- name
- debug-execution
- description
- Diagnose an n8n workflow that failed, produced the wrong result, or stopped running. Use for execution errors, unexpected output, intermittent failures, and workflow repairs.
- compatibility
- Requires Kiro with Agent Plugins support and n8n 2.34.0 or later with execution read access.
- metadata
- {"author":"n8n","version":"1.0.0"}
# Debug executions from Kiro
Apply [connect-n8n](../connect-n8n/SKILL.md), including its loading policy and
execution checks. Keep investigation and repair on that verified connection.
Load the [shared router](../connect-n8n/references/n8n-skills/skills/using-n8n-skills-official/SKILL.md)
and [debugging guide](../connect-n8n/references/n8n-skills/skills/n8n-debugging-official/SKILL.md)
when absent from context, before workflow investigation calls.
## Ground the shared investigation in the actual run
- Use the user's execution or search retained runs with `search_workflow_executions`
for the workflow and reported time before requesting another run. Filter failures
for execution errors; include successful runs for incorrect output. If it stopped
running, check the last run. This replaces the shared shortcut to request a rerun
when no execution ID is supplied.
- Compare against the workflow version saved with that execution; the current
draft may differ. Use available history and diffs before editing. If execution
data has expired, state which evidence is missing.
## Apply a repair
Read `update_workflow`'s advertised operation schema; these objects are not
Workflow SDK factory calls. Follow connect-n8n's missing-schema handling.
For an edit or test, apply [workflow lifecycle](../connect-n8n/references/n8n-skills/skills/n8n-workflow-lifecycle-official/SKILL.md),
loading it if absent from context, and follow the connection's execution checks.
Workflow execution tools do not expose standalone Agent conversations, except
for workflows those Agents ran as tools.
View on GitHub