Compliance audit workflow

Orchestrates a deployed Mistral Workflow for compliance auditing. Uses workflow_interact(action: "query") to inspect live findings mid-run, and workflow_interact(action: "signal") to approve, escalate, or skip at each decision checkpoint.

Profile note: workflow_execute, workflow_status, and workflow_interact are available in the core profile (default). files_upload (for uploading target documents) requires MISTRAL_MCP_PROFILE=full.

Temporal behavior: while the workflow is processing or waiting for a signal, its status remains RUNNING. Detect "waiting for human input" by querying a status handler, not by checking for a PAUSED status (which doesn't exist in this API).

Important: handler names are workflow-specific

Ask the user (or consult mistral://workflows) for:

• The query handler that exposes current findings (e.g. "get_findings")
• The query handler that exposes checkpoint state (e.g. "get_checkpoint")
• The signal handler that receives audit decisions (e.g. "audit_decision")

Steps

Step 1 — Collect inputs

Ask the user for:

1. workflowIdentifier — the deployed audit workflow name or ID
2. scope — compliance framework: "GDPR", "SOC2", "PCI-DSS", "ISO27001", etc.
3. Target documents: file_ids (upload via files_upload — requires full profile) or public URLs passed as workflow input
4. Optional: audit_depth — "quick" / "standard" / "full"
5. Handler names (if non-default)

Step 2 — Launch the audit

Call workflow_execute:

{
  "workflowIdentifier": "<workflow name or ID>",
  "input": {
    "scope": "<GDPR|SOC2|PCI-DSS|ISO27001>",
    "document_ids": ["<file_id_1>", "<file_id_2>"],
    "audit_depth": "standard"
  }
}

Note structuredContent.execution_id. Confirm: "Audit workflow started — execution ID: <execution_id>."

Step 3 — Poll and query mid-run findings

Loop:

1. Call workflow_status with { "executionId": "<execution_id>" }
2. Check structuredContent.status:

• COMPLETED → go to Step 5
• FAILED / CANCELED / TIMED_OUT → surface error and stop
• RUNNING → continue

While RUNNING, proactively query partial findings every ~15 seconds:

{
  "executionId": "<execution_id>",
  "action": "query",
  "name": "get_findings"
}

Surface partial findings to the user as they accumulate:

📊  AUDIT IN PROGRESS (RUNNING)
───────────────────────────────
Controls checked: [N from result]
Findings so far:
  ✅ [control name] — PASS
  ❌ [control name] — FAIL: [reason]
  ⚠️  [control name] — WARNING

Also probe for decision checkpoints:

{
  "executionId": "<execution_id>",
  "action": "query",
  "name": "get_checkpoint"
}

If the result indicates a decision is required, proceed to Step 4.

Step 4 — Handle decision checkpoints

When a checkpoint is detected from the query result:

1. Display the checkpoint context:

   ⏸  CHECKPOINT — Decision required
   ──────────────────────────────────
   Control: [control name from result]
   Finding: [description]
   Risk:    [level]

1. Ask the user: "Accept / Escalate / Skip this finding?"

1. Signal the decision:

Accept (finding acknowledged, workflow continues):

   {
     "executionId": "<execution_id>",
     "action": "signal",
     "name": "audit_decision",
     "input": { "decision": "accept", "note": "<optional note>" }
   }

Escalate (flag for external review):

   {
     "executionId": "<execution_id>",
     "action": "signal",
     "name": "audit_decision",
     "input": { "decision": "escalate", "note": "<reason for escalation>" }
   }

Skip (exclude this control from scope):

   {
     "executionId": "<execution_id>",
     "action": "signal",
     "name": "audit_decision",
     "input": { "decision": "skip", "note": "<justification>" }
   }

Return to Step 3. A full audit may have multiple checkpoints.

Step 5 — Deliver the audit report

When status === "COMPLETED", format structuredContent.result:

✅  COMPLIANCE AUDIT COMPLETE
──────────────────────────────
Framework: [scope]   Execution: <execution_id>

SUMMARY
  ✅ PASS:       [N] controls
  ❌ FAIL:       [N] controls
  ⚠️  WARNING:   [N] controls
  ↗️  ESCALATED: [N] controls
  ⏭  SKIPPED:   [N] controls

FAILED CONTROLS
───────────────
[control] — [finding] — Remediation: [suggestion]

ESCALATED ITEMS
───────────────
[control] — [reason]

RECOMMENDED REMEDIATIONS
─────────────────────────
[prioritized list from result]

Offer to export as JSON for GRC tooling or pass escalated items to mistral_chat with mistral-large-latest for remediation plan drafting.