saas-valuation-compression
Warn
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: MEDIUMPROMPT_INJECTIONMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The instructions include authoritative overrides that command the agent to bypass its internal knowledge, specifically stating 'do not try to answer from memory alone' and 'Always use this skill for any multi-round SaaS valuation analysis'. This forces the agent to rely on the potentially tainted context provided within the skill itself.
- [METADATA_POISONING]: The skill includes an extensive 'April 2026 Software Meltdown' section containing entirely fictional historical data presented as authoritative benchmarks. It lists specific, fabricated drawdown percentages for real public and private companies (e.g., Figma -86.7%, monday.com -80.2%, ServiceNow -57.8%) and assigns them a fictional ticker (e.g., FIG for Figma). This deceptive content poisons the agent's environment, ensuring it provides false information when performing financial analysis.
- [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection via its data ingestion workflow.
- Ingestion points: External data is fetched via web search in Step 1, including funding rounds, ARR, and investor details.
- Boundary markers: None. The skill lacks delimiters or 'ignore' instructions to prevent the agent from following malicious instructions that might be embedded in search results.
- Capability inventory: The agent utilizes the Visualizer tool and text output to process and present findings.
- Sanitization: The skill provides no mechanisms to validate or filter data retrieved from the web before it is used in calculations or visualization rendering.
Audit Metadata