Back to skill

Security audit

Rumour Buster

Security checks for vulnerabilities and agentic risk

Overview

This looks like a legitimate fact-checking search skill, but it embeds a Tavily API key and can store user API keys in plaintext, so it needs review before installation.

Install only if you are comfortable sending fact-checking queries and URLs to external search providers. Do not paste sensitive API keys, private messages, internal URLs, or confidential claims into this skill until the hardcoded Tavily key is removed, the dependency is pinned and verified, and credential storage is changed to a safer secret mechanism.

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 (3)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/tavily_search.py:11
Finding

Hardcoded Tavily API Credential in Source Code

Content
View full analysis

Vulnerability Details

File Location: scripts/tavily_search.py:11
Vulnerability Type: Hardcoded secret
Risk Level: High

Vulnerable Code

python
TAVILY_API_KEY = os.getenv("TAVILY_API_KEY", "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")

Technical Analysis

The Tavily client contains a live-looking API credential as the default value used when the TAVILY_API_KEY environment variable is absent. Embedding credentials in distributable source code exposes them to every user who downloads, installs, audits, or otherwise obtains the package.

The fallback also means the script silently authenticates with the embedded credential when the environment has not been configured. This behavior is independent of the API key collected by the setup script.

Attack Path

  1. An attacker downloads or otherwise obtains the Skill package.
  2. The attacker reads scripts/tavily_search.py.
  3. The attacker extracts the hardcoded Tavily API key from line 11.
  4. The attacker submits authenticated requests directly to the Tavily API using the exposed key.
  5. Requests consume the credential owner's quota and are attributed to the associated account until the key is revoked.

Impact Assessment

The exposed credential may allow unauthorized use of the Tavily account's API privileges. The practical scope is limited to the permissions assigned to that key, but may include quota consumption, service abuse, unexpected billing, account attribution, and denial of service to legitimate users through quota exhaustion.

Remediation
View remediation

Remediation Suggestions

  1. Revoke and rotate the exposed credential immediately.
  2. Remove the credential from the current source tree and repository history.
  3. Require the key through an environment variable or an operating-system-backed secret store.
  4. Fail closed with a clear configuration error if no credential is available; do not provide a secret fallback.
  5. Add secret scanning to pre-commit hooks and CI pipelines.
  6. Ensure tests use mocked credentials or explicitly invalid placeholders.
  7. Integrate the search script with the setup configuration securely rather than maintaining separate, inconsistent credential sources.

T09 · Insecure Skill Coding Practices

Warning
Location
sub-skills/setup/setup.py:99
Finding

Tavily API Key Stored in an Unprotected Plaintext Configuration File

Content
View full analysis

Vulnerability Details

File Location: sub-skills/setup/setup.py:99-115
Vulnerability Type: Plaintext sensitive-data storage with insufficient file protections
Risk Level: Medium

Vulnerable Code

python
"english": {
    "tavily": {
        "available": tavily.get("configured", False),
        "api_key": tavily.get("api_key", ""),
        "free_quota": tavily.get("free_quota", 0)
    },
    "multi_search_engine": {
        "available": multi_search["available"],
        "engines": multi_search.get("english_engines", [])
    }
}
python
with open(CONFIG_FILE, 'w', encoding='utf-8') as f:
    json.dump(config, f, ensure_ascii=False, indent=2)

The destination is defined as:

python
CONFIG_FILE = os.path.expanduser("~/.rumour-buster-config")

Technical Analysis

The setup process inserts the user's Tavily API key directly into a JSON configuration object and writes it in plaintext to ~/.rumour-buster-config.

The code does not explicitly create the file with owner-only permissions such as 0600. Consequently, confidentiality depends on the process umask and any permissions already associated with the file. It also does not verify file ownership, reject symbolic links, perform an atomic safe-file replacement, or use an operating-system secret store.

If the destination already exists as a symbolic link, the ordinary open(..., 'w') operation follows it. A local attacker able to manipulate the user's home-directory entry could therefore redirect the write to another writable target or arrange to capture the generated configuration.

Attack Path

Local disclosure path:

  1. The user runs the setup process and enters a valid Tavily API key.
  2. The setup process stores the key in the configuration object.
  3. The configuration is written in plaintext to ~/.rumour-buster-config.
  4. Another local user or compromised process with read access to th ...[truncated 1029 chars]
Remediation
View remediation

Remediation Suggestions

  1. Prefer an operating-system-backed secret store instead of storing API keys in the JSON configuration.
  2. If file storage is unavoidable, separate secrets from non-sensitive configuration.
  3. Create the secret file atomically with owner-only mode 0600, such as by using os.open with O_CREAT, O_EXCL, and an explicit mode.
  4. Reject symbolic links and verify that the destination is a regular file owned by the current user.
  5. Write through a securely created temporary file in the same directory, apply restrictive permissions, and atomically replace the destination.
  6. Correct permissions on existing configuration files during migration.
  7. Avoid printing, logging, or returning complete API keys.
  8. Document how users can rotate and remove stored credentials.

T08 · Insecure Dependencies

Warning
Location
sub-skills/setup/setup.py:173
Finding

Unpinned External Skill Dependency Installed from a Mutable Package Name

Content
View full analysis

Vulnerability Details

File Location: sub-skills/setup/setup.py:173-177
Additional Locations: SKILL.md:186-188; sub-skills/setup/SKILL.md:35-38
Vulnerability Type: Unpinned third-party executable dependency
Risk Level: Medium

Vulnerable Code

The executable setup script recommends installation by mutable package name:

python
print("请执行以下命令安装:")
print("  clawhub install api-multi-search-engine\n")
print("或访问: https://clawhub.com 搜索 'multi-search-engine'\n")
print("安装完成后重新运行 setup。")

The Skill documentation gives the same unpinned command:

text
clawhub install api-multi-search-engine

Dependency detection only checks whether a path exists:

python
skill_path = os.path.expanduser("~/.openclaw/workspace/skills/api-multi-search-engine")
if os.path.exists(skill_path):
    return {
        "available": True,
        "chinese_engines": ["sogou", "sogou_wechat", "toutiao"],
        "english_engines": ["duckduckgo", "startpage", "wolframalpha"]
    }

Technical Analysis

The project treats api-multi-search-engine as a required external Skill and directs users to install it without pinning a reviewed version, immutable digest, or verified publisher identity. A mutable package name can resolve to code that differs from the version originally reviewed.

The local dependency check establishes only that the expected directory exists. It does not verify a package signature, checksum, manifest, version, ownership, or expected files. A malicious or locally substituted directory can therefore satisfy the availability check.

This is a supply-chain weakness rather than evidence that the current dependency is malicious.

Attack Path

Registry compromise or malicious update path:

  1. An attacker compromises the package publisher, registry entry, or distribution channel for api-multi-search-engine.
  2. The attacker publishes a malicious release un ...[truncated 1209 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin the dependency to a reviewed release and immutable content digest.
  2. Verify the package publisher and cryptographic signature before installation.
  3. Maintain an allowlisted manifest containing the expected package identifier, version, digest, and trusted source.
  4. Validate the installed dependency's manifest and content hash instead of checking only whether its directory exists.
  5. Fail closed when integrity verification cannot be completed.
  6. Review dependency updates before changing the pinned version.
  7. Run third-party Skills with minimal filesystem, environment, network, and tool permissions.
  8. Document the exact trusted dependency source and update procedure.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (21)

Tainted flow: 'payload' from os.getenv (line 25, credential/environment) → requests.post (network output)

Critical
Category
Data Flow
Confidence
99% confidence
Finding

The file hardcodes a Tavily API key as the default value for TAVILY_API_KEY and then transmits it to an external service in the request body. This is a real secret-exposure issue: anyone with access to the code can reuse the credential, incur costs, or abuse the linked Tavily account, and the secret is also unnecessarily embedded in distributable source.

Content

Scanner excerpt · scripts/tavily_search.py (reported line 36)May include surrounding context.

python
}
    
    try:
        response = requests.post(url, json=payload, timeout=30)
        response.raise_for_status()
        return response.json()
    except requests.exceptions.RequestException as e:

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill requests an API key without any upfront warning that it is sensitive, may be retained in chat/session history, or may later be stored locally. This increases the likelihood of inadvertent credential disclosure and is especially dangerous in a conversational agent context where users may assume secure handling that is not actually provided.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The configuration example shows the full Tavily API key saved in a local config file, but the user-facing flow does not clearly warn that the credential will be persisted. Silent or under-disclosed persistence of secrets is dangerous because it expands exposure duration and attack surface beyond the immediate setup session.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README describes submitting claims and URLs for verification through multiple search providers, but it does not clearly warn users that their inputs may be transmitted to several third-party services. In a fact-checking skill, users may paste sensitive messages, private links, or unpublished content, so lack of disclosure creates a real privacy and data-handling risk rather than a purely documentation issue.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding

The skill describes capabilities that involve environment access, local file writes, and network use, but it does not declare any explicit tool scope or permissions boundary. That creates an authorization ambiguity where an agent may invoke broader capabilities than a user expects, especially during the automatic setup flow and external search operations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill sends user-provided queries and URLs to multiple external search providers, but the description does not clearly warn users that their inputs will be transmitted to third parties. This can expose sensitive personal, corporate, or investigative data to external services without informed consent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The manifest description is entirely in Chinese and the documented invocation examples use only Chinese commands such as '/验证' and '重新设置'. There is no indication that users may choose another language or that the Chinese-only interface is a justified region-specific constraint.

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 automatic first-run setup that generates a local config file and may handle Tavily API credentials, but it does not clearly warn the user before writing files or storing secrets. Silent or poorly disclosed credential handling increases the risk of accidental secret exposure, insecure local storage, and user surprise about persistent changes.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module description states it performs deep English search, and the usage/help text also requires 'English search keywords'. This imposes a language constraint in natural language without giving the user an option to choose another language or explaining a policy-justified reason for the restriction.

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 · scripts/tavily_search.py (reported line 23)May include surrounding context.

python
include_answer: 是否包含 AI 生成的答案总结
        max_results: 返回结果数量 (1-10)
    """
    url = "https://api.tavily.com/search"
    
    payload = {
        "api_key": TAVILY_API_KEY,

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This code transmits the user's search query to Tavily via an HTTP POST request, which is a network operation involving user-provided data. Although the script prints the query locally, it does not clearly disclose that the query and request payload are being sent to a third-party service.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
80% 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 · scripts/tavily_search.py (reported line 36)May include surrounding context.

python
}
    
    try:
        response = requests.post(url, json=payload, timeout=30)
        response.raise_for_status()
        return response.json()
    except requests.exceptions.RequestException as e:

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This markdown file defines invocation phrases, so vague-trigger review applies. The single-word trigger setup is highly generic and could match many unrelated contexts, with no narrowing constraints or exclusion examples to indicate when this specific sub-skill should or should not activate.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
91% confidence
Finding

This duplicated finding points to the same behavior: direct chat entry of an API key exposes the credential to session retention mechanisms. The risk is real because conversational systems commonly persist prompts and responses for history, quality review, or debugging.

Content

Scanner excerpt · sub-skills/setup/SKILL.md (reported line 91)May include surrounding context.

md
1️⃣ 访问 https://tavily.com/
2️⃣ 点击 "Sign Up" 注册账号
3️⃣ 进入 Dashboard → API Keys
4️⃣ 点击 "Create New Key"
5️⃣ 复制生成的 Key(格式:tvly-xxxxxxxx)

申请完成后,请直接粘贴你的 API Key:

Session Persistence

Medium
Category
Rogue Agent
Confidence
91% confidence
Finding

This duplicated finding points to the same behavior: direct chat entry of an API key exposes the credential to session retention mechanisms. The risk is real because conversational systems commonly persist prompts and responses for history, quality review, or debugging.

Content

Scanner excerpt · sub-skills/setup/SKILL.md (reported line 91)May include surrounding context.

md
1️⃣ 访问 https://tavily.com/
2️⃣ 点击 "Sign Up" 注册账号
3️⃣ 进入 Dashboard → API Keys
4️⃣ 点击 "Create New Key"
5️⃣ 复制生成的 Key(格式:tvly-xxxxxxxx)

申请完成后,请直接粘贴你的 API Key:

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The setup documentation explicitly stores the Tavily API key in a plaintext file under the user's home directory. This creates a durable credential exposure risk to other local users, malware, backups, logs, or accidental disclosure, and the credential-handling behavior is not safely designed or adequately justified in the documented flow.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script's prompts, instructions, and status messages are presented in Chinese, which imposes a specific language on all users. The policy allows locale constraints only when clearly justified or when users are offered a language choice, neither of which is present here.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script asks the user to paste an API key and then silently saves it to disk without clearly informing them that the credential will be persisted. This creates a real risk of unintended secret exposure because users may assume the key is used only for the current session, while the saved file can later be read by other software or included in backups.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The setup script persists a user-supplied Tavily API key into a plaintext config file under the user's home directory. Storing credentials locally is security-relevant behavior that expands the skill's trust boundary beyond simple search/verification and can expose the key to other local users, backups, logs, or later accidental disclosure if file permissions are weak.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

All user-facing instructions and examples in the README are presented only in Chinese, and the document does not indicate that users can choose another language or that the skill is intentionally restricted to a Chinese-language audience for a documented regional reason. This creates a natural-language locale policy concern under the rule for forced language without user opt-in.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill asks the user to paste the full API key directly into chat, which contradicts any expectation of protected secret entry and exposes the key to chat history, transcripts, model context, and operational logging. Even if the intent is setup convenience, collecting secrets in ordinary conversation is unsafe handling of credentials.

Content

No source excerpt is available for this finding.