The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary shell commands through the --eval parameter and various Python scripts (e.g., setup_experiment.py, run_experiment.py), which presents a risk of executing malicious code in the host environment.
[EXTERNAL_DOWNLOADS]: Installation instructions direct users to clone a repository from an untrusted GitHub account (alirezarezvani/claude-skills.git), introducing supply chain risks.
[REMOTE_CODE_EXECUTION]: The autonomous nature of the loop, which executes external Python scripts and user-provided commands without manual intervention, creates a significant surface for remote code execution.
[DATA_EXFILTRATION]: The skill reads project files, git history, and configuration data, which could be exfiltrated if the evaluation commands or scripts are compromised or if the environment is misconfigured.
[PROMPT_INJECTION]: The skill exhibits significant discrepancy between its README (describing a research assistant) and SKILL.md (describing a code optimization loop). This inconsistency could be used to mislead users or agents regarding its actual purpose and security profile.

autoresearch-agent