Vtell

Execution reports

vtell --report --round N --message-id msg-UUID < result.txt
vtell <planner-session> --report --round N --message-id msg-UUID < result.txt

Read vflow status <workflow-id> to obtain the current round. --report is for the assigned executor: the backend routes one submission according to the saved review setting. With independent review enabled, the full report goes to Review and a progress/reference notice to Plan, entering reviewing. With review disabled, the full report goes to Plan, entering summarizing; Plan does not perform technical review. Legacy workflows without a review setting retain planner review. An explicit target must resolve to the saved planner or reviewer. When omitting the target, supply the report on stdin. Use the recorded round, not a guessed or incremented value.

A blocked workflow still accepts ordinary messages and the assigned executor's current-round report. Submitting that report enters the appropriate review or summary phase without another dispatch or a round increment. Progress updates use ordinary vtell and leave the workflow state unchanged. A completed or explicitly stopped workflow cannot accept a new execution report.

Include changed files, what changed, checks actually run and their results, remaining omissions and review evidence. Submit only when this round is ready for review. Reporting is the last action that affects the work: do not edit files afterward, because the reviewer may inspect them immediately. End the turn and wait for feedback in the same conversation. A real blocker still uses vflow block. Planning, dispatch, review acceptance, final summary (finish), stopping and status remain vflow operations.

When independent review is enabled, inspect the returned deliveries for both targets. Each target uses a stable submission ID, and retries with the original report ID preserve already-successful deliveries.

Receipts and retries

Text comes from positional arguments or UTF-8 stdin and is limited to 65,536 bytes. Use a quoted message, a quoted heredoc or a file so shell syntax in prose cannot execute. Arguments after -- are literal.

For a distinct message, generate and retain a msg-UUID, target and exact text before sending. If omitted, the command generates an ID and prints it before delivery. Retain that printed ID on failure: retry with --message-id and unchanged target, text and report round. Never generate a new ID to bypass an uncertain receipt. The same ID with different content is rejected.

Report the actual sent, steered, blocked or queued receipt, and tell the user which one came back. sent started a new turn, steered joined a turn already running, blocked joined a turn whose agent is waiting for an answer to a question, and queued is waiting for the running turn to end. Both blocked and queued mean the recipient has not read the message, so report what it is waiting on. None of the four means the recipient has finished its task or accepted a report. A retained message or chat_submission_pending error is not proof of delivery; inspect the recipient with vrefer and workflow receipts with vflow status when applicable. Do not poll continuously or repeatedly send the same instructions.

The receiving chat and archived transcript retain the verified sender identity. This requires a VelaTerm build with the vtell backend and command shim; installing the skill alone cannot add that capability to an older running application.