Back to skill

Security audit

Duleheng Consulting

Security checks for vulnerabilities and agentic risk

Overview

This skill is a Markdown-only consulting guide, but it should be reviewed because it recommends installing multiple external skills without verification details.

Install the main skill only if you want Chinese-language engineering consulting guidance. Before running any referenced OpenClawmp install commands, inspect each target skill separately, confirm its publisher and permissions, and avoid batch installation unless the registry provides immutable versions and integrity verification.

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 (2)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:252
Finding

Unverified External Skill Installation Instructions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:252-269
Additional Location: SKILL.md:101
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

Vulnerable Code

bash
# Install Duleheng Consulting (ClawHub)
clawhub install duleheng-consulting

# Install Duleheng Consulting (OpenClawmp)
openclawmp install skill/[assetId]

# Install Master Series (Engineering Cost)
openclawmp install skill/4a3a9beaa9dc4a399d56b5a3a5063016  # Yin Yilin

# Install Master Series (Project Management)
openclawmp install skill/b42772e4cb38496096da58226657477c  # Ding Shizhao
openclawmp install skill/5f90e4007aec4faaa0728706a36219b6  # Ren Hong
openclawmp install skill/e0a2d954385b48828d867affac429af6  # Ding Lieyun

# Install Master Series (International Engineering)
openclawmp install skill/82851afe1e50453b82bad3af859e5a48  # Zhang Shuibo

The comments above are translated into English for report-language compliance; the commands and asset identifiers are reproduced unchanged.

Technical Analysis

The Skill directs users to install externally hosted Skills through the clawhub and openclawmp package ecosystems. The referenced packages are not included in the audited repository. Consequently, their source code, instruction content, permissions, manifests, and runtime behavior could not be reviewed.

The installation commands do not pin a cryptographically verifiable artifact hash, provide a trusted source URL, identify a validated publisher key, or require inspection of the downloaded package before activation. Asset identifiers can identify remote packages, but they do not independently demonstrate integrity or publisher authenticity. If an external account, registry, package, or resolution mechanism is compromised, the installed content could differ from what the author intended.

This is a supply-chain exposure rather than evidence that any referenced package is currently malicious. ...[truncated 1333 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin every external Skill to an immutable, versioned artifact rather than a mutable package reference.
  2. Publish and verify a cryptographic digest, such as SHA-256, for each reviewed artifact.
  3. Document the verified publisher identity and use registry signature verification where supported.
  4. Link to auditable source code and record the source revision corresponding to each distributed artifact.
  5. Document the permissions, tools, network access, and filesystem access required by every dependency.
  6. Require users to download and inspect each Skill's manifest, instructions, and scripts before activation.
  7. Run third-party Skills in a sandbox with least-privilege filesystem, process, credential, and network access.
  8. Replace the generic [assetId] placeholder with a documented, immutable, reviewed reference.
  9. Avoid presenting direct installation as a one-step trusted operation; add an explicit verification stage before installation and activation.
  10. Maintain a dependency inventory and periodically re-audit referenced packages for ownership or content changes.

T08 · Insecure Dependencies

Warning
Location
references/master-series-index.md:92
Finding

Batch Installation of Unreviewed Third-Party Skills

Content
View full analysis

Vulnerability Details

File Location: references/master-series-index.md:92-105
Related Individual Commands: references/master-series-index.md:5,17,30,39,48,61,72,83
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

Vulnerable Code

bash
# Install engineering cost master
openclawmp install skill/4a3a9beaa9dc4a399d56b5a3a5063016

# Install the complete engineering project management set
openclawmp install skill/b42772e4cb38496096da58226657477c
openclawmp install skill/5f90e4007aec4faaa0728706a36219b6
openclawmp install skill/e0a2d954385b48828d867affac429af6

# Install international engineering master
openclawmp install skill/82851afe1e50453b82bad3af859e5a48

The comments above are translated into English for report-language compliance; the commands and asset identifiers are reproduced unchanged.

Technical Analysis

The reference index recommends installing several externally hosted Skills, including a batch of three project-management Skills. None of the remote packages is present in the audited project, and the index supplies no immutable version, integrity digest, signature-verification procedure, permission declaration, or pre-activation review process.

Batch installation increases the supply-chain attack surface because several independent remote assets and publisher accounts must all remain trustworthy. It also encourages users to install multiple packages without separately assessing each package's instructions, scripts, dependencies, and requested capabilities.

No evidence establishes that the listed assets are malicious. The confirmed issue is the unsafe dependency-consumption pattern and the absence of controls needed to verify externally retrieved Skill content.

Attack Path

  1. An attacker gains control of any listed asset, its publisher account, the external registry, or the dependency-resolution infrastructure.
  2. The attacker publishes modif ...[truncated 1120 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove or discourage batch installation until every listed package has been independently reviewed.
  2. Pin each Skill to a specific immutable release and publish its expected cryptographic hash.
  3. Verify registry and publisher signatures before installation.
  4. Provide a trusted source repository and reviewed commit identifier for every asset.
  5. Add a per-package security summary covering scripts, dependencies, permissions, network behavior, and sensitive-data access.
  6. Introduce a staged workflow: download, verify signature and hash, inspect content, sandbox-test, and only then activate.
  7. Use least-privilege sandboxes and deny network, credential, process, and broad filesystem access by default.
  8. Continuously monitor the referenced assets for publisher changes, new versions, altered manifests, or ownership transfers.
  9. If reproducible verification cannot be provided, vendor the reviewed Skill content into the project instead of retrieving it from a mutable external source.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (7)

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The skill description and all operational instructions are presented exclusively in Chinese, with no indication that users may choose another language or that the skill is intentionally limited to a Chinese-language/regional audience. Under the policy, forcing a specific language without opt-in can be a natural-language policy violation unless clearly justified.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger phrases include broad natural-language patterns such as wanting to improve a domain capability, which can match ordinary conversation rather than explicit invocation. This can cause unintended activation of the skill's expert persona/workflow, leading to prompt-routing errors, overbroad influence on responses, or accidental use in contexts the user did not intend.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Short generic trigger terms like contract review, bidding, procurement, mediation, or dispute are common topic words and not reliable activation boundaries. If the host system uses these as routing or auto-invocation cues, normal user discussion may accidentally invoke specialized behavior or tools, creating context confusion and unsafe overreach.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Multiple sections define activation through broad domain keywords across contracts, cost control, procurement, mediation, and estimating. In aggregate, this greatly widens the accidental activation surface and increases the chance that ordinary engineering/legal discussion triggers specialized outputs or workflow assumptions without user consent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The title and the entire document are presented exclusively in Chinese, with no indication that users may choose another language or locale. Under the policy, forcing a specific language without user opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file contains user-facing instructional content only in Chinese, and there is no indication that the skill is region-specific or that users may opt into another language. Under the language/locale policy, forcing a specific language without user choice can be a natural-language policy violation.

Content

No source excerpt is available for this finding.