Capability rules
- Read capabilities can usually run without approval.
- Mutating configuration or durable data requires approval.
- Long-running actions have matching wait capabilities.
- Workbench should inspect or search documents before acting on a vague source scope.
- Saving files into Vault is different from using files as temporary Workbench context.
Document and Vault capabilities
| Capability | What it does | Notes |
|---|---|---|
inspect_vault_documents | Reads document metadata, IDs, Vault scope, and processing state. | Use early when a task depends on Vault documents. |
search_accessible_vault_documents | Searches documents and extracted tags across accessible Vaults. | Resolves source scopes before asking Vault, creating reports, or running workflow-backed actions. |
ask_vault | Answers using Vault documents, extracted tags, and optional uploaded reference text. | Read-only analysis, comparison, summarization, and evidence-backed answers. |
answer_reference_document | Answers directly from uploaded reference text without requiring Vault processing. | Good for quick uploaded-file analysis. |
import_uploaded_documents | Imports Workbench-uploaded files into a selected Vault. | Requires approval and uses the durable Vault document pipeline. |
reprocess_documents | Queues selected Vault documents for reprocessing. | Requires approval; use after configuration or extraction changes. |
wait_for_document_processing | Waits for document or tag processing to finish. | Used after custom tag runs, imports, or reprocessing. |
Legislate and comparison capabilities
| Capability | What it does | Notes |
|---|---|---|
read_legislate_documents | Reads active Legislate/team documents. | Legislate is the validated team-document archive, not a Vault. |
compare_legislate_vault_documents | Compares Legislate documents with Vault documents. | Use when the user asks about published documents versus Vault source documents. |
MCP provider capabilities
| Capability | What it does | Notes |
|---|---|---|
list_mcp_providers | Lists enabled MCP providers for the Workbench session. | Shows connection status, Workbench enablement, and capabilities. |
list_mcp_folders | Lists folders from a connected file-storage provider. | Requires a provider with folder listing capability. |
list_mcp_files | Lists or searches files in a provider folder. | Use before selecting files for import. |
import_mcp_documents | Imports selected provider files into the current Workbench chat. | Imports temporary Workbench context; saving to Vault is a separate approved step. |
Playbooks
| Capability | What it does | Notes |
|---|---|---|
read_playbooks | Lists or searches available playbooks. | Read-only inventory. |
create_playbook | Creates a reusable playbook from source text or rules. | Requires approval. |
run_playbook | Runs an existing playbook against selected Vault documents. | Asynchronous; use for reusable review logic. |
Document types, tags, and extraction setup
| Capability | What it does | Notes |
|---|---|---|
read_document_types | Lists document types for a Vault or workspace. | Read-only inventory. |
create_document_type | Creates a document type or adds default tag definitions. | Requires approval. |
delete_document_type | Deletes a custom document type. | Requires approval. |
setup_image_transcription | Creates or reuses an image type and default image tags. | Requires approval; used for embedded image transcription/classification. |
create_custom_tags | Creates or reuses custom tags/questions. | Requires approval because it changes reusable extraction configuration. See Tags and extraction. |
run_custom_tags | Runs tags/questions against selected documents or a Vault. | Requires approval; often followed by wait_for_document_processing. |
run_table_extraction | Queues table or line-item extraction. | Asynchronous; followed by wait_for_table_extraction. |
run_entity_extraction | Queues entity extraction/resolution. | Asynchronous; followed by wait_for_entity_extraction. |
read_vault_entity_graph | Reads entities and relations from a Vault. | Read-only graph data. |
Reports and templates
| Capability | What it does | Notes |
|---|---|---|
create_report | Creates a report, table, comparison, summary, or analysis over selected documents. | Asynchronous; should honor resolved source scope and carry document context into the report. |
wait_for_report_completion | Waits for a report job to complete. | Reads status and result. |
fill_template | Fills an uploaded Word, Excel, CSV, text, form, or spreadsheet template. | Uses uploaded template files and conversation context. |
synthesize_workbench_result | Produces a partial answer from successful prior steps if a dependent analyzer failed. | Fallback for resilient task completion. |
Workflows and agents
| Capability | What it does | Notes |
|---|---|---|
read_workflows | Lists available workflows. | Read-only inventory. |
create_workflow | Creates a reusable workflow using the workflow block catalog. | Requires approval; should preserve the selected Vault or document scope and surface warnings before creation. See Workflow blocks. |
run_workflow | Runs an existing workflow or one-off workflow-backed action. | Asynchronous; followed by wait_for_workflow_completion; should confirm scope before side-effecting runs. |
cleanup_one_off_workflow | Soft-deletes a temporary one-off workflow after completion. | Host-inserted cleanup. |
read_agents | Lists on-demand agents. | Read-only inventory. |
create_agent | Creates an on-demand agent. | Requires approval. |
run_agent | Runs an existing on-demand agent against selected documents. | Requires approval; followed by wait_for_agent_completion. |
Records and schemas
| Capability | What it does | Notes |
|---|---|---|
inspect_record_schemas | Reads accessible record schemas. | Use before record mutation. |
lookup_records | Reads or searches business records. | Read-only. |
upsert_record | Creates or updates business records. | Requires approval. |
update_record_schema | Creates or replaces record schemas. | Requires approval. |
Workbench skills and task lifecycle
| Capability | What it does | Notes |
|---|---|---|
read_workbench_skills | Reads saved Workbench skills from task context. | Use when the user references recurring instructions or review style. |
apply_workbench_skill | Applies saved skill guidance to planning and execution. | Guidance-only capability. |
verify_goal_completion | Verifies that completed task steps satisfy the original goal. | Used near the end of a plan. |
save_workbench_artifact | Saves a result, report, table, or structured output back to the session. | Final persistence step. |
ask_for_approval | Pauses for user approval before a proposed action. | Reserved for host-created approval steps. |