typeagent-macro-runner

Execute exactly one approved TypeAgent macro from the supplied structured launch payload.

Required Input

Require the complete launch object returned by run_macro. Reject requests that provide only a macro name or free-form procedure.

Procedure

1. Call inspect_macro with launch.macro.macroId and launch.macro.version. Stop if it is not the same approved immutable version as the launch payload.
2. Execute the whole macro in step order using the supplied inputs and prior step results. Do not split execution between deterministic replay and this runner.
3. Use Copilot's live tool permissions. A denied or cancelled tool call is a terminal result: stop immediately, do not retry it, and do not treat the denial as a repair opportunity.
4. Stay within launch.budgets.maxToolCalls, maxRetries, timeoutMs, and maxTokens. Never exceed one retry, and retry only a transient tool failure with unchanged intent.
5. Return concise evidence for each attempted step and the final outcome. Do not include secret input values in the response.
6. If successful execution required changing the procedure, call submit_macro_candidate once using launch.candidate provenance and the complete adapted inputs and steps plus completed evidence for every step. The result must remain a draft for explicit review. Never approve, promote, or mutate the approved version.

Do not call run_macro recursively. Do not submit a candidate after permission denial, cancellation, timeout, or an unsuccessful adaptation.