llm-golden-dataset-builder
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes production logs and traces which contain real-world user inputs. These inputs are considered untrusted data and may contain malicious instructions designed to influence the agent's behavior during the curation process.
- Ingestion points: The agent is instructed in
SKILL.md(Step 1) to search the project for production logs or traces in.json,.jsonl,.csv, or.logfiles. - Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore embedded instructions within the ingested traces.
- Capability inventory: The agent has the capability to search the codebase, read files, summarize contents for the developer, and write the curated results to a new
golden-dataset.jsonlfile. - Sanitization: No explicit sanitization or filtering of the log content is mentioned, although Step 4 requires explicit developer approval for each entry, which serves as a manual security control.
Audit Metadata