Back to skill

Security audit

ClawRiver 经验之河

Security checks for vulnerabilities and agentic risk

Overview

ClawRiver is a coherent public experience-sharing skill, but it embeds a shared API key and exposes public upload/update actions that need review before use.

Install only if you are comfortable using a public third-party ClawRiver service. Do not use the embedded demo key for real work; configure your own API key or leave it empty for read-only access if supported. Review any content before upload and remove API keys, tokens, internal hostnames, private logs, customer data, and confidential configuration values.

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 (2)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:30
Finding

Shared Hardcoded API Credential in Published Configuration

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:30-36 and README.md:24-30
Vulnerability Type: Hardcoded shared API credential
Risk Level: Medium

Vulnerable Code

SKILL.md:30-36:

json
{
  "mcpServers": {
    "clawriver": {
      "url": "https://clawriver.onrender.com/mcp",
      "headers": { "X-API-Key": "sk_test_demo_key_999999" }
    }
  }
}

README.md:24-30:

json
{
  "mcpServers": {
    "clawriver": {
      "url": "https://clawriver.onrender.com/mcp",
      "headers": { "X-API-Key": "sk_test_demo_key_999999" }
    }
  }
}

Technical Analysis

The documentation publishes an API key and instructs every user to copy it into the MCP configuration. Because the credential is public and shared, it cannot provide reliable authentication or attribution. Any party that reads the package can replay it outside the intended client.

This also conflicts with the version 1.0.9 changelog in clawhub.json, which states that the default demo API key was removed. Whether the key is labeled as a test key does not make embedding it safe if the remote service still accepts it.

Exploitability and resulting privileges depend on the server-side permissions assigned to this key. The package describes MCP tools capable of retrieving, uploading, updating, rating, and verifying experiences, so a valid shared key could potentially authorize both read and write actions.

Attack Path

  1. An attacker obtains sk_test_demo_key_999999 from SKILL.md or README.md.
  2. The attacker configures an MCP client with the published key or sends requests directly to https://clawriver.onrender.com/mcp.
  3. The remote service treats requests as belonging to the shared credential.
  4. The attacker invokes any operations granted to that credential, potentially consuming shared quotas or interacting with resources associated with the shared identity.
  5. Audit records cannot reliably ...[truncated 648 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the literal API key from both SKILL.md and README.md.

  2. Replace it with an environment-variable reference:

    json
    {
      "mcpServers": {
        "clawriver": {
          "url": "https://clawriver.onrender.com/mcp",
          "headers": {
            "X-API-Key": "${MEMORY_MARKET_API_KEY}"
          }
        }
      }
    }
    
  3. Revoke or rotate sk_test_demo_key_999999 on the server because publication must be treated as credential compromise.

  4. Give each user an individual, revocable credential with least-privilege scopes.

  5. Default users without a key to explicitly documented read-only access.

  6. Apply server-side rate limits, expiration, scope restrictions, and per-credential audit logging.

  7. Add secret scanning to the release process and verify that package documentation is consistent with security-related changelog claims.

T08 · Insecure Dependencies

Note
Location
README.md:99
Finding

Mutable External Source and Unlocked Dependency Installation

Content
View full analysis

Vulnerability Details

File Location: README.md:99-105 and clawhub.json:102-120
Vulnerability Type: Unpinned source and dependency supply chain
Risk Level: Low

Vulnerable Code

README.md:99-105:

bash
git clone https://github.com/Timluogit/clawriver.git && cd clawriver
pip install -r requirements.txt
python3 -m uvicorn app.main:app --port 8000

clawhub.json:102-120:

json
{
  "type": "git",
  "command": "git clone https://github.com/Timluogit/clawriver.git",
  "description": "Clone from GitHub"
},
{
  "type": "docker",
  "command": "docker-compose up -d",
  "description": "Deploy with Docker"
},
{
  "type": "pip",
  "command": "pip install -r requirements.txt",
  "description": "Install Python dependencies for self-hosting"
}

The package metadata also declares a non-exact dependency constraint:

json
"dependencies": {
  "python": ">=3.11",
  "packages": [
    "httpx>=0.24.0"
  ]
}

Technical Analysis

The self-hosting procedure clones the current default branch of an external repository rather than a reviewed commit or signed release. It then installs dependencies from that mutable checkout. The effective code and dependency graph can therefore change after this skill package has been reviewed.

The declared httpx>=0.24.0 constraint permits future matching versions rather than identifying a reproducible version. The referenced requirements.txt, application source, Docker configuration, and lockfiles are not included in the audited artifact, so their integrity and security could not be verified.

This is a supply-chain hardening issue rather than evidence that the current upstream repository or dependency is malicious.

Attack Path

  1. An attacker compromises the upstream repository, its default branch, a maintainer account, or a dependency release channel.
  2. The attacker modifies application code, `requirem ...[truncated 1076 chars]
Remediation
View remediation

Remediation Suggestions

  1. Direct users to a reviewed, immutable release tag or commit rather than the default branch:

    bash
    git clone --branch v1.0.9 --depth 1 https://github.com/Timluogit/clawriver.git
    cd clawriver
    git verify-tag v1.0.9
    
  2. Publish signed release tags or artifacts and provide cryptographic checksums through an independently protected release channel.

  3. Lock all direct and transitive Python dependencies to reviewed versions.

  4. Install with hash enforcement, such as pip install --require-hashes -r requirements.lock.

  5. Pin container images by digest and avoid floating image tags.

  6. Use automated dependency scanning while requiring review before lockfile updates.

  7. Run installation and the service as a dedicated, unprivileged account in an isolated virtual environment or container.

  8. Include the relevant source, lockfile, and deployment configuration in future audit artifacts so the full execution path can be reviewed.

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 (8)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The README includes an API key directly in the MCP configuration example without any warning that it is a demo credential or guidance on safe secret handling. In an agent-skill context, users may copy this verbatim into configs, normalize unsafe credential practices, or accidentally treat embedded keys as production-safe, increasing the risk of unauthorized use or leakage.

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 · README.md (reported line 74)May include surrounding context.

bash
# 注册
curl -X POST https://clawriver.onrender.com/api/v1/agents \
  -H "Content-Type: application/json" \
  -d '{"name": "MyAgent"}'

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 · SKILL.md (reported line 79)May include surrounding context.

bash
# 注册
curl -X POST https://clawriver.onrender.com/api/v1/agents \
  -H "Content-Type: application/json" \
  -d '{"name": "MyAgent"}'

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The trigger phrases are broad enough to match ordinary troubleshooting or knowledge-seeking requests, which can cause the skill to activate outside a clearly intended context. Because this skill routes users toward a third-party experience-sharing service, overbroad activation increases the chance of unsolicited external data exposure or uploads during normal support conversations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill encourages uploading debugging logs, integration tips, and configuration workarounds, but does not prominently warn against including secrets, personal data, internal system details, or proprietary configuration values. In this context, the omission is dangerous because the platform is public and the advertised content types commonly contain API keys, tokens, stack traces, endpoint URLs, and other sensitive operational data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest exposes remote MCP tools that can search, upload, update, rate, and perform account-affecting actions against an external HTTP service, but it does not present any explicit consent, warning, or action-safety guidance in the tool metadata. In an agent setting, this can lead to unintended data exfiltration, unreviewed sharing of local/project information, or unwanted mutations to a user’s remote account state if the agent invokes these tools automatically.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

Natural-language content in the primary README is presented in a single language, which can functionally force a locale on users who arrive at the default entry point. Because the file does not explicitly offer a language choice up front beyond a link, this may conflict with a policy requiring user language/locale opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The package metadata is primarily in English, but the MCP tool descriptions are presented only in Chinese. This creates a locale/language policy concern because users are not offered a language choice and no region-specific justification is documented.

Content

No source excerpt is available for this finding.