The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a Python script that takes a configurable evaluation command as an argument. The variable {eval_cmd} is interpolated into a bash command line: python {skill_path}/scripts/result_ranker.py --eval-cmd \"{eval_cmd}\" .... If this command is sourced from untrusted project configurations or agent-influenced data, it represents a command injection vulnerability.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes outputs from potentially untrusted agents, creating a vulnerability to indirect prompt injection. 1. Ingestion points: The skill reads git diff outputs and agent-generated result files located at .agenthub/board/results/agent-{i}-result.md as described in SKILL.md. 2. Boundary markers: No explicit boundary markers or 'ignore embedded instructions' warnings are mentioned for processing agent result data. 3. Capability inventory: The skill performs shell execution of local Python scripts (result_ranker.py, session_manager.py) and interacts with the Git repository as seen in SKILL.md. 4. Sanitization: No sanitization, escaping, or validation of the external agent-generated content is described before it is processed for ranking.

eval