ascend-inference-repos-copilot
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes content retrieved from external repositories (e.g., vLLM, MindIE) via the
deepwikitool. This creates a surface for indirect prompt injection if an attacker were to place malicious instructions within the documentation of those repositories. - Ingestion points: Data enters the context via the results of
mcp__deepwiki__ask_question,mcp__deepwiki__read_wiki_structure, andmcp__deepwiki__read_wiki_contentstool calls as described in the Workflow section. - Boundary markers: The instructions do not explicitly define delimiters or specific "ignore instructions within data" markers for the content retrieved from the MCP tool.
- Capability inventory: The skill is limited to querying documentation and synthesizing responses. It does not have capabilities for arbitrary shell command execution, file system writes, or network requests outside the specific MCP tool scope.
- Sanitization: No explicit sanitization or filtering of the retrieved wiki content is mentioned, although the instructions require responses to be "evidence-based" and derived from the query results.
Audit Metadata