Back to skill
Skillv1.2.0
ClawScan security
Memegen · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
BenignApr 13, 2026, 5:35 AM
- Verdict
- Benign
- Confidence
- high
- Model
- gpt-5-mini
- Summary
- The skill's requirements and instructions are coherent with a meme-generation utility; it requires no credentials and no install, but it does ask agents to include its guidance in system prompts and to optionally run helper scripts — review those choices before giving the skill full control.
- Guidance
- This skill behaves like a normal memegen helper: it builds memegen.link URLs, downloads images, and optionally runs included scripts (deep-fry, Imgflip/Reddit/Giphy fetchers). It requests no credentials by default. Two things to consider before installing: (1) SKILL.md suggests copying its entire content into your model's system prompt — that will change the assistant's behavior, so don't paste it into privileged system prompts unless you trust the content. Instead, extract just the parts you need (URL rules, template taxonomy). (2) The optional helper scripts fetch external data and may require API keys (Reddit/Giphy) or pip installing Pillow for deep-frying; run those scripts only in a sandbox or review their code. Otherwise the skill is internally consistent and proportionate for its stated purpose.
- Findings
[system-prompt-override] expected: SKILL.md explicitly instructs integrators to include its content in the LLM system prompt to improve template selection. This behaviour is functionally expected for skills that provide a knowledge file, but it increases the skill's ability to influence agent decisions and should be reviewed before blindly pasting into privileged prompts.
Review Dimensions
- Purpose & Capability
- okName/description, examples, and included scripts all align with a memegen skill: building memegen.link URLs, optional Imgflip/Reddit/Giphy trending fetchers, and a Pillow-based deep-fry post-processor. No unrelated credentials or binaries are requested.
- Instruction Scope
- noteRuntime instructions include examples that tell an agent to download external images, run curl/subprocess/execSync, and (explicitly) to include SKILL.md content in the model/system prompt. These are expected for improving template selection but giving the skill content direct control of the system prompt increases its influence over agent behaviour — review before embedding verbatim.
- Install Mechanism
- okNo install spec is declared (instruction-only). Scripts are included in the package but nothing will be auto-downloaded or executed at install time.
- Credentials
- okNo required environment variables or credentials are declared. Optional features (Reddit/Giphy trending fetchers) need their own API keys as documented, which is reasonable and scoped to those optional features.
- Persistence & Privilege
- notealways:false and user-invocable:true (normal). The guidance suggests agents may maintain a local meme-history.json — that implies writing to the agent workspace but is limited in scope. The main privilege concern is the recommendation to copy SKILL.md into system prompts (see instruction_scope).