[power-automate][Bug] diagnose_run throws "ctx.getClient(...).diagnoseRun is not a function"

[AJx03]

Plugin

power-automate

Plugin Version

2.2.0

Skill / Command

debug-flow

Bug Description

Calling the diagnose_run MCP tool always throws a TypeError instead of returning a diagnosis, for any valid flow/run combination. The same class of bug (an unwired/missing method on the FlowAgent client) was already tracked for edit_flow in #314, now fixed by PR #318 (adding FlowClient.editFlow()). copy_flow has the same unfixed pattern in #289/#294. This appears to be the same root cause affecting diagnoseRun, which doesn't yet have its own issue.

Steps to Reproduce

1. Have a flow with at least one failed run (any cloud flow, any environment).
2. Call diagnose_run with a valid flow (flow ID) and run (run ID) that get_run_actions can successfully retrieve for the same flow/run.
3. Observe the tool call fails immediately.

Expected Behavior

diagnose_run returns the classified failed/timed-out actions with a remediation for each, as described in its tool definition.

Actual Behavior

Tool call fails with:

TypeError: ctx.getClient(...).diagnoseRun is not a function

get_run_actions works fine against the same flow/run, confirming the run/flow IDs and connection are valid — the failure is isolated to diagnose_run's client wiring.

Relevant Logs / Screenshots

{
  "code": "tool-error",
  "message": "ctx.getClient(...).diagnoseRun is not a function",
  "errorName": "TypeError"
}

Environment

• OS: Microsoft Windows 11 Enterprise, 10.0.26100.0
• Claude Code version: 2.1.220 (Claude Code)

[SaiprasadDash]

Hi! I'd like to work on this. I see this is the same root-cause pattern as
#314 (fixed in PR #318 by adding FlowClient.editFlow()) — I'll follow the
same approach and add a FlowClient.diagnoseRun() method wired to the
diagnose_run MCP tool. Will also check if copy_flow (#289/#294) shares
this exact gap while I'm in this code, in case that's useful context.

[radioblazer]

This is the same root-cause pattern as #314 — the MCP tool definitions call methods on FlowClient that were never implemented in the class body. The fix adds all 5 missing run management methods to FlowClient:

• cancelRun: POSTs to /runs/{id}/cancel
• resubmitRun: Auto-detects the trigger name from the run, then POSTs to /triggers/{trigger}/histories/{runId}/resubmit
• getRunActionRepetitions: Paginated GET for loop iteration details
• diagnoseRun: Fetches run details + actions, classifies each failed/timed-out action with a remediation (auth → re-authenticate, throttling → retry-after, 404 → verify resource, etc.)
• cancelAllRuns: Tries Dataverse CancelAllCloudFlowRuns bulk action first (single atomic call), falls back to per-run cancel when no Dataverse instance is available

All 6 associated unit tests now pass. Will include in the next plugin bundle PR.

[radioblazer]

Fixed in v3.0.1 — resolved in PR #393, merged 2026-08-12.

Please update to v3.0.1 and let us know if you're still seeing this.