comment_on_task
ChatGPTCreate a comment on a task via
action (create). Reversible authoring action.Overview
Manage Writesonic Agents tasks, playbooks, grids, knowledge base, and pages from chat — triage work, follow what agents did, inspect playbook runs, read grids, search your Knowledge Base and Brand Kit, and pull Pages reports. Connects to the hosted Writesonic Agents MCP server (20 tools plus batch execution, OAuth sign-in) and ships playbooks for task triage, agent-run review and workspace status snapshots. Only task creation, updates and comments write data; everything else is read-only and writes are confirmed first. Requires Writesonic Agents access.
Tools
create_task
ChatGPTCreate a task in the project. Purely additive: nothing is overwritten and the task is visible and deletable in the product.
download_attachment
ChatGPTGet a short-lived presigned download URL for a task or comment attachment. Returns the URL plus filename and content type - not the file bytes. For target 'task', pass task_id.
execute_batch
ChatGPTRun up to 10 tool calls in one request, passing results between steps. Use instead of separate calls whenever you already know three or more calls to make: independent reads (identity, labels and settings in one request) or query then act on each result. Each step runs with full auth and RBAC, exactly as if called directly, and can only call tools available on this mount. A failed step stops the batch; completed results are returned alongside the error. Not for loops or conditionals. Each step's
args is the payload you'd send calling that tool directly; the args wrapper some tools take is added automatically. Reference a prior step's result with "$N.path" (e.g. "$0.items[0].id").execute_plan
ChatGPTRun JavaScript that chains tools with loops, conditionals, and transforms. The code runs in a sandbox with exactly two capabilities:
await tool(name, args) calls any tool available on this mount (same auth as calling it directly), and output(value) sets the result returned to you. Use this when a plan needs iteration ("comment on every overdue task") or when a large intermediate payload should be reduced before you read it. For fixed linear sequences prefer execute_batch. Hard timeout 30s. The whole plan is scoped to a single project - the project_id you pass to execute_plan. Every tool() call inherits it; do not pass project_id inside tool() calls (it is rejected). To act on a different project, ask the user which one and run a separate execute_plan - a plan cannot fan out across projects.get_agent_runs
ChatGPTGet agent run history for a task or an agent, or step-by-step progress for one run. Scope task_runs needs task_id, agent_runs needs agent_id, run_progress needs task_id and run_id.
get_brand_kit
ChatGPTReturns the project's brand guidelines as text: business, audience, objectives, authors, voice and visual design. Use it to write or design content in the brand's style.
view picks the format: catalog lists the guideline sections and the options they offer; compose returns all the guidelines as one block of text; compose_by_section returns them split by section; compose_image_prompt returns the visual-design guidelines written as an image description, optionally built around a subject. Optional sections, content_type, author, audience and voice narrow the guidelines returned.get_ci_dashboard
ChatGPTProject-level content-intelligence reads, selected by
view. view=pages is also how a page named in natural language (by URL/title) is resolved to the page_id get_page_report needs. view=filters is the discovery call for the pages list's sort/filter metric keys and named presets (e.g. 'declining pages') — call it before guessing a sort key. view=groups resolves a page-group NAME (filter with search) to the group_id view=pages/summary's group_ids param needs — group_ids is never guessable, always resolve it here first.get_grid_data
ChatGPTRead a grid's columns, rows, one cell, a cell's run detail, or search rows via the
view enum. A workflow column's workflow_version_id (view=columns) names its pinned VERSION - pass to get_workflow(view=version_detail); cell_run's workflow_run_id instead names the RUN.get_knowledge_source
ChatGPTList the project's knowledge sources, list its folders, or fetch one source's content (kih-hub). Does NOT cover the Brand Kit — that is a separate resource, fetched with get_brand_kit, not a knowledge source you list/read here. view=list enumerates sources (filter with folder_id/type/status/q, paginate with cursor/limit); view=folders lists the project's knowledge folders (id + name) — use it to resolve a folder NAME to its folder_id before filtering or ingesting into that folder; view=content returns the parsed text of one source (needs source_id).
get_page_report
ChatGPTPer-page content-intelligence report selected by
report_type (SEO, AI citations, funnels, traffic, SERP, keywords, lifecycle, activity). Some reports require market_id. page_id is resolved from get_ci_dashboard view=pages, search=<url or title> — never fabricate it.get_task
ChatGPTGet one task with full detail (attachments and labels are always included). Pass
include to also fetch the task's timeline, relations, and/or approvals. timeline_cursor and timeline_kind apply only when the timeline is included.get_workflow
ChatGPTGet one workflow and its related data. view selects the slice: 'detail', 'versions', 'version_detail' (needs version_id), 'runs' (filter by run_status/run_search), or 'run_detail' (needs workflow_run_id).
list_grids
ChatGPTList a project's grids, or read one grid's detail/processing status, via the
view enum (detail and status need grid_id). view=column_options is the live catalog of column types you can actually add - not just the fixed input data-type enum, but also every workflow this account can use as a workflow column; call it live to get the available types rather than guessing them.list_integrations
ChatGPTList the workspace's connected apps and integration connections. view=apps returns the app catalogue; apps_overview groups connections by app with active/inactive counts; connections lists individual connections (filter with q/app/status/access, and project_id to scope either list to one project); connection_detail returns one connection (needs connection_id). Credentials are never returned.
list_project_resources
ChatGPTList a project's reference resources by
resource: users (pass the id whoami returned for you as resource_id to get your own project-scoped record, e.g. your role in this project), labels, settings, skills, agents, or agent_settings. Pass resource_id to fetch one item's detail (users, skills, agents only).list_workflow_components
ChatGPTList workflow building blocks. view='hub_capabilities' is the compact catalog (id, name, description per block; page with page/size) - use it to browse or answer what blocks or capabilities exist, filtered by skill_id or capability_id when named. view='components' returns the same blocks with their full input/output schemas and is very large (several thousand characters per block) - use it with search or a small size only when you need the schemas. view='batch' returns full schemas for known capability_ids.
list_workflows
ChatGPTList the project's workflows or their status counts. Use view='list' for the paginated inventory (filter by type/status/search) or view='counts' for totals.
query_tasks
ChatGPTQuery the project's tasks. view='list' returns a cursor page of task list items (filter/sort/pagination); view='counts' returns totals grouped by group_by (status/priority/assignee_id/creator_id).
search_knowledge
ChatGPTSemantic search over the project's knowledge sources (kih-hub RAG). Returns the matched chunks and, when generate_answer is True, a generated answer; set it False to get raw chunks only. Optionally scope the search to specific source_ids/folder_ids/tag_ids or workflow_status; omit all to search the whole project. A source is only indexed here once it finishes syncing, so an unscoped query already excludes sources still queued, syncing, or failed - no source_ids scoping needed to search "synced only".
update_task
ChatGPTUpdate a task: edit fields (action=update_fields), transition its status (action=transition_status), or (re)assign it (action=assign). New values replace the old ones.
whoami
ChatGPTReturn the caller's identity, HQ teams, and the projects in the active workspace. Call this first to bootstrap: you're scoped to one workspace automatically, so its projects come back here — pass a project's id as the
project_id argument on project-scoped tools. identity.role is account-wide, not per-project (see list_project_resources resource=users for your own project-scoped record).Capabilities
Writes
Example Prompts
Click any prompt to copy it.
