openfinclaw-quantitative-research
Warn
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to install the OpenFinClaw CLI using
npx @openfinclaw/cli@latest install. This method downloads and executes code from the NPM registry, which is a third-party source not included in the trusted vendor list. - [COMMAND_EXECUTION]: The provided code examples and usage patterns explicitly instruct the AI agent to use shell execution tools such as
subprocess.run(Python) andchild_process.exec(Node.js) to interact with the local operating system and the OpenFinClaw CLI. - [DATA_EXFILTRATION]: The skill contains functionality to upload local strategy files and workspace data to remote servers via the
strategy_publishtool andopenfinclaw publishcommand. This represents a potential vector for exfiltrating sensitive local files if the agent is directed to publish unintended directories. - [INDIRECT_PROMPT_INJECTION]: The skill processes research data, metrics, and strategy information fetched from remote API endpoints (
hub-gw.openfinclaw.ai), creating an attack surface for indirect prompt injection. - Ingestion points: Remote research results and strategy metadata are ingested through the
fin_deepagent_research_poll,fin_deepagent_research_finalize, andstrategy_get_infotools. - Boundary markers: The instructions do not define boundary markers or safety directives to prevent the agent from following instructions potentially embedded in the remote research output.
- Capability inventory: The skill provides the agent with the ability to execute system commands (
exec,subprocess.run), write to the filesystem (strategy_fork), and perform network operations. - Sanitization: There is no evidence of sanitization or validation of the content received from the remote API before it is added to the agent's context.
- [DYNAMIC_EXECUTION]: The skill demonstrates patterns for dynamically generating and executing logic based on remote tool output, such as parsing backtest IDs from CLI stdout to programmatically construct subsequent API requests.
Audit Metadata