Documentation
Runbooks, approvals and audit
How a standard operating procedure becomes a governed runbook, and what the audit trail records.
Authoring
- Start from an existing SOP (document or ticket template).
- REAP compiles it into steps in natural language with explicit inputs, checks and stop conditions.
- Classify each step: read (diagnostic) or write (remediation).
- Attach guardrails: required approvals, and protected devices, forbidden commands and blast-radius limits which are enforced today.
- Review the generated commands (validated when the runbook is compiled); dry-run against a device ; publish to the roles allowed to run it.
Autonomy modes
Design and today The table is the design. Today: recommend-only, and approve-then-act for approved runbooks through the console. Per-action autonomy configuration is in development.
| Mode | Behavior | Status today |
|---|---|---|
| Recommend only | REAP proposes the runbook and the evidence; a person executes or declines | Available · the default |
| Approve then act | REAP executes after a named approver confirms in the console (in Slack (opens in a new tab) or Teams where confirmed per pilot) | Approved runbooks; a pilot deliverable |
| Act and notify | REAP executes within guardrails and notifies | Not offered; per-action autonomy control in development |
Execution safeguards
- Pre-execution change impact analysis against topology, reachability, security policy and known CVEs, with reasons shown.
- Reverse steps authored into the runbook where the change is reversible; automatic rollback on breached postchecks is in development.
- Credentials revocable by you at any time.
Audit trail
| Recorded | Detail |
|---|---|
| Investigation | Trigger, evidence gathered (with sources and timestamps) and tool calls, conclusion, engineer scoring; a visible log of hypotheses tested and rejected is in development |
| Approval | Approver, time, channel, evidence presented |
| Action | Commands issued, device, output, result |
| Access | Console logins, role changes, credential events |
| Export | Audit export formats and mechanism |
In a pilot
- Two or three runbooks under approval (week 4); zero unapproved actions is a scorecard measure.
- Diagnostic (read-only) runbooks are available; approval-gated remediation is delivered in pilots.
Last reviewed: October 7, 2026 (draft) · Owner: solutions engineering · Corrections: security@reapnet.ai