artifact_binary_upload_v1
ChatGPTUpload an artifact to the signed-in account.
Overview
Cmdexa connects ChatGPT to a Windows computer you choose through an account-scoped relay. Inspect files and task status, then request commands with local approval on Windows. Each Cmdexa account can access only its own paired devices and data. Windows only. The app content is suitable for users under 18, but this beta is intended for adults using Cmdexa for business or professional purposes; consumers may not use this beta. Installers are not code-signed. Review every approval prompt and connect only computers you are authorized to administer.
Tools
artifact_cleanup_v1
ChatGPTPreview or remove artifacts belonging only to the signed-in account.
artifact_list_v1
ChatGPTList artifacts belonging to the signed-in account.
artifact_task_log_bundle_v1
ChatGPTCreate a task log artifact for a task owned by the signed-in account.
cancel_task_v2
ChatGPTCancel a task owned by the signed-in account.
devices_v2
ChatGPTList paired devices for the signed-in Cmdexa account. Use view=active for normal device selection; view=history shows revoked pairings and view=all preserves the full audit view. Offline active devices retain their last reported hostname and agent/helper versions. Pairing time is not connection time; only report a connection time when connected_at or last_seen is present. If an offline device has no last_seen or reported metadata, say it has not reported a connection yet. If none are paired, use the getting_started guidance.
execute_batch_v2
ChatGPTStart asynchronous commands on a device paired to the signed-in account. Always send a stable request_id and reuse it unchanged if the response is lost: Cmdexa then returns the original tasks instead of dispatching again, and rejects changed content under the same id. Without it a lost response cannot be recovered safely; use tasks_resume_v1.
execute_v2
ChatGPTStart one asynchronous command on a device paired to the signed-in account. Local confirmation rules still apply. Always send a stable request_id (one unique id per logical step) and resend the identical request with the same id if the response is lost: Cmdexa then returns the original task instead of running it twice, and rejects changed content under the same id. Without a request_id a lost response cannot be told from a deliberate repeat; recover with tasks_resume_v1 before re-sending.
execution_preflight_v1
ChatGPTStart an asynchronous PowerShell syntax and command availability diagnostic on an authorized paired Windows PC. Creates and dispatches a task record, so this is a state-changing tool even though the probe does not execute the submitted script or discovered commands. Retrieve the task ID using task_result_v1 and inspect PowerShell version, tools_available and syntax.valid. Local approval and device permissions still apply.
list_task_history_v1
ChatGPTList existing task IDs, statuses and outcomes belonging to the signed-in account. Read-only lookup of retained and expired task metadata after a lost execute response; this never starts, restarts or resumes a command. Filter by device, exact execute request_id or workflow prefix. Follow pagination when has_more or expired_has_more is true; inspect terminal tasks with task_result_v1.
longscribe_failed_jobs_v1
ChatGPTRead bounded historical failure events from the fixed LongScribe HTTPS origin via a paired Windows user session. No token or target URL may be passed in arguments. Network confirmation and desktop approval are required. Poll task_result_v1 and follow next_cursor until null.
ssh_diagnostics_v1
ChatGPTStart bounded SSH diagnostics as the logged-in Windows user with fresh local approval. Local mode checks metadata and known-host entries; network mode additionally checks DNS/TCP after confirmation. No remote commands, private-key content, handshake, authentication or trust changes. Poll task_result_v1 for the structured report; reachable TCP and known-host entries never verify server identity.
support_diagnostics_bundle_v1
ChatGPTCreate a bounded diagnostics bundle containing only data belonging to the signed-in account.
task_output_chunks_v1
ChatGPTRead output chunks for a task owned by the signed-in account.
task_result_v1
ChatGPTWait briefly for a task owned by the signed-in account and return its redacted result. Use include_output=false for status-only reads, read output incrementally with task_output_chunks_v1, and stop polling when terminal=true. Reuse request_id on execute retries; do not create duplicate tasks or poll hundreds of tasks in parallel.
task_status
ChatGPTRead a task owned by the signed-in account.
tasks_resume_v1
ChatGPTRecover recent task IDs and states after a chat disconnect or a lost execute response. Read-only, bounded metadata without command text or stdout. Each task has an outcome (succeeded, failed, timed_out, cancelled, not_started, or outcome_unknown, which is never success). Filter by device, the exact execute request_id, or a whole <workflow>/ request_id_prefix; expired (retention-removed) tasks are listed separately. Lists hold at most 20 entries and are complete only when has_more and expired_has_more are false: follow cursor / expired_cursor. Works while the device is offline. This reads existing tasks and never executes or resumes commands. Fetch a terminal result once using task_result_v1; for active tasks use include_output=false and respect poll_after_ms.
tool_history_v1
ChatGPTReturn recent persisted MCP tool calls and retained usage counts. Arguments are secret-redacted and each JSONL record is bounded to 4 KiB.
wait_task_v2
ChatGPTWait for new events on a task owned by the signed-in account.
Capabilities
Writes
Example Prompts
Click any prompt to copy it.
