posttrainer

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill explicitly instructs the agent to collect user API tokens/keys (HF, RunPod, WandB), to "walk them through" using them, to "ask them to confirm their token works," and to "write complete commands they can copy‑paste" and submit jobs via the RunPod API/CLI — all of which create situations where the LLM would need to embed secret values verbatim into generated commands or requests.

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

• Potentially malicious external URL detected (high risk: 0.90). The RunPod deployment pulls and runs the Docker image ghcr.io/chrisvoncsefalvay/posttrainer:latest at runtime, which executes remote code (the posttrainer worker) and is relied on by the skill to perform training.