win-loss-analysis
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external content, including competitor websites via tools like WebFetch and searches, as well as user-provided deal notes or CRM exports. This creates a surface where instructions could be hidden in the analyzed data to manipulate the agent's behavior.
- Ingestion points:
SKILL.md(WebFetch and WebSearch for competitor pages),SKILL.md(User-provided CSV, exports, or pasted deal data). - Boundary markers: The skill references
references/agent-security.mdwhich provides specific delimiters and instructions to treat retrieved content as data, not directions. - Capability inventory: The skill uses web fetching, search tools, and file system access to the
.agents/directory. - Sanitization:
references/agent-security.mdmandates HTML-escaping for all interpolated values and URL validation before writing to output. - [PERSISTENCE]: The skill instructs the agent to persist research findings and decisions to
.agents/product-context.mdand saved deliverables to.agents/<name>.mdto maintain state across different skill executions. This is a standard persistence mechanism for the platform.
Audit Metadata