skills/ztemerbekov/next-move-theory-canon-and-skills/nmt-craft-value-proposition/Gen Agent Trust Hub
nmt-craft-value-proposition
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_CONTEXT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources including web reviews (Capterra, G2, Reddit, Product Hunt) and user-supplied documentation such as Notion exports or strategy notes. This creates a potential surface for indirect prompt injection where instructions embedded in the external content could influence the agent's logic. The skill mitigates this by enforcing a policy to treat all external inputs as hypotheses requiring validation.\n- [DYNAMIC_CONTEXT_INJECTION]: The instructions reference the
!commandsyntax used for environment awareness, specifically naming development tools such as `!`git logand!gh pr diff``. These commands execute in the shell environment when the skill is loaded to provide context to the agent.\n- [EXTERNAL_DOWNLOADS]: In high-depth research modes, the skill fetches data from the web using recognized external services like Firecrawl and Exa to perform automated review mining and competitive analysis.
Audit Metadata