d-official-market-analysis
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is built to ingest and process data from external market reports, PDFs, and web pages, which represents a potential attack surface for instructions embedded in third-party content.
- Ingestion points: Processes external CSV/JSONL datasets, PDF reports, and web content from industry sources.
- Boundary markers: The skill mandates strict data schemas (e.g., source_url, evidence_level) but lacks explicit delimiters or instructions to ignore commands within the ingested text.
- Capability inventory: Includes the ability to execute a local Python script (
scripts/validate_dataset.py) to process these datasets. - Sanitization: The validation script checks for structural correctness and format (e.g., date patterns) but does not verify the semantic content for prompt injection risks.
- [COMMAND_EXECUTION]: The skill instructions direct the agent to run a local Python script for dataset validation.
- Evidence:
SKILL.mdexplicitly instructs the agent to runpython scripts/validate_dataset.py <dataset.csv|dataset.jsonl>.
Audit Metadata