The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The test suite file tests/test_run_post_merge_task.py utilizes the exec() function to execute Python code that is dynamically extracted from the scripts/run-phase.sh shell script. While used for unit testing internal logic, the execution of code strings extracted at runtime is a significant security risk.
[COMMAND_EXECUTION]: The core logic in scripts/run-phase.sh, scripts/seed-board.sh, and scripts/archive-run.sh relies heavily on subprocess.run to call the hermes CLI and other system utilities. These calls often involve dynamically generated prompts and arguments, creating a large surface for potential command injection if input data is not perfectly sanitized.
[EXTERNAL_DOWNLOADS]: The documentation (README.md and SKILL.md) instructs users to download the skill repository directly from a third-party GitHub repository (https://github.com/magnus919/hermes-SkillOpt). This represents a supply-chain risk as the repository is not from a verified or well-known trusted organization.
[PROMPT_INJECTION]: The skill is designed to ingest and execute external task instructions (via test-suite.json) during the 'Rollout' and 'Validate' phases. This creates a significant surface for Indirect Prompt Injection, where malicious instructions in a test suite could cause the agent to perform unauthorized actions during the optimization process.
[COMMAND_EXECUTION]: The skill uses shell-to-python heredocs (python3 << 'PYEOF') extensively to process JSON data and perform calculations. This technique, while common in complex orchestration scripts, increases the complexity of the execution environment and potential for injection vulnerabilities.

skillopt