Back to skill

Security audit

OpenClaw Agent Control Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill appears intended to deploy an agent-control service, but it runs mutable remote code and npm scripts with broad host and network impact that users should review before installing.

Review scripts/deploy_project.sh and the target repository before running it. Prefer a pinned, trusted commit; run it as an unprivileged user outside /root; restrict MONITOR_PORT and PORT to trusted interfaces; and avoid using a custom REPO_URL unless you fully trust that repository and its npm dependencies.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
Findings (3)

T03 · Remote Payload Retrieval and Execution

Error
Location
scripts/deploy_project.sh:4
Finding

Mutable Remote Repository Is Retrieved and Executed Without Verification

Content
View full analysis
/tmp/openclaw-agent-control-backend.log 2>&1 & ``` ### Technical Analysis The deployment script clones or updates the mutable `main` branch of a remote Git repository and immediately starts a Python application from that checkout. It does not pin an audited commit, validate a cryptographic checksum, verify a signed commit or tag, or inspect the retrieved files before importing the remote `app` module. The `REPO_URL` environment variable also allows the caller or execution environment to substitute an arbitrary repository. When Uvicorn loads `app:app`, Python executes module-level code in the remotely retrieved `app` module. Consequently, the effective executable payload can change after this skill package has been reviewed. ### Attack Path 1. An attacker compromises the configured upstream repository or gains the ability to influence `REPO_URL`. 2. The attacker places malicious module-level Python code in the remote application's `app` module or one of its imports. 3. An operator e ...[truncated 1002 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Error
Location
scripts/deploy_project.sh:28
Finding

Unverified npm Lifecycle and Project Scripts Are Executed

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/deploy_project.sh:23
Finding

Backend Service Is Exposed on All Network Interfaces

Content
View full analysis
/tmp/openclaw-agent-control-backend.log 2>&1 & ``` ### Technical Analysis The Uvicorn server binds to `0.0.0.0`, making the backend listen on every available IPv4 network interface. However, the skill documentation presents the validation endpoint as `http://127.0.0.1:8787/api/status`, which can lead operators to believe that the service is restricted to local access. The audited package does not add firewall restrictions, transport encryption, a reverse proxy, or an authentication boundary around the exposed listener. The authentication behavior of the remotely retrieved backend cannot be established from this package, so lack of application authentication is not asserted; nevertheless, the all-interface binding unnecessarily expands the reachable attack surface. ### Attack Path 1. An operator executes the deployment script on a host reachable from another network system. 2. Uvicorn binds the backend port to all network interfaces. 3. A remote client discovers or otherwise learns the host and configured monitor port. 4. The client connects directly to the backend rather than through the documented loopback address. 5. Any endpoint-level weaknesses in the downloaded backend become remotely reachable from networks permitted by host or perimeter firewall rules. ### Impact Assessment The immediate impact is unintended network exposure of the backend and all routes it serves. The reachable scope depends on the host's network topology and firewall configuration and may include a local subnet, container network, cloud virtual network, or the public Internet. The final privileges obtainable depend o ...[truncated 272 chars]
Remediation
View remediation
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The documentation instructs users to execute a deployment shell script directly, but provides no warning about what the script changes, what privileges it may require, or the operational impact on the host. In a skill intended to deploy backend and frontend services, this raises supply-chain and system-modification risk because users may run the script blindly in a privileged environment.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
65% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · scripts/deploy_project.sh (reported line 24)May include surrounding context.

sh
fi

echo "[skill] starting backend"
nohup uv run --with fastapi --with uvicorn \
  python -m uvicorn app:app --app-dir "$PROJECT_DIR" --host 0.0.0.0 --port "$MONITOR_PORT" \
  > /tmp/openclaw-agent-control-backend.log 2>&1 &

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The document is explicitly titled and presented as Chinese-only usage instructions, with no indication that users may choose another language or that the locale restriction is required. 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.