Skip to content

v0.16.12

Choose a tag to compare

@rcb0727 rcb0727 released this 17 Aug 02:43
· 3 commits to main since this release

[0.16.12] - 2026-08-16

  • Bug fix — an action whose inputs is an array is no longer stored as an object. typeof [] === "object", and the preprocessor that fixes expression-bearing payloads spreads its argument into an object literal — so a Compose holding a literal list came back as {"0":"a","1":"b"}. create_flow reported success and the flow then failed at runtime with "the result of the evaluation of 'foreach' expression … is of type 'Object'. The result must be a valid array." Found while provisioning columns across 18 SharePoint libraries from a list held in a Compose. An array has no body/parameters to rewrite, so it now passes through untouched; update_flow's deep merge already guarded this, which is why only newly created flows were affected.
  • Bug fix — a destructive tool called without confirm: true reported SUCCESS while doing nothing. The confirmation prompt was returned as ordinary text, so a refusal was indistinguishable from a completed delete: the MCP result carried no isError, and a script that deleted 59 SharePoint items and checked only that flag reported 59 deletions and performed none. All 36 refusals across 15 tools (delete_flow, delete_sharepoint_item, delete_environment, delete_powerapp, solution/connector/work-queue/Power Pages deletes, and the restore/reset/import gates) now raise a ConfirmationRequiredError, which is rendered verbatim so the prompt keeps the id you must confirm against. The interactive flow is unchanged — read the prompt, call again with confirm: true. delete_flow and delete_custom_connector also had their best-effort name lookup restructured: the prompt now names the flow or connector instead of falling back to the id.
  • Bug fix — get_flow now tells you which connection a flow is bound to. A flow's connection reference carries three easily-conflated things: the reference key (shared_jira), the connector type, and the actual bound connection id (connectionName). Only the first two were reported, so "which connection is this trigger using?" could not be answered from the tool — during a live trigger outage the binding had to be read out of the maker portal by hand while the tool confidently displayed the connector name. connectionName (and source) now appear in both the structured output and the summary, which also makes read → rebind a round trip since it is exactly the field update_flow accepts back.
  • Regression test — an explicitly requested environment now provably reaches the connections API URL. list_connections honouring its environment argument was fixed in 0.16.10, but the wire-level test only ever called the API with no argument, so it pinned the configured environment and would have stayed green through the original defect. Connections are environment-scoped, and getting this wrong answers confidently about the wrong environment.