wikieval
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes a local script
harness/scripts/wikieval.pyusing the system'spython3interpreter. This is used to process golden benchmarks and candidate outputs to generate reports and manage regression baselines. - [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from external sources, including golden benchmark files (
llmwiki/wiki/sources/evals/*.md) and candidate output files (outputs.json). - Ingestion points: Evaluation data is read from the specified markdown and JSON files.
- Boundary markers: The skill utilizes YAML frontmatter in markdown files to structure inputs, expectations, and assertion rules.
- Capability inventory: The skill executes a local Python script that performs file read/write operations for metrics and baselines. It has no network access and model invocation is disabled.
- Sanitization: The primary logic uses deterministic code-based assertions (equals, contains, regex, is-json) rather than natural language interpretation, which prevents embedded instructions from influencing the evaluation logic.
Audit Metadata