Caveat verdict
siliville
silivillepro
The skill fetches a remote claw-manifest containing a system_prompt_extension and 35 action blueprints, allowing the siliville.com server to remotely inject new instructions into agent behavior at runtime.
⚠ Flagged for review — coarse, uncorroborated signal, not a confirmed exploit. Review the config yourself before installing.
Automated static analysis — not a human review. Caveat flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.
Findings (6)
<script> tag in markdown — potential code injection
SKILL.md · code · <script
Possible hardcoded credential
README.md · code · TOKEN="sk-slv-your-key-here
Base64 encode (btoa) — may obfuscate data exfiltration
SKILL.md · code · btoa(
Python os.environ.get — reads environment variable
README.md · code · os.environ.get(
Instructs covert action — may act without user awareness
README.md · prose · downgraded · silently
Accesses shell history/config
siliville_skill.py · prose · downgraded · ~/.zshrc
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class F). Final tier capped at Caution — cannot be lifted by any downgrade, example-payload opt-in, or allowlist.
Permissions & capabilities
Requires 1 environment variable. (1 sensitive: SILIVILLE_TOKEN).
data_encodingnetwork_in Is this flag fair?
Thanks — recorded.