Back to skill

Security audit

Nirvana

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent local AI routing plugin, but its privacy promises, cloud-routing controls, permissions, and Docker install guidance are too broad and inconsistent for automatic installation.

Review this before installing on any sensitive OpenClaw environment. Use local-only settings, verify that cloud routing is actually disabled, do not enter secrets in prompts unless you have confirmed no cloud provider can be invoked, bind Docker services to localhost, replace fixed credentials, pin images/releases, and narrow manifest permissions to plugin-owned files and Ollama API access.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Findings (6)

T09 · Insecure Skill Coding Practices

Error
Location
src/router.js:15
Finding

Cloud fallback opt-out is ignored and raw queries bypass privacy sanitization

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
plugin-manifest.json:67
Finding

Plugin manifest requests unnecessary access to identity, memory, session, and execution capabilities

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
INSTALL.md:68
Finding

Installation instructions expose services on all interfaces and deploy a hardcoded database credential

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
INSTALL.md:68
Finding

Installation workflow executes mutable unpinned images and plugin releases

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
src/privacy-auditor.js:15
Finding

Configuration-controlled log paths allow arbitrary writable-file modification

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
src/router.js:440
Finding

Truncated Base64 routing-cache keys permit cross-query decision collisions

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (30)

Privileged Container / Container Escape

High
Category
Privilege Escalation
Confidence
87% confidence
Finding

Passing host device nodes such as '/dev/kfd' into a container materially increases host attack surface and weakens isolation. If the container image or workload is compromised, device access can enable deeper host interaction, data exposure, or kernel-driver attack paths.

Content

Scanner excerpt · INSTALL.md (reported line 229)May include surrounding context.

bash
docker run -d \
  --device /dev/kfd \
  --device /dev/dri \
  --group-add video \
  ollama/ollama:rocm

Privileged Container / Container Escape

High
Category
Privilege Escalation
Confidence
87% confidence
Finding

Exposing '/dev/dri' to the container grants direct access to GPU/DRM interfaces and similarly reduces container isolation. In an installation guide for local inference this may be operationally necessary, but it remains a real security-sensitive instruction that should be clearly caveated.

Content

Scanner excerpt · INSTALL.md (reported line 230)May include surrounding context.

bash
docker run -d \
  --device /dev/kfd \
  --device /dev/dri \
  --group-add video \
  ollama/ollama:rocm

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The documentation provides a configuration snippet to disable privacy enforcement while only labeling it 'Not Recommended,' without clearly describing the consequences. Disabling context-boundary enforcement can permit sensitive data leakage across requests, plugins, or cloud routing paths and undermines the core privacy claims of the skill.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · MIGRATION.md (reported line 39)May include surrounding context.

cp memory/audit.log archive/cloud-era/$(date +%Y%m%d)/

Start fresh Nirvana logs

rm memory/nirvana-metrics.json memory/nirvana-audit.log

text

### Migrate Cached Responses

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The schema advertises cloud model fallback without a clear user-facing warning that prompts, context, or derived data may be transmitted to external APIs. In a skill handling agent memory and identity files, omission of this warning materially increases the chance that operators will enable fallback without understanding the privacy consequences.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The comment says certain identity files are 'Never exported', but reconfigure() allows identityFilesNeverExport to be replaced, weakening or removing those protections entirely. Because isIdentityFile() relies solely on this mutable list, an attacker or unsafe caller can reconfigure the stripper to permit export of files that the documentation claims are permanently blocked.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
75% confidence
Finding

Docker image references without a specific tag (:latest is implicit) or digest (@sha256:...) can be silently replaced by a malicious image.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
75% confidence
Finding

Docker image references without a specific tag (:latest is implicit) or digest (@sha256:...) can be silently replaced by a malicious image.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The guide enables cloud fallback after presenting the system as 'thinking locally' and privacy-protective, but it does not prominently warn that queries may be transmitted to a third-party provider once fallback is enabled. This can mislead users into sending sensitive prompts or context off-host under the assumption that privacy guarantees still hold.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · INSTALL.md (reported line 138)May include surrounding context.

docker-compose ps

Check Ollama health

curl http://localhost:11434/api/tags

text

## Verify Installation

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The documentation includes destructive log removal commands without an explicit warning about data loss, retention obligations, or the need to confirm backups. Users may copy-paste the commands and permanently delete audit or metrics records needed for incident response, compliance, or troubleshooting.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The guide states that identity 'stays local, never exported' while the provided configuration explicitly enables cloud fallbacks to third-party providers. This creates a misleading privacy guarantee that could cause operators to deploy under false assumptions and inadvertently send sensitive data off-device.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The expected-results section claims 'Zero identity files exported' even though the migration path keeps cloud fallback enabled. This inconsistency can mislead users into believing the deployment is privacy-preserving when it may still transmit sensitive content to external providers.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The success criteria require 'Zero identity files exported to third parties' despite the documented rollback and steady-state configurations allowing cloud fallback. This sets an unattainable or misleading compliance target and may conceal actual privacy exposure during operation.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The README instructs users to run docker run -d -p 11434:11434 ollama/ollama without pinning the image to a specific tag or digest. This causes deployments to pull whatever image is current at execution time, creating a supply-chain and reproducibility risk if the upstream image changes unexpectedly or is compromised.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The description advertises optional cloud fallback but does not provide a prominent warning, at the point of introduction, that user prompts may be sent to third-party providers such as Anthropic, OpenAI, or Gemini. In a privacy-focused skill that markets 'AI sovereignty' and 'zero API keys required,' this omission is especially risky because users may reasonably assume all processing remains local and may disclose sensitive data under that assumption.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The installation instructions run the ollama/ollama container without pinning a specific tag or immutable digest, so users may pull whatever image is current at execution time. This creates supply-chain risk because a future upstream image change, compromise, or breaking update could introduce malicious code or unsafe behavior into the local environment.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill states that any normal OpenClaw interaction will automatically route through Nirvana and that the plugin internally decides between local and cloud providers. Broad automatic interception of ordinary usage can cause users to invoke the skill without clear intent, increasing the chance that sensitive prompts are processed or forwarded in ways the user did not explicitly request.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

Automatic model download is enabled by default, which creates implicit outbound network activity and pulls code or model artifacts without an explicit user warning. While this is not necessarily malicious, it can violate operator expectations in restricted or sensitive environments and may introduce supply-chain or egress-governance concerns.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The schema enables cloud fallback but does not define sufficiently strict conditions for when local requests may be escalated to external providers. In an agent setting, ambiguous fallback behavior can cause prompts or context to be sent off-box more broadly than users expect, increasing risk of unintended data disclosure even if some privacy controls exist elsewhere in the schema.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest explicitly advertises a 'cloud fallback optional' design but does not define any gating conditions, user-consent requirements, or data-classification limits for when local prompts may be sent to a remote provider. In a plugin that also claims privacy enforcement and reads sensitive agent state files, ambiguous fallback behavior can cause unintended exfiltration of prompts, memory, or identity-bearing context to external services.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The plugin declares broad 'query-router' and 'decisionEngine' capabilities without clear scope boundaries, exclusions, or routing policy constraints. Because the same manifest also grants access to sensitive context files and response/memory integration, an overly broad router could intercept arbitrary user or agent queries and redirect them in ways that bypass expected privacy or trust boundaries.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The module-level documentation asserts a privacy boundary ('Removes identity before sending queries to cloud APIs'), but strip() returns the original context unchanged whenever enforceContextBoundary is false. In a security-sensitive context-stripper, this creates a fail-open path that can expose all context, including personal or identity-bearing data, to external services if configuration is absent, disabled, or tampered with.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The cloud execution path transmits user queries and stripped context to an external provider, creating a data egress channel for potentially sensitive information. Although a context stripper is used, this file provides no explicit consent gate, disclosure, or verification that stripping is sufficient before external transmission.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The plugin records raw query content into metrics, which can capture sensitive prompts, secrets, personal data, or proprietary information. Even if intended for observability, collecting full queries without minimization, consent, or clear controls increases privacy risk and broadens the impact of any log exposure or insider access.

Content

No source excerpt is available for this finding.