Back to skill

Security audit

Ai Secretary Bundle

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent AI secretary bundle, but it asks users to connect email accounts, schedule automation, and install unreviewed dependencies without enough safeguards or disclosure.

Install only after reviewing the dependency skills and the cloned install script, preferably using pinned versions or a trusted release. Treat email configuration as sensitive: use app-specific passwords or OAuth where possible, keep config files out of version control, and restrict mailbox permissions. Require confirmation before sending emails or sharing summaries, and review or disable recurring cron jobs that access private data.

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
README.md:9
Finding

Unpinned Third-Party Dependencies and Unverified Repository Script Execution

Content
View full analysis

Vulnerability Details

File Location: README.md:9, README.md:19-24, and SKILL.md:72-82
Vulnerability Type: Unpinned dependencies and execution of unverified remote repository content
Risk Level: Medium

Evidence

README.md:9:

powershell
# Windows
pip install openclaw

README.md:19-24:

powershell
# Clone the bundle repository
git clone https://github.com/clawhub/ai-secretary-bundle.git
cd ai-secretary-bundle

# Install all dependent skills
./install.ps1

SKILL.md:72-82:

powershell
# Install email management
py -m clawhub install email-mail-master

# Install scheduled reminders
py -m clawhub install qqbot-cron

# Install news collection
py -m clawhub install ai-news-collector

# Install task tracking
py -m clawhub install task-tracker

Technical Analysis

The installation instructions retrieve packages and repository content without pinning exact versions, immutable Git commit identifiers, or cryptographic hashes. Consequently, the content installed by these commands can change after the Skill has been audited.

The README also instructs the user to execute install.ps1 immediately after cloning the repository's default branch. That script is absent from the audited artifact, so its contents and security properties could not be reviewed. No signature, checksum, commit validation, or manual inspection step is required before execution.

Although the reviewed files do not contain a malicious payload, this installation pattern creates a supply-chain trust boundary. Compromise of a package registry account, dependency release, repository account, or mutable default branch could cause attacker-controlled code to be installed and executed.

Attack Path

  1. An attacker compromises one of the referenced package publisher accounts, upstream Skill packages, or the Git repository's mutable default branch.
  2. The attacker publishes a malicious p ...[truncated 1175 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin openclaw and every ClawHub Skill to exact, reviewed versions rather than resolving the latest available release.
  2. Clone an immutable, reviewed Git commit or signed release tag instead of executing content from the repository's default branch.
  3. Publish and verify SHA-256 or stronger cryptographic hashes for downloaded packages and scripts.
  4. Require signature verification for package releases and Git tags where supported.
  5. Include install.ps1 in the distributed artifact so it can be reviewed alongside the documentation and configuration.
  6. Display or review the installation script before execution rather than invoking it immediately after cloning.
  7. Use lockfiles or a signed dependency manifest containing exact versions, source locations, and integrity hashes.
  8. Perform installation in a non-administrative, isolated environment with minimal filesystem and network permissions.
  9. Add automated dependency provenance, vulnerability, and integrity checks to the release process.
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 (11)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The README tells users to copy and edit an email configuration file but does not warn that it will likely contain sensitive data such as mailbox usernames, server details, and app passwords. This increases the chance users will store secrets insecurely, commit them to version control, or share them during troubleshooting, leading to credential exposure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The README instructs users to start an AI secretary with access to email, reminders, and news sources without clearly disclosing that the skill will read potentially sensitive communications and connect to external services. Users may enable it without understanding the privacy and network exposure implications, which is risky in an executive assistant context handling high-value data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill advertises unified mailbox access, intelligent summarization, auto-classification, and scheduled sending, but does not clearly warn users that it will process mailbox contents and may send emails on their behalf. In a secretary-style automation context, this creates a real risk of privacy exposure and unintended external communications if users enable the capability without informed consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The configuration example instructs users to provide email passwords and app passwords via environment variables, but gives no warning about the sensitivity of these credentials or safe handling practices. This is dangerous because users may paste secrets into insecure locations, commit configs accidentally, or grant broad mailbox access without understanding the security implications.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The example workflow states that the AI secretary has already sent a meeting agenda to participants, but does not mention user approval, recipient verification, or safeguards for external communications. In a tool positioned for busy professionals and executives, this could lead to accidental disclosure of confidential information or unintended messages being sent to the wrong audience.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The comments and user-facing reminder names/messages in this configuration are written in Chinese, indicating a fixed language choice. Under the policy criteria, forcing a specific language without offering a user choice or documenting a justified region-specific constraint is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The reminder names and messages are natural-language content that appears intended for end users, but they are fixed to Chinese only. Because the file does not provide an opt-in, locale selection, or justification for a Chinese-only deployment, this violates the language/locale policy criterion.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The example shows the assistant reading and summarizing emails, schedules, and tasks, but it does not warn users that these operations involve potentially sensitive personal or business information. This increases the risk of users enabling the skill without understanding the privacy implications, especially in enterprise or executive contexts where mailbox and calendar contents may contain confidential data.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The shortcut triggers are very broad, natural-language phrases like '检查邮件' and '结束今天', which can easily overlap with ordinary conversation. In a voice or chat environment, this can cause unintended invocation of the skill and accidental access to or display of sensitive email, calendar, or task data without clear user intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

This file appears to require Chinese for user interaction, including example commands such as the command quick reference, but it does not indicate that the skill is intentionally region-specific or provide an opt-in language choice. That can violate the language/locale policy when a specific language is effectively forced without user choice or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The natural-language instructions and examples are presented entirely in Chinese, which can amount to a language policy constraint if users are not given an option to use another language. There is no explicit statement that the skill is China-region-specific or that language selection is configurable.

Content

No source excerpt is available for this finding.