The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection vulnerability surface where untrusted data could influence agent behavior.
Ingestion points: Data enters the agent context through ze.Dataset.pull() from external benchmark repositories or via local CSV files using ze.Dataset() in SKILL.md and references/dataset-lifecycle.md.
Boundary markers: Absent. No specific delimiters or instructions are provided to separate data content from system instructions.
Capability inventory: Ingested row data is processed within evaluation tasks (dataset.eval()) or downstream logic via custom task functions (@ze.task) as shown in references/dataset-lifecycle.md.
Sanitization: Absent. The data is parsed and used directly without validation or input sanitization.
[EXTERNAL_DOWNLOADS]: The skill installs the zeroeval Python package from public registries during initialization in SKILL.md. This dependency matches the verified author context.
[COMMAND_EXECUTION]: The skill executes shell commands for environment setup and version control management, including pip install zeroeval, zeroeval setup, and standard Git operations (git clone, git commit, git push) to interface with git.llm-stats.com as outlined in SKILL.md and references/git-workflow.md.

manage-data