vmware-policy

name	vmware-policy
description	Unified audit logging, policy enforcement, and input sanitization for the entire VMware MCP skill family. Use when querying audit logs, managing policy rules, or when any VMware skill needs audit/policy infrastructure. Provides the @vmware_tool decorator that wraps all 156+ MCP tools across 8 skills. Use when user asks to "show audit log", "check denied operations", "view policy rules", "audit stats", or "query audit trail". For VM lifecycle use vmware-aiops, for monitoring use vmware-monitor, for networking use vmware-nsx, for load balancing use vmware-avi.
installer	{"kind":"uv","package":"vmware-policy"}
allowed-tools	["Bash"]
user-invocable	false
metadata	{"openclaw":{"requires":{"bins":["vmware-audit"],"config":["~/.vmware/rules.yaml"]},"optional":{"env":["CLAUDE_SESSION_ID","OLLAMA_HOST"]},"homepage":"https://github.com/zw008/VMware-Policy","emoji":"🛡️","os":["macos","linux"]}}

VMware Policy

Disclaimer: This is a community-maintained open-source project and is not affiliated with, endorsed by, or sponsored by VMware, Inc. or Broadcom Inc. "VMware" and "vSphere" are trademarks of Broadcom. Source code is publicly auditable at github.com/zw008/VMware-Policy under the MIT license.

Unified audit logging, policy enforcement, and input sanitization -- the infrastructure layer for the entire VMware MCP skill family.

Infrastructure dependency: All 8 VMware skills depend on vmware-policy. It is auto-installed and provides the @vmware_tool decorator, sanitize(), and the shared audit database. Family: vmware-aiops (VM lifecycle), vmware-monitor (read-only monitoring), vmware-storage (iSCSI/vSAN), vmware-vks (Tanzu Kubernetes), vmware-nsx (NSX networking), vmware-nsx-security (DFW/firewall), vmware-aria (metrics/alerts/capacity), vmware-avi (AVI/ALB/AKO). | vmware-pilot (workflow orchestration)

What This Skill Does

Category	Components	Count
Audit Logging	AuditEngine (SQLite WAL), log rotation, agent detection	3
Policy Engine	deny rules, maintenance windows, hot-reload	3
Sanitization	sanitize() -- prompt injection defense, control char stripping	1
Decorator	@vmware_tool -- pre-check + execute + post-log + metadata	1
CLI	vmware-audit log, vmware-audit export, vmware-audit stats	3

Quick Install

uv tool install vmware-policy
vmware-audit stats          # verify installation

vmware-policy is automatically installed as a dependency of all VMware skills. Manual install is only needed for standalone audit querying.

When to Use This Skill

• Query the unified audit trail across all VMware skills
• View denied operations and policy violations
• Check audit statistics (by skill, by status, by time range)
• Export audit logs as JSON for external analysis
• Configure deny rules or maintenance windows
• Integrate the @vmware_tool decorator into a new VMware skill

This skill is auto-loaded as a dependency -- you do not need to invoke it directly. It activates when:

• Any VMware skill tool function is called (via @vmware_tool decorator)
• User asks about audit logs, denied operations, or policy rules
• User runs vmware-audit CLI commands

Related Skills -- Skill Routing

User Intent	Recommended Skill
VM lifecycle, deployment, guest ops	vmware-aiops (uv tool install vmware-aiops)
Read-only monitoring, zero risk	vmware-monitor (uv tool install vmware-monitor)
Storage: iSCSI, vSAN, datastores	vmware-storage (uv tool install vmware-storage)
Tanzu Kubernetes (vSphere 8.x+)	vmware-vks (uv tool install vmware-vks)
NSX networking: segments, gateways, NAT	vmware-nsx (uv tool install vmware-nsx-mgmt)
NSX security: DFW rules, security groups	vmware-nsx-security (uv tool install vmware-nsx-security)
Aria Ops: metrics, alerts, capacity	vmware-aria (uv tool install vmware-aria)
Load balancer, AVI, ALB, AKO, Ingress	vmware-avi (uv tool install vmware-avi)
Multi-step workflows with approval	vmware-pilot
Audit log query, policy rules	vmware-policy -- this skill

Common Workflows

Query Recent Audit Activity

1. View last 20 audit entries: vmware-audit log --last 20
2. Filter by skill: vmware-audit log --skill vmware-nsx --last 50
3. Check denied operations: vmware-audit log --status denied --since 2026-03-28
4. View aggregate stats: vmware-audit stats --days 7

Set Up Policy Rules for Production

1. Copy default rules: cp $(python -c "import vmware_policy; print(vmware_policy.__file__.replace('__init__.py','rules_default.yaml'))") ~/.vmware/rules.yaml
2. Edit ~/.vmware/rules.yaml -- add deny rules for production:

   deny:
     - name: no-delete-in-prod
       operations: ["delete_*", "cluster_delete"]
       environments: ["production"]
       reason: "Destructive operations blocked in production"
   maintenance_window:
     start: "22:00"
     end: "06:00"
   

3. Rules hot-reload automatically -- no restart needed
4. Verify: vmware-audit log --status denied to see blocked operations

Export Audit Logs for Compliance

1. Export all logs as JSON: vmware-audit export --format json > audit-export.json
2. Filter by skill: vmware-audit export --skill vmware-aiops --since 2026-01-01
3. Import into your SIEM or compliance tool

Usage Mode

Scenario	Recommended	Why
Query audit logs	CLI	vmware-audit provides rich table output
Integrate into a skill	Python API	from vmware_policy import vmware_tool, sanitize
Automated compliance export	CLI	vmware-audit export --format json pipes to any tool

CLI Quick Reference

# View recent audit entries
vmware-audit log --last 20
vmware-audit log --skill vmware-nsx --status denied
vmware-audit log --since 2026-03-28 --tool delete_segment

# Export for compliance
vmware-audit export --format json > audit.json
vmware-audit export --skill vmware-aiops --since 2026-01-01

# Aggregate statistics
vmware-audit stats --days 7
vmware-audit stats --days 30

Full CLI reference: see references/cli-reference.md

Python API

from vmware_policy import vmware_tool, sanitize

# Wrap every MCP tool function
@vmware_tool(risk_level="high", sensitive_params=["password"])
def delete_segment(name: str, env: str = "") -> dict:
    ...

# Sanitize untrusted API responses before returning to LLM
clean_text = sanitize(api_response_text, max_len=500)

Read-Only Gate

This package implements the family's read-only mode; the skills only call it. A prompt instruction ("never modify anything") is advisory and a weak model can ignore it, so apply_read_only_gate() makes the guarantee structural: with the mode on, every write tool is removed from the FastMCP registry before the server serves, and list_tools() never offers it. The model cannot call what it cannot see.

from vmware_policy import apply_read_only_gate

# Call once, after every tool module has registered and before the server runs.
WITHHELD_WRITE_TOOLS: list[str] = apply_read_only_gate(
    mcp, "vmware-aria", config_flag=_config_read_only()
)

It returns the sorted names of the tools it removed (empty when the mode is off) so the caller can log what was withheld, and it is idempotent. skill is the hyphenated skill name, normalised to a per-skill env var (vmware-nsx-security -> VMWARE_NSX_SECURITY_READ_ONLY). config_flag carries the skill's own read_only: setting and is consulted only when neither env var is set -- pass None for skills with no config file (vmware-pilot, vmware-debug, vmware-harden). Resolution order: per-skill env -> family VMWARE_READ_ONLY -> config_flag -> off, default off.

Two helpers answer the same question without touching a registry: read_only_enabled(skill, config_flag) -> bool, and read_only_status(skill, config_flag) -> ReadOnlyStatus(enabled, source, raw, recognised), which adds where the answer came from so a skill's doctor can report it. Doctors should call read_only_status rather than walking the chain themselves -- one implementation means a doctor cannot disagree with the gate that enforces it.

Classification. A tool is withheld unless it is provably read-only. Signals, highest priority first: FORCE_WRITE membership, [WRITE] docstring prefix, readOnlyHint=False annotation, [READ] docstring prefix, readOnlyHint=True annotation, and anything inconclusive counts as a write. The docstring marker outranks the MCP annotation because it has full family coverage, while vmware-harden and vmware-debug register tools through a build_server() factory that passes no annotations at all.

FORCE_WRITE overrides three tools whose [READ] marker under-reports their real effect:

Tool	Skill	Why
vm_guest_download	vmware-aiops	Reads from the guest OS, but writes to an operator-supplied local_path and takes guest credentials
get_supervisor_kubeconfig	vmware-vks	Materialises a session-token credential file at a model-supplied local path
get_tkc_kubeconfig	vmware-vks	Same shape as above

All three are read-only against the managed infrastructure but write a file to a caller-supplied local path with credentials involved -- side effects a locked-down deployment should opt into explicitly rather than receive by default. Tools that write only to a skill's own local store (vmware-harden's DuckDB twin) stay exposed: that store is a cache of observations, not managed infrastructure.

Fail-closed, precisely. Exactly two conditions abort start-up with ReadOnlyGateError: the FastMCP tool registry cannot be enumerated (typically an incompatible mcp version), or a removal did not take effect and a write tool survived the sweep. A switch value that cannot be parsed (VMWARE_READ_ONLY=ture) does not abort -- it resolves to on, with a warning naming the accepted values, because a typo must never leave write tools exposed. A read-only mode that silently degrades to read-write is worse than none, since operators stop checking. Skill authors shipping to operators who run local or small models: see references/agent-guardrails.md for what the gate, the list envelope and the policy engine each remove from the prompt those operators would otherwise hand-write.

MCP Tools (0)

vmware-policy does not expose MCP tools. It is a Python library and CLI consumed by other VMware skills.

Component	Type	Description
@vmware_tool	Decorator	Wraps all 156+ MCP tools across 8 skills
sanitize()	Function	Prompt injection defense for API responses
apply_read_only_gate()	Function	Removes write tools from a skill's MCP registry when read-only mode is on (see above)
read_only_status()	Function	Resolves read-only state + its source, for a skill's doctor to report
AuditEngine	Class	SQLite WAL audit logger with rotation
PolicyEngine	Class	YAML rule evaluation with hot-reload
vmware-audit	CLI	Typer CLI for querying audit trail
detect_agent()	Function	Infers calling AI agent from env vars (see below)

Agent Detection (Transparency Note)

The detect_agent() function in audit.py checks the following environment variables to identify which AI agent is calling the tools. This is read-only inspection for audit logging purposes — no credentials are extracted or stored:

Env Var	Detected Agent	Purpose
CLAUDE_SESSION_ID or CLAUDE_CODE	claude	Claude Code session
OPENAI_API_KEY or CODEX_SESSION	codex	OpenAI Codex session
OLLAMA_HOST	local	Local Ollama model
DEERFLOW_SESSION	deerflow	DeerFlow session
(none matched)	unknown	Unrecognized agent

The detected agent name is stored in the agent column of ~/.vmware/audit.db for audit trail purposes only. No API keys or tokens are logged.

Troubleshooting

"Cannot initialize audit DB" warning

The audit database directory ~/.vmware/ must be writable. Create it manually: mkdir -p ~/.vmware && chmod 700 ~/.vmware.

Policy rules not taking effect

Rules are loaded from ~/.vmware/rules.yaml. Verify the file exists and contains valid YAML. The PolicyEngine hot-reloads on file change -- no restart needed.

Audit log growing too large

The AuditEngine automatically rotates at 100MB, keeping the 5 most recent archives. For manual cleanup: ls ~/.vmware/audit.*.db to see archives.

"PolicyDenied" exception in skill

This means a deny rule in ~/.vmware/rules.yaml matched the operation. Check vmware-audit log --status denied to see the rule name and reason. To temporarily bypass: VMWARE_POLICY_DISABLED=1 (still logged as bypassed).

Decorator not detecting skill name

The @vmware_tool decorator infers the skill name from the module path (e.g., vmware_aiops.ops.vm_lifecycle -> aiops). If the module does not follow the vmware_<skill> convention, the skill is logged as "unknown".

SQLite "database is locked" error

Multiple concurrent skill processes can write to the same audit.db via WAL mode. If locks persist beyond 5 seconds, check for zombie processes holding the database file.

Setup

uv tool install vmware-policy
mkdir -p ~/.vmware

vmware-policy is auto-installed as a dependency of all VMware skills. The ~/.vmware/ directory is created automatically on first audit write.

Full setup guide, security details, and integration instructions: see references/setup-guide.md

Security

• Source Code: github.com/zw008/VMware-Policy
• Config File Contents: ~/.vmware/rules.yaml contains only rule definitions, no credentials
• Webhook Data Scope: N/A -- vmware-policy does not send data externally
• TLS Verification: N/A -- vmware-policy does not make network connections
• Prompt Injection Protection: sanitize() truncates to 500 chars and strips C0/C1 control characters
• Least Privilege: Audit database is local-only (~/.vmware/audit.db), no network exposure

License

MIT -- github.com/zw008/VMware-Policy