qiqi-project-financial-analysis
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides templates for executing shell commands via the
qiqiCLI to retrieve financial data. These commands interpolate user-provided parameters such as project names, dates, and IDs (e.g.,qiqi query list Project --where name contains <project-name>). This is the primary mechanism of the skill and relies on the execution environment to sanitize inputs. - [INDIRECT_PROMPT_INJECTION]: The skill processes project-related data retrieved from external ERP systems, which serves as a potential vector for indirect prompt injection if record fields (like project descriptions or contract terms) contain adversarial content.
- Ingestion points: Data enters the context via the
qiqitool's query results, as detailed inreferences/query-playbook.mdandreferences/budget-and-profit.md. - Boundary markers: The instructions emphasize 'evidence paths' and 'transparent derivation,' but do not specify technical delimiters (like XML tags or specific quotation markers) to isolate external data from instructions.
- Capability inventory: The agent uses the
qiqiCLI for reading records, describing object metadata, and performing server-side aggregations. - Sanitization: No specific data sanitization or instruction-filtering logic is mentioned in the skill files.
Audit Metadata