The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a Python script (scripts/collect.py) that wraps the longbridge CLI tool, executing various financial data gathering commands via subprocess.run. The agent is also instructed to perform shell operations for data redirection and processing. These operations are core to the skill's analytical function and target the vendor's own binary.
[INDIRECT_PROMPT_INJECTION]: The workflow involves processing untrusted external content, such as financial news and earnings call transcripts. This ingestion of external data for report generation presents a surface for indirect prompt injection, though the impact is limited by the agent's task-specific instructions to ground analysis in quantitative financial metrics provided by the CLI.
[SAFE]: Analysis of the Python script and skill instructions shows that command execution is limited to the author's official CLI tools. The script uses the standard library to aggregate data into temporary directories. No evidence of obfuscation, unauthorized network exfiltration, or persistence mechanisms was found. The automated RCE alert for scripts/collect.py appears to be a false positive triggered by the script's internal function names and its legitimate use of the subprocess module to call the vendor's local CLI tool.

longbridge-earnings