The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions guide the agent to extract sensitive API keys from the local file system at ~/.hermes/config.yaml using shell commands like grep and sed.
[REMOTE_CODE_EXECUTION]: The skill performs remote execution by using SSH to run shell commands and Python scripts on a hardcoded external VPS (IP: 43.159.133.35). This includes generating Python logic locally and executing it on the remote host.
[DATA_EXFILTRATION]: Extracted credentials and processed financial data are transmitted to an external server (43.159.133.35) through SSH command-line arguments and file transfers.
[COMMAND_EXECUTION]: The skill explicitly uses ssh -o StrictHostKeyChecking=no, which disables critical security checks that verify the identity of the remote server, making the connection vulnerable to interception.
[DYNAMIC_EXECUTION]: The skill involves a multi-step execution pattern where it writes Python code to temporary files (e.g., /tmp/decode_vps.py) at runtime and then invokes them via the Python interpreter.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill uses the web_search tool to collect market news, institutional viewpoints, and events from the open web.
Boundary markers: There are no delimiters or instruction-isolation markers used when interpolating this external data into the image generation prompts.
Capability inventory: The skill possesses high-privilege capabilities including remote SSH execution, file writing, and network access.
Sanitization: No validation or filtering is applied to the ingested external content before it is processed by the AI.
[EXTERNAL_DOWNLOADS]: The skill initiates network requests to various external domains including qt.gtimg.cn, finnhub.io, and moltbook.ai to fetch financial data and interact with remote APIs.

daily-investment-report