Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
OpenClaw DLP Guard
v2.0.0Automatically detects and blocks prompt injection attempts during AI content submission to social media, APIs, web forms, and file outputs.
⭐ 0· 56·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
The skill claims to detect/block prompt injection across social media, APIs, web forms and file writes and its SKILL.md and plugin manifest contain many detection patterns and triggers that match that purpose. Minor inconsistency: README and SKILL.md reference an install command (clawhub install prompt-guard) and a configuration path (~/.openclaw/.../prompt-guard-config.json) even though the package is instruction-only and the registry metadata declares no required config paths or install spec. This is not necessarily malicious but should be clarified (how is configuration persisted and who/what writes that file?).
Instruction Scope
Runtime instructions list concrete triggers (pre_submit, pre_post, pre_send) and pattern-based checks; they do not instruct the agent to read arbitrary unrelated system files, exfiltrate data, call external endpoints, or request secrets. The SKILL.md contains many injection phrases and regex examples — these appear as detection patterns (expected), not as active commands to exfiltrate data.
Install Mechanism
No install spec or code files are provided (instruction-only), which is low-risk. However the README advertises a CLI install command and CLI commands (/guardian ...) even though no binary or install metadata is present in the package. The absence of an actual install artifact means the skill will only be guidance/instructions for the agent rather than executable code installed on disk.
Credentials
The skill requests no environment variables, no credentials, and no config paths in the registry metadata. The included detection lists identify many types of secrets and PII (OpenAI, AWS, SSH keys, SSN etc.) for detection purposes only; there is no unexplained request for access to those secrets.
Persistence & Privilege
The skill is not declared always:true and is user-invocable. It does reference persisting configuration under the user's workspace, which is reasonable for a guard plugin but should be validated at install/runtime. It does not request system-wide privileges or modification of other skills.
Scan Findings in Context
[ignore-previous-instructions] expected: The phrase is part of the detection pattern list (the guard must look for 'ignore previous instructions'); its presence in SKILL.md is expected and not evidence of malicious behavior.
[you-are-now] expected: This is part of the 'role manipulation' detection patterns (e.g., 'you are now a hacker'). Its presence in the pattern library is expected for a prompt-guard.
Assessment
This skill is internally consistent with being a pattern-based prompt-injection guard and appears to be instruction-only (no binaries or credentials requested). Before installing or enabling it for automated submissions, verify: (1) how configuration is actually stored/managed (the README mentions ~/.openclaw/... though the package provides no installer), (2) how alerts/notifications are delivered (ensure they don't leak content to an external webhook or third party), and (3) whether the agent runtime will actually enforce these checks (test in a safe environment with harmless injection examples). If you need a guard that runs as code, obtain a real installable implementation (or confirm the platform provides the enforcement hooks) rather than relying only on this instruction-only package.README.md:103
Prompt-injection style instruction pattern detected.
SKILL.md:43
Prompt-injection style instruction pattern detected.
About static analysis
These patterns were detected by automated regex scanning. They may be normal for skills that integrate with external APIs. Check the VirusTotal and OpenClaw results above for context-aware analysis.Like a lobster shell, security has layers — review code before you run it.
latestvk97bbrd9f483xvafezjvhxpjgn83m591
License
MIT-0
Free to use, modify, and redistribute. No attribution required.