WPHammer
Steward
Supervised WordPress operations

Steward

A governed agent layer for WPHammer. Steward investigates your WordPress fleet, plans the safest next action, and waits for your approval before executing. Not unrestricted automation. Supervised operation with bounded autonomy.

How it works

Two layers, one operating surface.

Layer A: Agent control plane

The operator brain

Built on Laravel AI. Four specialized agents understand your fleet, plan work, gather context, and decide when to ask for approval. Each agent has focused instructions and a scoped tool set.

WordPress Caretaker
Updates, patches, auto-update policies, security remediation
Server Diagnostician
Health metrics, load correlation, incident analysis
Site Operations
Clone, backup, restore, staging plans
Security Triage
Finding prioritization, patch assessment, remediation recommendations
Layer B: Operational tool plane

The hands

Built on Laravel MCP. 22 typed tools that expose WPHammer capabilities through a structured, authenticated API. Every tool enforces team scope and produces structured output.

Discovery
11 tools
List sites, server health, WordPress risk queue, security findings
Planning
4 tools
Plan patch, plan clone, plan restore, plan remediation
Execution
7 tools
Queue patch, enable auto-update, start backup, clone site, restore
MCP tool reference

Every tool the agent can reach.

All tools are team-scoped via Sanctum authentication. Read-only tools are safe by default. Write tools validate preconditions, check approval mode, and emit activity records.

Read-only discovery
ListSites
Sites with server, WP collection, canary state
ListServers
Servers with site count, disk, provider
GetSiteSummary
Full site detail with WP inventory, security, backup, uptime
GetServerSummary
Server detail with hosted sites, onboarding, disk
GetWordPressRiskQueue
Pending updates with version, auto-update, status
GetServerHealthSummary
24h metrics with min/max/avg per metric type
GetSiteSecuritySummary
Active findings grouped by severity and patch status
GetBackupSummary
Config, last backup, recent history, time capsule state
GetCanarySummary
Baseline state, recent checks
GetUptimeSummary
Monitoring config, current status, recent incidents
GetActivitySummary
Recent job activity and activity logs
Planning
PlanWordPressPatch
Segmented plan: auto-update, inactive, needs-review, skipped
PlanSiteClone
Source/dest validation, domain/DB plan, deployment steps, risks
PlanBackupRestore
Backup validation, age check, restore steps, pre-restore backup
PlanSecurityRemediation
Severity-prioritized recommendations per finding
Execution
QueueAutoUpdateEnable
Enable WordPress auto-update for a plugin or theme
QueueWordPressPatch
Dispatch update job across specified sites
QueueCanaryCheck
Run visual change detection against baseline
QueueSecurityScan
Run integrity, file, vulnerability, and cron checks
QueueBackup
Start a backup deployment
QueueSiteClone
Clone site to another server (dual-review required)
QueueRestore
Restore from backup (dual-review required)
Approval workflow

Every write action declares how it wants to be reviewed.

Auto

Safe actions allowed by team policy. Enable auto-update, run scans, start backups.

Review

Default write path. Patch plugin across 12 sites. Human sees the plan and approves.

Dual review

Destructive or high-risk. Clone site, restore backup. Requires typed CONFIRM.

Blocked

Action not allowed in current scope. Agent explains why and what would need to change.

Runtime policies

Define exactly what each pilot can touch.

Every agent run is bound to a named runtime policy. The policy defines the server and site allowlist, available tools, action budgets, and whether destructive operations are permitted. This is enforced server-side, not by prompt instructions.

Allowed servers
Explicit server IDs or all
Allowed sites
Explicit site IDs or all
Allowed tools
Whitelist of MCP tools
Max sites per action
Blast radius cap
Daily action budget
Auto-resets each day
Weekly action budget
Auto-resets each week
Autonomy mode
Plan only, review required, or low-risk auto
Destructive actions
Explicitly enabled or blocked
Dry-run mode
Full plans, zero execution
Ready to see it

Steward is built into WPHammer. Log in to try it.

Create a runtime policy, scope it to a few sites, and let the WordPress Caretaker review your fleet. See what it finds before you give it any write permissions.