Security audit

Agentic Governance

Security checks for vulnerabilities and agentic risk

Overview

The skill is a disclosed governance helper that works on workspace governance files, with a supply-chain caution because its optional dependency install commands are not pinned.

Before installing the full governance stack, verify the publisher and contents of the referenced leegitw packages and prefer pinned or reviewed versions where possible. Expect this skill to read and write governance-related workspace files and generated indexes.

A.I.G

Vulnerability Patterns

Insecure DependenciesIntroduces malicious components through unsafe dependency sources
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
Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally

Findings (1)

T08 · Insecure Dependencies

Warning

Location: SKILL.md:26
Finding: Unpinned Third-Party Skill Installation Creates a Supply-Chain Risk## Vulnerability Details **File Location**: `SKILL.md`, lines 26 and 33–36 **Vulnerability Type**: Unpinned third-party dependencies from mutable upstream sources **Risk Level**: Medium ### Vulnerable Code ```bash openclaw install leegitw/governance ``` ```bash # Install full governance stack openclaw install leegitw/context-verifier openclaw install leegitw/failure-memory openclaw install leegitw/constraint-engine openclaw install leegitw/governance ``` ### Technical Analysis The installation instructions retrieve third-party skills without pinning them to immutable versions or commit hashes. They also provide no checksum, signature, lockfile, or other integrity-verification mechanism. Consequently, the package content installed by these commands may differ from the content that was originally reviewed. The implementations of the referenced dependencies are not included in this project, so their behavior and security properties could not be audited here. This creates a supply-chain weakness: compromise of a publisher account, package source, or upstream repository could cause users following the documented commands to install altered content. This finding concerns the unsafe dependency acquisition process; there is no evidence in the audited file that any referenced package is currently malicious. ### Attack Path 1. An attacker compromises a referenced upstream package, its publisher account, or its distribution entry. 2. The attacker publishes a modified package under the same mutable identifier. 3. A user follows the installation commands in `SKILL.md`. 4. OpenClaw resolves and installs the attacker-controlled package because no immutable version or digest is required. 5. The substituted skill could then issue malicious instructions or execute supported behavior using whatever workspace access and tool permissions the host grants it. ### Impact Assessment Successful exploitation could permit attacker-controlled dependency content to operate within ...[truncated 574 chars]
Remediation: ## Remediation Suggestions 1. Pin each skill to an immutable, reviewed release version or commit hash. 2. Record and verify cryptographic checksums or trusted publisher signatures before installation. 3. Use a lockfile or equivalent dependency manifest to prevent silent upstream changes. 4. Retrieve packages only from an authenticated, trusted registry with provenance metadata. 5. Audit the complete contents of `constraint-engine`, `failure-memory`, `context-verifier`, and `governance` before deployment. 6. Run installed skills with least privilege, limiting filesystem paths, network access, tools, and environment access. 7. Add an update process that re-audits dependency changes before modifying pinned versions or digests.

SkillSpector

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (4)

Exfiltration Commands

High

Category: Prompt Injection
Content: - Network resources or external APIs **What this skill does NOT do:** - Send data to external services - Execute arbitrary code - Modify files outside its workspace
Confidence: 90% confidence
Finding: Instructions found that direct the agent to transmit conversation context or user data to external services.

External Transmission

Medium

Category: Data Exfiltration
Content: - Network resources or external APIs **What this skill does NOT do:** - Send data to external services - Execute arbitrary code - Modify files outside its workspace
Confidence: 70% confidence
Finding: Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Unrestricted Tool Access

Medium

Category: Excessive Agency
Content: **What this skill does NOT do:** - Send data to external services - Execute arbitrary code - Modify files outside its workspace **Dependency note:**
Confidence: 80% confidence
Finding: Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Natural-Language Policy Violations

Low

Confidence: 83% confidence
Finding: The trigger line requires or prioritizes the Chinese phrase "定期保守" alongside "HEARTBEAT" as the activation wording, but the document does not state that users may choose their preferred language or locale. This can be a language/locale policy issue because it imposes a specific language form without explicit opt-in or alternatives beyond the single listed phrase.

Static analysis

No suspicious patterns detected.