ai-human-xlsx
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
uv runto execute its included Python scriptscripts/export_xlsx_cli.py. This is a controlled execution model where the script logic is provided within the skill package. It also utilizes benign system utilities likemktempandgit rev-parsefor path management. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided Markdown files, creating an indirect prompt injection surface. This is addressed with high-quality mitigations: 1. Ingestion points: Processes local Markdown files, folder structures, and the
.ai-skills.tomlconfiguration file. 2. Boundary markers: The skill contains a dedicated 'Untrusted Content' section that explicitly instructs the agent to treat all ingested data as non-instructional and to report any found imperatives as findings rather than executing them. 3. Capability inventory: Capabilities are limited to reading repository files, writing a single Excel workbook to a confirmed path, and executing a local Python script. 4. Sanitization: The conversion script parses Markdown into a structured data model (headings, tables, prose blocks) before writing to the grid, preventing raw injection strings from being interpreted as instructions by the agent.
Audit Metadata