feat(workflow-executor): audit trail for ai-filled and ai-submitted action forms (PRD-513)

[Scra3]

What

Audit trail for the action-forms epic (PRD-57), on top of PRD-511/512. The raw data (AI-filled values, submitted values, submission outcome) is already persisted; this surfaces it.

Changes

• submittedBy discriminator on the trigger-action executionResult: 'ai' (Full AI — executor submitted) vs 'user' (AI-assisted — human submitted natively).
• Trigger-action step-summary formatter (downstream-AI context / audit): distinguishes an executed action from one submitted for approval ("AWAITING APPROVAL — has NOT been executed"), names the submitter, lists AI-pre-filled fields, and reports the fields the human edited (diff prefill ↔ submitted).

Full AI = first-hand executor data; AI-assisted = derived from the confirmation payload (per the settled architecture).

Tests

1106 tests pass. New formatter tests: pending-approval not-executed, Full AI submitted-by-AI, AI-assisted human-edit diff, skipped → null.

Stacked on feature/prd-512-executor-full-ai-forms. Completes the executor track (509 → 511 → 512 → 513).

fixes PRD-513

Note

Add audit trail for AI-filled and AI-submitted action forms in trigger-action steps

• Extends TriggerRecordActionStepExecutionData to include an optional submittedBy field ('ai' | 'user') recorded when a trigger-action step completes.
• Updates TriggerRecordActionStepExecutor.doExecute to persist submittedBy: 'ai' for fully automated runs and submittedBy: 'user' for user-confirmed flows.
• Adds a formatTriggerAction formatter in StepExecutionFormatters that produces a human-readable audit summary: pending-approval state, submitter identity, AI pre-filled fields, and any fields edited by the user (diffed against AI values).

^{Macroscope summarized 2b56b54.}

[linear-code]

PRD-513

[qltysh]

1 new issue

Tool	Category	Rule	Count
qlty	Structure	Function with many returns (count = 5): format	1