The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill refers users to an external repository (github.com/joaquinhuigomez/agent-eval) for installation. This source is not from a well-known or trusted organization, representing a potential supply-chain risk if the external tool is malicious.
[COMMAND_EXECUTION]: The evaluation workflow relies on executing arbitrary shell commands specified within YAML task configurations (e.g., using the pytest or npm run build commands in the judge block). This creates a surface where malicious commands could be executed if configuration files are sourced from untrusted locations.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests task definitions from YAML files that include natural language prompts and executable commands without explicit sanitization.
Ingestion points: YAML task definition files within the tasks/ directory.
Boundary markers: None identified in the documentation or example YAML structures to separate instructions from untrusted data.
Capability inventory: The agent is granted Bash, Read, Write, and Edit tools which can be leveraged by injected instructions.
Sanitization: There is no evidence of input validation or command escaping before the commands are passed to the shell.

agent-eval