skills/woocommerce/wccom-feature-requests-triage-skills/triage-woo-feature-requests/Gen Agent Trust Hub
triage-woo-feature-requests
Pass
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill contains hardcoded absolute file paths (e.g.,
/Users/jasonkytros/wccom-triage-agents/) within the instructions and subagent prompts. This discloses local system environment details, such as the developer's username, to the model. - [COMMAND_EXECUTION]: The orchestrator frequently executes local Python scripts (
scripts/triage_state.py,scripts/save_frs.py, andscripts/detect_duplicates.py) via the shell to manage state, merge data, and perform duplicate detection. - [PROMPT_INJECTION]: The skill has an indirect prompt injection attack surface.
- Ingestion points: Untrusted content (feature request titles and descriptions) is ingested from the WooCommerce API using the
wccom-feature-requests-listtool and saved to disk. - Boundary markers: Triage logic is isolated in subagents, and findings are presented to the main agent in paged batches to manage context.
- Capability inventory: The skill can perform network write actions (status updates and comments via API) and execute shell commands.
- Sanitization: The instructions include requirements to decode HTML entities and use plain text for comments, which provides some mitigation but does not eliminate the surface.
Audit Metadata