skill-social-performance-review
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from platform exports and user inputs, creating an indirect prompt injection surface.
- Ingestion points: Phase 1 collects data from CSV files, dashboard screenshots, and user-provided descriptions across multiple social platforms.
- Boundary markers: Phase 3 instructions do not specify explicit delimiters or "ignore previous instructions" safety markers for the agent when interpreting script results and raw data.
- Capability inventory: Phase 3 executes a local Python script via the shell; Phase 4 optionally utilizes web scraping tools (Firecrawl/Playwright MCP); Phase 6 writes generated reports to the filesystem.
- Sanitization: The skill implements a data standardization step in Phase 2/3, mapping raw inputs to a structured JSON schema which provides a layer of data isolation before interpretation.
- [COMMAND_EXECUTION]: The skill executes a local script to process data.
- Evidence: Phase 3 utilizes
python3to run a local script (scripts/review.py) for calculating engagement rates and rankings. This is a standard and safe implementation for performing deterministic calculations that are difficult for an LLM to do accurately by hand.
Audit Metadata