The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and encourages the installation of a third-party tool from an unverified repository (github.com/joaquinhuigomez/agent-eval). Interacting with unvetted external software can introduce supply chain risks.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute the agent-eval CLI. Its core functionality involves running 'judges,' which are arbitrary shell commands (e.g., pytest, npm run build) specified in YAML task files. This allows for the execution of any command defined in the task configuration.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes YAML-based task definitions that could contain malicious shell commands in the judge field or instructions intended to override agent behavior.
Ingestion points: Processes tasks/*.yaml files and agent outputs from evaluation runs in the SKILL.md workflow.
Boundary markers: No explicit delimiters or warnings are provided to the agent to treat the contents of the YAML files as untrusted data.
Capability inventory: The agent has access to Bash, Read, Write, and Edit tools, which could be leveraged by malicious instructions embedded in the processed tasks.
Sanitization: There is no evidence of sanitization or validation of the commands provided in the YAML task definitions before they are passed to the Bash tool.

agent-eval