startup-pressure-test

Pass

Audited by Gen Agent Trust Hub on Jun 6, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to ingest and analyze untrusted external data, which could contain instructions meant to influence agent behavior.
  • Ingestion points: The skill explicitly instructs the agent to read content from external sources including PowerPoint presentations (.pptx), PDFs, Word documents (.docx), screenshots, Excel spreadsheets (.xlsx), and URLs via WebFetch, as defined in the 'First Move' section of SKILL.md.
  • Boundary markers: The prompt does not provide specific delimiters or 'ignore' instructions to isolate the ingested data from the analysis instructions.
  • Capability inventory: The skill uses platform tools for file reading and web fetching. Its operations are limited to text extraction and analysis; it does not demonstrate capabilities for writing to the filesystem, executing shell commands, or performing unauthorized network operations.
  • Sanitization: There is no mention of sanitizing or escaping the data retrieved from external documents or websites before it is processed.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 6, 2026, 03:12 AM
Security Audit — agent-trust-hub — startup-pressure-test