skills/smartcontractkit/chainlink-agent-skills/chainlink-confidential-ai-attester-skill/Gen Agent Trust Hub
chainlink-confidential-ai-attester-skill
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted document content (PDF, PNG, HTML) via an external AI service, which presents a surface for indirect prompt injection.
- Ingestion points: The POST /v1/inference endpoint consumes Base64-encoded files or fetches data from remote URLs as described in references/api-reference.md.
- Boundary markers: System prompt templates in references/prompts.md lack robust delimiters to isolate untrusted content from the agent's instructions.
- Capability inventory: The skill is granted Bash, WebFetch, Read, and Write tools in SKILL.md, enabling local file access and network communication via curl.
- Sanitization: There is no explicit validation or sanitization of document content before it is passed to the inference engine.
Audit Metadata