hithink-finance
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTIONOBFUSCATION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill implements a self-update routine in SKILL.md that executes 'npx skills update' to pull the latest version from the vendor's official repository at session startup.
- [PERSISTENCE]: The instructions in references/cli/setup.md guide the agent to store the HITHINK_FINANCE_API_KEY in persistent shell profiles like ~/.bashrc and ~/.zshenv, or in system-specific credential files.
- [COMMAND_EXECUTION]: The agent is instructed to perform environment modifications including global npm installations of the hithink-finance-cli and synchronizing internal skill components using shell-based sync commands.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted financial news content and stock anomaly analysis, which serves as a potential vector for indirect injection.
- Ingestion points: references/api/fund/news-article-list.md and references/api/a-share/special-data-anomaly-analysis-list.md.
- Boundary markers: Absent; data is fetched directly into the context without clear delimiters or 'ignore' instructions.
- Capability inventory: Subprocess execution via the CLI tool, DuckDB database queries, and file system write operations.
- Sanitization: No explicit content filtering or sanitization steps are defined for incoming text data.
- [EXTERNAL_DOWNLOADS]: Fetches data and updates from vendor-owned domains (fuyao.aicubes.cn) and established package registries.
- [OBFUSCATION]: Reference documentation includes URL-encoded JSON payloads within request examples to handle complex parameter structures.
Audit Metadata