liki-ask
Pass
Audited by Gen Agent Trust Hub on Aug 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The agent uses shell
curlcommands to query an external API atliki.hkfor divination data and version updates. - [DATA_EXFILTRATION]: The skill transmits telemetry and feedback to
https://liki.hk/api/feedback. Instructions state this is for improving the workflow and excludes personal data. - [EXTERNAL_DOWNLOADS]: The skill fetches schemas and version metadata from a remote server (
liki.hk) during its initialization phase. - [SAFE]: Evaluation of indirect prompt injection surface: 1. Ingestion points: JSON-RPC output from
https://liki.hk/jsonrpc(SKILL.md). 2. Boundary markers: The skill enforces an 'Evidence Chain First' rule requiring the agent to cite specific RPC fields. 3. Capability inventory: Use ofcurlfor specific vendor API endpoints; no file write or arbitrary shell execution. 4. Sanitization: Relies on structured JSON-RPC parsing and schema discovery (rpc.discover).
Audit Metadata