The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs users to install a CLI tool from an external GitHub repository (github.com/joaquinhuigomez/agent-eval). This repository originates from a personal account and is not associated with a well-known service or trusted organization.
[COMMAND_EXECUTION]: The evaluation workflow uses YAML task definitions that specify arbitrary shell commands in the judge field (e.g., pytest tests/test_http_client.py -v, npm run build). These commands are executed on the host system via the Bash tool, creating a risk of arbitrary code execution if task files are sourced from untrusted environments.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted task definitions from the tasks/ directory.
Ingestion points: YAML files within the tasks/ directory containing prompt and judge fields.
Boundary markers: None identified; instructions and data are not explicitly delimited.
Capability inventory: The skill uses high-privilege tools including Bash, Write, Edit, Read, Grep, and Glob across its operational flow.
Sanitization: There is no evidence of command sanitization or validation of the task definitions before execution.

agent-eval