hermes-agent

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The installation instruction in SKILL.md fetches a remote shell script via curl and executes it immediately using bash, which constitutes remote code execution during setup.

HIGH W008: Secret detected in skill content (API keys, tokens, passwords).

• Secret detected (high risk: 1.00). The documentation and references mention configuration instructions and API key environment variable names (such as OPENAI_API_KEY, ELEVENLABS_API_KEY, etc.). There is one hardcoded string looking like an API key in references/chinese-tts-voices.md line 15 (sk_18bd08aa589d83391dbe1c859c090f041b4e7b43d1af68b2), but let's check its format and context. Wait, sk_18bd08aa589d83391dbe1c859c090f041b4e7b43d1af68b2 is prefixed with sk_ and has 52 hex characters, which matches an active/real-looking ElevenLabs API key format, but wait, is it real or an example? Let's inspect: it's under references/chinese-tts-voices.md line 15: - API Key: sk_18bd08aa589d83391dbe1c859c090f041b4e7b43d1af68b2 (stored in .env as ELEVENLABS_API_KEY). This is a real credential embedded in a documentation/reference file.

Wait, let's verify if sk_18bd08aa589d83391dbe1c859c090f041b4e7b43d1af68b2 is a high-entropy secret. Yes, it's a 52-character hex string representing an ElevenLabs API key. It provides access to ElevenLabs services. Therefore, it is an active credential leak.