Back to skill

Security audit

Doc Collaboration Watcher

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches a document-change notifier, but it exposes a real-looking ClawHub token and enables broad automatic monitoring and notification behavior that needs review.

Review before installing. The publisher should revoke and remove the exposed ClawHub token, republish a cleaned version, pin dependencies, run the watcher without root, replace broad process-kill commands, and require explicit channel selection or confirmation before any document metadata or local paths are sent outside the host.

Vulnerability Patterns
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (4)

T09 · Insecure Skill Coding Practices

Error
Location
PUBLISH.md:43
Finding

Plaintext ClawHub Access Token Committed to Documentation

Content
View full analysis
" ``` The repository contains a complete token value beginning with the `clh_` prefix. The credential value is redacted from this report to avoid further disclosure. ### Technical Analysis A ClawHub access token is embedded directly in a tracked documentation file. Because `PUBLISH.md` is distributed with the Skill and may also exist in repository history, anyone who downloads or clones the project can recover the credential without requiring local system access. Environment variables are an appropriate runtime mechanism for supplying credentials, but documentation must use placeholders rather than real values. Committing the token defeats the confidentiality provided by the environment variable. Even if the token is later removed from the current file, it may remain accessible through Git history, package archives, mirrors, caches, forks, or previously published Skill versions. ### Attack Path 1. An attacker downloads the Skill package or clones the public repository. 2. The attacker opens `PUBLISH.md` or searches the repository for token prefixes such as `clh_`. 3. The attacker extracts the plaintext `CLAWHUB_TOKEN`. 4. The attacker submits the token to the ClawHub CLI or API. 5. If the token remains valid, the attacker performs any operations allowed by its assigned scope, potentially including unauthorized package publication or modification. 6. A malicious Skill version could then be distributed to downstream users as a supply-chain attack. ### Impact Assessment If the token is valid, the attacker may obtain the ClawHub account or publication privileges associated with it. The exact privileges cannot be determined from the repository alone and depend on server-side token scope. Potential ...[truncated 447 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
bin/doc-watcher.py:224
Finding

Busy-Wait Loop Causes Persistent CPU Resource Exhaustion

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
skill.yaml:49
Finding

Broad Process Matching Can Terminate Unrelated Processes

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
skill.yaml:13
Finding

Third-Party Dependency Is Not Reproducibly Pinned or Integrity-Verified

Content
View full analysis
=6.0.0 ``` A related installation instruction appears in `SKILL.md`, lines 53-55: ```bash # 2. Install dependency pip3 install watchdog ``` ### Technical Analysis The version constraint `watchdog>=6.0.0` permits any future release satisfying the lower bound. The manual installation command is even broader because it installs whichever version the package index currently resolves as latest. The project includes no audited lockfile, exact version pin, package hash, or explicit trusted-index policy. Consequently, two installations performed at different times may execute different dependency code. There is no evidence that `watchdog` is malicious, misspelled, or retrieved from a suspicious source. The finding is an integrity and reproducibility weakness rather than evidence of an active dependency attack. ### Attack Path 1. A user installs the Skill and follows the documented dependency installation process. 2. The package manager resolves `watchdog` to the latest eligible release. 3. If the upstream package, publisher account, package index, or dependency chain has been compromised, the altered package is downloaded. 4. Python imports and executes the installed dependency when `doc-watcher.py` starts. 5. Malicious dependency code would run with the same filesystem and account permissions as the Skill. This exploitation path depends on compromise or unexpected behavior in a future dependency release; the repository itself does not contain such a payload. ### Impact Assessment Potential impact is limited by the privileges of the account running the Skill but may include: - Reading files accessible to the OpenClaw user. - Modifying workspace data and logs. - Accessing environment var ...[truncated 307 chars]
Remediation
View remediation
\ --hash=sha256: ``` 3. Install dependencies with hash enforcement: ```bash python3 -m pip install --require-hashes -r requirements.txt ``` 4. Use a dedicated virtual environment with minimal permissions. 5. Configure an approved package index explicitly and require TLS certificate validation. 6. Periodically review and update the pinned version through a controlled dependency-update process. 7. Add automated vulnerability and provenance scanning for dependencies. 8. Keep the manifest and installation documentation synchronized so both enforce the same reviewed version. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (17)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README promotes OpenClaw memory integration as enabled by default and says change events are stored in SQLite, but it does not clearly warn users that document names, timestamps, actors, and change metadata may be retained locally by default. Even without full document contents, this metadata can reveal sensitive project activity, internal structure, and timelines.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README states the skill will automatically use all enabled channels from the user's OpenClaw configuration, but it does not prominently warn that document-change metadata will be sent outbound over every configured messaging channel. In a collaboration-monitoring skill, this can cause unintended data disclosure to third-party services or recipients if users enable the skill expecting local-only behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill explicitly describes sending document change notifications to external messaging channels such as Feishu, WeChat, iMessage, and WebChat, and its sample notification includes a local file URI. Without a clear user-facing warning and consent boundary, this creates a real data exposure risk because document names, paths, timestamps, and potentially sensitive collaboration context may be transmitted off-host automatically.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
84% confidence
Finding

The documentation instructs users to start the service with 'sudo systemctl', normalizing elevated execution for a file-watching and message-sending skill that does not appear to require root privileges. Running such a component as root increases the blast radius of any bug, dependency compromise, or future code injection, especially because it processes filesystem events and communicates externally.

Content

Scanner excerpt · SKILL.md (reported line 102)May include surrounding context.

python3 bin/doc-watcher.py &

或使用 systemd(Linux)

sudo systemctl start doc-watcher

text

### 查看变更历史

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

This code file contains natural-language instructions and runtime messages that force a specific language for all users. Under the policy, language constraints should either offer user choice or be clearly justified as locale-specific, which is not present here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

该 Markdown 文件的标题与全文说明均使用中文呈现,但没有提供其他语言选项、用户选择机制,或说明这是面向特定中文地区/组织的限定文档。根据语言/locale 政策,这构成了未获用户选择即强制特定语言的自然语言策略问题。

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The description advertises broad real-time monitoring and automatic notification behavior without clearly constraining what events trigger it, who is considered 'relevant,' or what approval gates exist. In an automation skill with file monitoring, message sending, and Python execution privileges, vague activation language increases the risk of over-collection, unintended triggering, and surprise data disclosure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill is explicitly designed to automatically notify related sub-agents and communication channels, yet the manifest does not present a user-facing warning or consent flow about outbound messaging. Because the skill also reads and writes files, this creates a realistic risk that document contents or metadata are propagated to recipients without the user's informed awareness.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
82% confidence
Finding

The skill requests a combination of persistent file access, message-sending, and Python execution for a background watcher that stores events in OpenClaw memory. In this context, those capabilities can facilitate session persistence or long-lived monitoring behavior beyond a single user action, increasing the blast radius of misconfiguration or abuse.

Content

Scanner excerpt · skill.yaml (reported line 20)May include surrounding context.

yaml
# 权限
permissions:
  - file:read
  - file:write
  - message:send
  - exec:python

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Automatic channel discovery combined with auto-notification enables the skill to select recipients from platform configuration and send document-related events without an explicit recipient allowlist. This materially increases the chance of sensitive collaboration data being shared to unintended channels, especially in multi-agent or multi-tenant environments.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file is entirely in Chinese, including all contribution instructions, and does not provide an alternative language option or state that the project is intentionally Chinese-only. The policy requires flagging language or locale constraints when they are imposed without user opt-in or clear justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file uses Chinese throughout for the skill's publishing and usage information, but does not state that the language is optional, user-selectable, or region-specific. Under the policy rule for language/locale constraints, forcing a single language without opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill documentation presents all usage guidance, installation steps, and operational details only in Chinese. Under the stated policy, forcing a specific language without user opt-in or justification is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

The surrounding documentation frames the feature as zero-config channel discovery for a document watcher, but the inline code example shows direct access to ~/.openclaw/config/openclaw.json. This is not necessarily unsafe by itself, but it contradicts the narrower impression that the skill only watches docs and stores minimal change metadata.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The natural-language instructions, headings, workflow descriptions, and examples are all in Chinese, with no indication that the user can select another language or that the skill is intended only for a Chinese-language environment. Under the stated policy, forcing a specific language without opt-in is a locale-policy issue.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The document states it applies to all OpenClaw users 'regardless of what communication tool' they use and later says it 'automatically adapts to all supported communication tools.' However, the concrete supported channels listed earlier are a finite set, and later regional examples mention LINE and KakaoTalk, which are not present in the supported-channel tables or configuration examples. This is an intent/documentation contradiction rather than mere incompleteness.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The manifest presents the primary skill description in Chinese only, with no indication that users can choose another language or that the skill is intentionally limited to a Chinese-speaking audience. This may violate language/locale policy where skills should not force a specific language without opt-in or justification.

Content

No source excerpt is available for this finding.