Security audit
Security checks for vulnerabilities and agentic risk
AstroClaw is a playful horoscope skill that makes disclosed requests to its own service and has optional, user-directed persistence for daily delivery.
Install only if you are comfortable with the agent contacting astroclaw.xyz for horoscope text. Keep activation to explicit horoscope or astrology requests, and approve any memory, AGENTS.md/CLAUDE.md, state-file, symlink, or cron setup deliberately rather than letting it happen automatically.
The skill explicitly instructs the agent to make unauthenticated network requests to external endpoints, but it does not declare any tool scope or allowed network permissions in the manifest. That mismatch weakens policy enforcement and review, making it easier for the skill to trigger outbound requests in environments where operators expect capabilities to be explicitly declared.
The description includes broad activation phrases such as 'creative rut' and 'entropy injection' that could match many normal conversations unrelated to horoscopes. Overbroad triggering can cause the skill to activate unexpectedly and inject untrusted external content into unrelated workflows, increasing prompt-injection and data-handling risk even though the skill says the horoscope is for fun.
The eval prompts double as activation examples and include very broad trigger phrases such as horoscope, zodiac signs, cosmic alignment, entropy injection, creative rut, and even generic routine setup. This can cause the skill to activate in loosely related conversations, increasing unintended network calls and opportunities for prompt-surface abuse or user confusion about when the skill should run.
No suspicious patterns detected.
