Back to skill

Security audit

Doubao Capture

Security checks for vulnerabilities and agentic risk

Overview

The skill describes a normal Doubao subtitle-capture workflow, but its plugin entrypoint points to an external Python file that is not included in the reviewed package.

Review before installing. The skill may be legitimate, but the package should include the Python implementation and use a package-relative entrypoint so users can verify what will run. Avoid installing this version unless you control and have audited the referenced /mnt/h/... script path.

Vulnerability Patterns
  • Tool Hijacking and SpoofingModifies or replaces tools so legitimate-looking calls execute attacker logic
  • 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)

T07 · Tool Hijacking and Spoofing

Error
Location
openclaw.plugin.json:9
Finding
Plugin Entrypoint References an External Unaudited Script## Vulnerability Details **File Location**: `openclaw.plugin.json`, line 9 **Vulnerability Type**: External tool entrypoint replacement **Risk Level**: High ### Vulnerable Code ```json "entry": "/mnt/h/AI/openclaw_workspace/video_translate/skills/capture_doubao_scroll_v2.py" ``` ### Technical Analysis The plugin delegates execution to an absolute filesystem path outside the audited package. The referenced Python implementation is not included among the project files, so its behavior and integrity cannot be verified as part of this package. This configuration also conflicts with the package-relative command documented in `SKILL.md`: ```bash python scripts/capture_doubao_scroll_v2.py --hwnd <window_handle> --output-dir <dir> --stop-auto --json-output ``` Because the external target can be modified independently of the reviewed package, control over that path permits replacement of the implementation while retaining the trusted `capture_translation` tool identity. The effective code executed by the plugin can therefore differ from the code—or behavior—reviewed at installation time. ### Attack Path 1. The package is installed or loaded with `openclaw.plugin.json` as its plugin descriptor. 2. An attacker who can write to `/mnt/h/AI/openclaw_workspace/video_translate/skills/` creates or replaces `capture_doubao_scroll_v2.py`. 3. The agent or user invokes the legitimate-looking `capture_translation` tool. 4. The plugin resolves its entrypoint to the attacker-controlled external file. 5. The substituted Python code executes with the privileges and environmental access of the plugin host. Exploitation requires write access to the referenced external path or another means of controlling its resolution. ### Impact Assessment Successful exploitation permits arbitrary Python execution in the plugin host's security context. The resulting scope may include access to files, environment variables, GUI resource ...[truncated 303 chars]
Remediation
## Remediation Suggestions 1. Include `capture_doubao_scroll_v2.py` within the distributed package and configure a package-relative entrypoint. 2. Canonicalize the resolved entrypoint and reject startup if it escapes the package root. 3. Ensure the packaged implementation and its parent directories are not writable by untrusted users. 4. Record and verify a cryptographic hash or signed manifest for executable plugin files before invocation. 5. Fail closed when the bundled entrypoint is missing rather than falling back to an external path. 6. Align `openclaw.plugin.json` with the package-relative location documented in `SKILL.md`. 7. Audit the referenced Python implementation before deployment, because it was not included in the supplied project.
Vulnerability Patterns
  • 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep

Static analysis

No suspicious patterns detected.