Prompt Injection Guard
v1.0.0Prompt injection defense. Detect and block malicious prompts, protect system instructions, sanitize user input.
⭐ 3· 6.5k·25 current·25 all-time
by@maorun
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name and description match the SKILL.md: it defines detection patterns, handling levels, and filtering rules for prompt injection. No unrelated env vars, binaries, or installs are requested.
Instruction Scope
Instructions stay within the stated defensive purpose (input boundary markers, pattern scanning, output filtering, context separation). Minor operational notes: it instructs logging of 'input parts' and detection results but does not specify storage/retention or destinations; auto_trigger and keyword matching may produce false positives or over-blocking without tuning.
Install Mechanism
No install spec and no code files — instruction-only skill. This minimizes installation risk because nothing is written or executed by the skill itself.
Credentials
The skill requests no credentials, binaries, or config paths. The only privacy/credential consideration is the logging guidance (records of input and detection events) which could capture sensitive user content if stored — but that is an operational concern, not a mismatch with purpose.
Persistence & Privilege
always:false and user-invocable:true. SKILL.md sets auto_trigger/keyword triggers (appropriate for a guard). Agent autonomous invocation is allowed by default but is not combined with other concerning privileges here.
Assessment
This skill is an instruction-only prompt-injection defense and appears internally consistent. Before installing, consider: 1) where detection logs will be stored and who can read them (the skill suggests logging parts of inputs — could capture sensitive data), 2) tune and review the detection patterns to avoid false positives and to catch obfuscated payloads, and 3) test the guard in a safe environment to ensure it doesn't inadvertently block legitimate tasks. Remember this is a policy the agent must follow; it does not itself enforce rules at system level unless the host agent implements them.Like a lobster shell, security has layers — review code before you run it.
latestvk9787jw7xfv1w60kdq573gj4vh817n4a
License
MIT-0
Free to use, modify, and redistribute. No attribution required.