs Salesforce record data, aggregates, relationship query results, or record identifiers before a later action.
• Inputs: query, a valid SOQL query string.
• Output: JSON array of Salesforce records or aggregate results.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Use only fields confirmed by describe_sobjects or trusted instructions. Apply LIMIT where possible.

DML SObjects

• Primitive API name: dml_sobjects
• Also known as: Insert/Update Records, Salesforce DML, write records.
• Description: Performs Salesforce INSERT, UPDATE, UPSERT, or DELETE operations for records.
• When to use: Use when the agent is authorized to create, update, upsert, or delete Salesforce records.
• Inputs: JSON object with a records array. Each record includes RecordType, Operation, optional Id, optional ExternalId, optional Name, and a Fields array of fieldName and string value pairs.
• Output: Record IDs for successful changes plus errors for failed records.
• Estimated credit usage: 1 credit per event.
• Selection cautions: This is a write tool. Use describe_sobjects first, do not mix object types or operations in one call, and require confirmation when the user is not explicitly asking to write.

Add Note

• Primitive API name: add_note
• Also known as: Add Note, Salesforce note creation.
• Description: Creates a Salesforce note linked to the current or specified record.
• When to use: Use when the user asks to record a summary, decision, observation, or follow-up note in Salesforce.
• Inputs: Note body, title or summary where supported, and target record context.
• Output: Created note or content record details.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Notes become durable CRM content. Do not store sensitive or speculative content unless the user asks.

Create Task

• Primitive API name: create_task
• Also known as: Create Task, Salesforce follow-up task.
• Description: Creates Salesforce tasks for follow-up work.
• When to use: Use when a user asks for a reminder, assignment, next step, or CRM task.
• Inputs: Subject, description, due date, owner or assignee when available, and related record context.
• Output: Created Salesforce Task ID and status details.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Confirm ambiguous due dates, owners, or task scope before creation.

Invoke Flow

• Primitive API name: invoke_flow
• Also known as: Invoke Flow, autolaunched Salesforce Flow action.
• Description: Invokes the Salesforce Flow configured on the current skill.
• When to use: Use when the agent must call governed Salesforce automation already defined by an administrator.
• Inputs: Skill-defined input variables, often including current record ID or user-provided values.
• Output: Flow output variables, commonly output or a structured response.
• Estimated credit usage: 1 credit per event.
• Selection cautions: The useful schema depends on the configured rooms__AISkill__c record and Flow variables. Do not invent unavailable input variables.

Invoke Apex

• Primitive API name: invoke_apex
• Also known as: Invoke Apex, invocable Apex action.
• Description: Calls a configured Apex REST or invocable action through the Salesforce connection.
• When to use: Use when an agent needs custom Salesforce-side business logic that should not be reimplemented in prompts.
• Inputs: Apex action name or endpoint as configured by the skill, plus JSON-compatible input arguments.
• Output: Apex action result, errors, or status details.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Use only for approved actions. Treat the Apex implementation as authoritative for validation and side effects.

Deploy Field

• Primitive API name: deploy_field
• Also known as: Deploy Field, Salesforce custom field deployment.
• Description: Deploys a custom field to Salesforce metadata.
• When to use: Use when a Salesforce admin or developer intentionally asks to create or modify a field.
• Inputs: Target object, field API name, label, type, and type-specific metadata such as length, precision, scale, or picklist values.
• Output: Metadata deployment status and any errors.
• Estimated credit usage: 2 credits per event.
• Selection cautions: Metadata changes can affect packaging, validation, automation, and user interfaces. Require explicit approval and precise field details.

Deploy Metadata

• Primitive API name: deploy_metadata
• Also known as: Salesforce metadata deploy, metadata deployment.
• Description: Deploys broader Salesforce metadata snippets or components.
• When to use: Use for advanced admin/developer workflows that need metadata changes beyond a single field.
• Inputs: Metadata type, component name, and component body or deployment package details as supported by the skill.
• Output: Deployment status, component identifiers, and errors.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Not in the provided primitive picklist, but mapped in the Java runtime. Use only when the target agent is explicitly authorized for metadata deployment.

Platform Event

• Primitive API name: platform_event
• Also known as: Raise Platform Event, Salesforce event publish.
• Description: Publishes a Salesforce platform event for downstream automation.
• When to use: Use when an agent needs to notify Salesforce automation, integrations, or event-driven processes.
• Inputs: Event API name and event field values.
• Output: Publish status and event result details.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Events can trigger automations. Confirm event type and payload before publishing.

Salesforce Send Email

• Primitive API name: salesforce_send_email
• Also known as: Send Email via Salesforce, Salesforce email.
• Description: Sends outbound email using the Salesforce connection.
• When to use: Use when the agent must send an email through Salesforce identity, templates, governance, or record context.
• Inputs: Recipient addresses, subject, body, optional related record context, and any template or sender options supported by the skill.
• Output: Send status, message identifiers, or errors.
• Estimated credit usage: 2 credits per event.
• Selection cautions: This is externally visible. Confirm recipients and content unless the workflow is explicitly autonomous.

User Interaction And Orchestration Skills

Ask User Choice

• Primitive API name: ask_user_choice
• Also known as: Choice control, user choice prompt.
• Description: Renders a single-choice or multi-select control for the user.
• When to use: Use when the agent needs a constrained decision from the user before proceeding.
• Inputs: Prompt text, options, selection mode, optional default, and labels.
• Output: Structured user choice submitted back into the conversation.
• Estimated credit usage: Not listed.
• Selection cautions: Use for decisions, approvals, or selection among meaningful alternatives. Do not use for open-ended text.

Ask User Form

• Primitive API name: ask_user_form
• Also known as: Form control, structured user input, line-item form.
• Description: Renders a stacked form or line-item grid for review or data entry.
• When to use: Use when the agent needs structured values such as fields, quantities, addresses, or repeatable rows.
• Inputs: Form title, field definitions, current values, validation hints, and line-item definitions when needed.
• Output: Structured form submission JSON.
• Estimated credit usage: Not listed.
• Selection cautions: Use before DML when values must be reviewed or corrected by a user.

Run Background Task

• Primitive API name: run_background
• Also known as: Run Background, asynchronous agent task.
• Description: Starts a background task for long-running agent work.
• When to use: Use for tasks that should continue asynchronously, such as multi-step research, longer processing, or jobs that may outlive the current chat turn.
• Inputs: Task prompt, target agent or model context when applicable, and execution settings supported by the skill.
• Output: Background job or dialogue status.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Background work can continue without active user review. Use only with clear task boundaries and expected output.

Knowledge, Search, And Web Skills

Search Knowledge

• Primitive API name: search_knowledge
• Also known as: Knowledge search, vector knowledge retrieval.
• Description: Searches an organization-specific knowledge base or vector store.
• When to use: Use for product docs, help center content, internal procedures, support articles, or other indexed knowledge.
• Inputs: query, a focused self-contained search query.
• Output: Structured search results with snippets, titles, scores, and source metadata where available.
• Estimated credit usage: 5 credits per event.
• Selection cautions: Call at most once per user question unless a clearly new query is needed. Answer from results and cite relevant sources when available.

Semantic Search

• Primitive API name: semantic_search
• Also known as: Vector search, retrieval search.
• Description: Performs semantic retrieval against configured files, records, or vector spaces.
• When to use: Use when the agent needs lower-level retrieval behavior configured by skill properties rather than public support knowledge search.
• Inputs: Search query and skill-configured knowledge/vector parameters.
• Output: Matching chunks, file or record references, and relevance information.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Active in the Java runtime but not in the provided Salesforce primitive picklist. Prefer search_knowledge for public support/documentation answers.

Search Web

• Primitive API name: search_web
• Also known as: Perplexity web search, public web search.
• Description: Searches the public web through the configured search provider.
• When to use: Use for current public information, external validation, company research, market context, or facts outside Salesforce and iDialogue knowledge.
• Inputs: query, a focused web search query.
• Output: Search answer or results with source material depending on provider behavior.
• Estimated credit usage: 2 credits per event.
• Selection cautions: Web results may be incomplete or stale. Use crawl_web for known URLs that must be read directly.

Crawl Web

• Primitive API name: crawl_web
• Also known as: Web crawler, page fetch.
• Description: Crawls a target URL and returns cleaned page content for reasoning.
• When to use: Use when the URL is known and the agent needs the actual page text, not a search summary.
• Inputs: URL and crawl depth or related crawl options.
• Output: Extracted page content, links, or crawl summary.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Keep depth low unless the user asks for broader crawling. Do not crawl private or sensitive pages unless access is authorized.

Brave Search

• Primitive API name: brave_search
• Also known as: Brave search proxy.
• Description: Runs Brave-powered public web search.
• When to use: Use as an alternative public search provider when Brave results are desired or configured.
• Inputs: Search query and provider-supported result options.
• Output: Search results with links and snippets.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Prefer search_knowledge for iDialogue/product docs and crawl_web for known URLs.

File, Document, Artifact, And Analysis Skills

Process File

• Primitive API name: process_file
• Also known as: File processing, file transcription, file-to-markdown.
• Description: Creates or refreshes a canonical markdown artifact for a Salesforce File owned by the current org.
• When to use: Use when a file must be extracted, converted, inspected, analyzed, or normalized before later reading.
• Inputs: contentVersionId required; optional contentDocumentId, sourceUrl, intent, outputName, and force.
• Output: JSON with processing status, artifact key, and related file metadata.
• Estimated credit usage: 10 credits per event.
• Selection cautions: Call at most once per source file unless the user asks to reprocess. Do not invent file IDs or S3 URLs.

Read File

• Primitive API name: read_file
• Also known as: Read processed file, read file artifact.
• Description: Reads a canonical markdown artifact for a Salesforce File.
• When to use: Use when the agent must inspect, quote, summarize, or reason over processed file content.
• Inputs: contentVersionId required; optional artifact name, offset, and maxChars.
• Output: Artifact body, truncation status, and artifact key.
• Estimated credit usage: 5 credits per event.
• Selection cautions: Prefer default artifact name file.md unless the user asks for a different artifact.

Write File

• Primitive API name: write_file
• Also known as: Write file artifact, save markdown artifact.
• Description: Saves markdown content as an org-scoped File I/O artifact.
• When to use: Use when the agent needs to persist derived markdown, notes, summaries, or transformed content for a file.
• Inputs: contentVersionId, markdown body, optional name, and optional overwrite.
• Output: Artifact key, bytes written, checksum, and write status.
• Estimated credit usage: 5 credits per event.
• Selection cautions: Write markdown only. Use overwrite=true only when replacing existing content is intentional.

Generate PDF

• Primitive API name: generate_pdf
• Also known as: PDF generation, document rendering.
• Description: Converts raw HTML or Markdown into a business-ready PDF.
• When to use: Use for reports, summaries, proposals, briefs, handouts, or customer-ready documents.
• Inputs: Raw HTML or Markdown as the custom tool input. A PDF filename may be set with HTML metadata, title, or heading.
• Output: PDF artifact details, download URL, and page count when available.
• Estimated credit usage: 10 credits per page.
• Selection cautions: Use HTML when layout matters. For chart PDFs, embed contentUrl values returned by render_chart or render_image.

Generate Excel

• Primitive API name: generate_excel
• Also known as: Spreadsheet generation, Excel generation.
• Description: Creates an Excel workbook artifact from structured workbook instructions.
• When to use: Use for tabular deliverables, reports, operational exports, or spreadsheet artifacts.
• Inputs: Workbook name, sheets, columns, rows, formulas, formatting, and output options as supported by the tool schema.
• Output: Excel artifact metadata and download URL.
• Estimated credit usage: 15 credits per page.
• Selection cautions: Use structured data. For simple tables inside a chat answer, a tool call may be unnecessary.

Render Chart

• Primitive API name: render_chart
• Also known as: Chart rendering, chart image.
• Description: Renders a chart as a signed PNG artifact for reports, PDFs, and Salesforce records.
• When to use: Use when the user needs a visual chart from explicit data.
• Inputs: Chart type, title, labels, numeric datasets, axis labels, and visual options.
• Output: Chart artifact JSON with contentUrl, signedUrl, file metadata, and dimensions.
• Estimated credit usage: Not listed.
• Selection cautions: Use exact provided data. Do not invent values for charts unless the user asks for mock data.

Render Image

• Primitive API name: render_image
• Also known as: Image generation, generated PNG.
• Description: Generates a PNG image artifact from a prompt.
• When to use: Use when the user asks for an image, illustration, icon, cover image, visual concept, or generated visual for a PDF.
• Inputs: Prompt, size, quality, background, format, and file name.
• Output: Image artifact JSON with image IDs, file metadata, contentUrl, signedUrl, dimensions, and prompt details.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Use render_image, not generate_image, for the active iDialogue image-generation skill.

Interpret Code

• Primitive API name: interpret_code
• Also known as: Code Interpreter, Python analysis, run_code legacy alias.
• Description: Enables OpenAI native code interpreter for analysis, transformations, calculations, and artifact generation.
• When to use: Use for data analysis, calculations, code execution, file transformations, plots, and computational workflows that exceed normal reasoning.
• Inputs: The agent supplies instructions and files through the Responses API code interpreter container rather than a normal JSON function schema.
• Output: Code interpreter call outputs, logs, generated artifacts, files, images, or analysis results.
• Estimated credit usage: 50 credits per event.
• Selection cautions: This is comparatively expensive. Use when computation or artifact generation is actually needed.

Memory Skills

Save Memory

• Primitive API name: save_memory
• Also known as: Save Memory, persistent rule, agent memory.
• Description: Creates persistent memories with system, record, or user scope.
• When to use: Use when the user explicitly asks the agent to remember a durable rule, preference, instruction, or business logic.
• Inputs: scope, summary, body, recordId, and userId.
• Output: Saved memory record details and status.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Do not save casual facts, speculation, or sensitive information unless the user clearly asks for persistence and scope is appropriate.

Delete Memory

• Primitive API name: delete_memory
• Also known as: Delete Memory, remove memory.
• Description: Deletes an existing agent memory.
• When to use: Use when a user asks to remove, forget, replace, or invalidate a saved memory.
• Inputs: Memory id.
• Output: Deletion status and any error details.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Confirm the target memory when multiple memories may match.

Set Memory

• Primitive API name: set_memory
• Also known as: Legacy memory save alias.
• Description: Legacy alias that maps to save_memory.
• When to use: Prefer save_memory for new agents. Use only for backward compatibility.
• Inputs: Same practical inputs as save_memory.
• Output: Saved memory status.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Do not add to new agents unless a legacy configuration requires the name.

Communication Skills

Send SMS Text

• Primitive API name: send_sms
• Also known as: Send SMS, text message.
• Description: Sends an SMS text message through the configured messaging provider.
• When to use: Use for reminders, urgent notifications, outreach, or workflow messages when SMS is explicitly desired.
• Inputs: Recipient phone number, message body, and optional sender or context fields supported by the skill.
• Output: Send status, message ID, and errors.
• Estimated credit usage: 5 credits per event.
• Selection cautions: SMS is externally visible and may have telecom compliance requirements. Confirm recipient and text before sending unless the workflow is explicitly autonomous.

Address And Geo Skills

Verify Address

• Primitive API name: verify_address
• Also known as: Address verification, address validation.
• Description: Validates and standardizes an address.
• When to use: Use before updating CRM address fields, generating shipping/visit instructions, or routing when address quality matters.
• Inputs: Street, city, state, postal code, country, or a single formatted address depending on schema.
• Output: Verification status, standardized address fields, and validation messages.
• Estimated credit usage: 5 credits per event.
• Selection cautions: Do not overwrite CRM addresses with uncertain results without confirmation.

Geocode Address

• Primitive API name: geocode_address
• Also known as: Geocode, address-to-coordinates.
• Description: Converts an address into latitude and longitude.
• When to use: Use for mapping, distance estimates, territory logic, routing, or location enrichment.
• Inputs: Address string or structured address fields.
• Output: Coordinates and geocoding metadata.
• Estimated credit usage: 3 credits per event.
• Selection cautions: For poor or ambiguous addresses, use verify_address first.

Geo Route

• Primitive API name: geo_route
• Also known as: Route calculation, directions.
• Description: Calculates route distance, duration, and route details between points.
• When to use: Use for travel planning, service routing, territory visits, and distance/duration estimates.
• Inputs: Origin and destination addresses or coordinates, plus routing options where supported.
• Output: Route summary, distance, duration, and route metadata.
• Estimated credit usage: 3 credits per event.
• Selection cautions: Confirm whether the user needs driving route distance rather than straight-line distance.

Apollo Skills

Apollo Enrich Person

• Primitive API name: apollo_enrich_person
• Also known as: Apollo person enrichment, Lead/Contact enrichment.
• Description: Enriches one person candidate with Apollo.io data.
• When to use: Use for one Lead or Contact when the agent has a strong identifier or enough name/company context.
• Inputs: Strong person identifier such as email, hashed email, Apollo person ID, LinkedIn URL, or name plus company/domain context; optional Salesforce record ID/object type.
• Output: Apollo person profile data and enrichment metadata.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Treat output as candidate data. Ask before revealing personal emails or writing Salesforce records.

Apollo Enrich People

• Primitive API name: apollo_enrich_people
• Also known as: Apollo bulk person enrichment, multi-Lead/Contact enrichment.
• Description: Bulk enriches up to the configured limit of people.
• When to use: Use when the user asks to enrich several known Leads, Contacts, or person candidates.
• Inputs: List of person identifiers or person candidate objects.
• Output: List of enriched person profiles and per-record status.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Use search first when the people are not selected. Confirm before bulk writes or email reveal.

Apollo Enrich Organization

• Primitive API name: apollo_enrich_organization
• Also known as: Apollo account enrichment, company enrichment.
• Description: Enriches one organization or Account candidate with Apollo.io organization data.
• When to use: Use for one Account or company when domain, Apollo organization ID, or company context is available.
• Inputs: Domain, organization name, Apollo organization ID, Salesforce Account ID, or related identifiers.
• Output: Organization profile, firmographic details, and enrichment status.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Do not overwrite CRM Account values with weak or conflicting data without review.

Apollo Enrich Organizations

• Primitive API name: apollo_enrich_organizations
• Also known as: Apollo bulk organization enrichment, multi-Account enrichment.
• Description: Bulk enriches multiple organizations.
• When to use: Use when a user asks to enrich several known Accounts or company candidates.
• Inputs: List of domains, Apollo organization IDs, organization names, or Salesforce Account candidates.
• Output: List of enriched organization profiles and per-organization status.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Keep batches intentional and review target fields before Salesforce DML.

Apollo Get Organization

• Primitive API name: apollo_get_organization
• Also known as: Apollo organization lookup, get company by Apollo ID.
• Description: Gets complete Apollo.io organization details by Apollo organization ID.
• When to use: Use after person enrichment or search returns an Apollo organization ID and the agent needs full company details.
• Inputs: Apollo organization ID.
• Output: Organization details from Apollo.io.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Use this for known Apollo IDs. Use organization search when only partial company information is available.

Apollo Search People

• Primitive API name: apollo_search_people
• Also known as: Apollo people search, prospect search.
• Description: Searches Apollo.io for partial person prospects.
• When to use: Use before enrichment when the user asks to find people by title, seniority, location, company domain, organization ID, or keyword.
• Inputs: Search filters such as titles, seniorities, person locations, organization locations, organization domains, organization IDs, keywords, page, and per-page limit.
• Output: Candidate people search results with partial profiles.
• Estimated credit usage: 0 credits per event.
• Selection cautions: Search results are partial. Use enrichment before writing high-confidence Lead or Contact data.

Apollo Search Organizations

• Primitive API name: apollo_search_organizations
• Also known as: Apollo organization search, company search.
• Description: Searches Apollo.io for company prospects.
• When to use: Use when the user asks to find organizations by domain, keyword, location, industry, employee count, or similar filters.
• Inputs: Organization search filters, page, and per-page limit.
• Output: Candidate organization search results.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Treat results as candidates until enriched or confirmed by the user.

Apollo Usage Stats

• Primitive API name: apollo_usage_stats
• Also known as: Apollo usage, Apollo rate limits.
• Description: Reports Apollo.io API usage stats and limits for the configured credential.
• When to use: Use before or during Apollo-heavy workflows to understand available quota.
• Inputs: Usually no user-supplied business input is required beyond the configured org credential.
• Output: Apollo usage and rate-limit information.
• Estimated credit usage: 0.5 credits per event.
• Selection cautions: Useful for diagnostics and quota planning, not for CRM enrichment itself.

Internal, Testing, And Runtime-Specific Entries

Mock Function

• Primitive API name: mock_function
• Also known as: Test mock JSON function.
• Description: Testing utility for exercising function-call behavior.
• When to use: Use only in development, tests, or demonstrations.
• Inputs: Test query or payload.
• Output: Mock function result.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Do not add to production agents.

Mock Freeform

• Primitive API name: mock_freeform
• Also known as: Test mock freeform tool.
• Description: Testing utility for freeform/custom tool behavior.
• When to use: Use only in development, tests, or demonstrations.
• Inputs: Freeform test content.
• Output: Mock freeform result.
• Estimated credit usage: Not listed.
• Selection cautions: Do not add to production agents.

Run Code

• Primitive API name: run_code
• Also known as: Legacy Code Interpreter alias.
• Description: Deprecated alias for interpret_code.
• When to use: Prefer interpret_code. Use run_code only for legacy agent records that already reference it.
• Inputs: Same runtime behavior as code interpreter after canonicalization.
• Output: Code interpreter outputs and artifacts.
• Estimated credit usage: 50 credits per event.
• Selection cautions: New agents should use interpret_code.

Generate Image

• Primitive API name: generate_image
• Also known as: Legacy image generation primitive.
• Description: Declared and priced image-generation primitive.
• When to use: Prefer render_image, which is the active Java tool for generated image artifacts.
• Inputs: Image prompt and generation options if supported by a target runtime.
• Output: Generated image artifact if supported.
• Estimated credit usage: 1 credit per event.
• Selection cautions: Do not add to new agents unless the runtime has explicit support for generate_image.

Generate Document

• Primitive API name: generate_document
• Also known as: Document generation.
• Description: Declared document-generation primitive for template-based documents.
• When to use: Prefer generate_pdf, generate_excel, or file tools unless a target runtime explicitly supports generate_document.
• Inputs: Template and document data when supported.
• Output: Generated document artifact.
• Estimated credit usage: 10 credits per page.
• Selection cautions: Not a normal agent-facing Java function-factory tool in the current runtime.

Non-Selectable Primitive Families

The following primitives or prompt helpers are declared or present in supporting files but should not normally be selected for new production agents unless the runtime has explicit support:

Public-Agent Safety Checklist

Before adding a skill to an agent that may act for customers, prospects, or public website visitors:

1. Confirm the skill is necessary for the agent objective.
2. Confirm the agent has instructions for when to call the skill and when not to call it.
3. Confirm read tools are available before write tools when the agent needs Salesforce state.
4. Confirm side-effect tools have approval or autonomy rules.
5. Confirm the agent has a response pattern for tool failures, empty results, and ambiguous user requests.
6. Confirm credit-heavy tools such as interpret_code, generate_pdf, file processing, SMS, and Apollo enrichment are only used when needed.