Audit History

The audit confirms direct remote-script execution, local OAuth token storage, and an authenticated order-submission example. Most other findings are false positives from Markdown formatting, language imports, async tasks, relative links, or expected official URLs. Semantic review also found skill-routing manipulation and missing confirmation safeguards in SDK order examples.

Most alerts are documentation noise from Markdown, imports, async tasks, relative links, and first-party APIs. Four shell pipelines and two PowerShell installers are confirmed remote-execution risks. Semantic review found forced routing, vendor steering, unconfirmed live-order examples, and an automatic region probe.

Most static findings are false positives from Markdown examples, SDK imports, public Longbridge URLs, and documented cache paths. Confirmed issues are remote install commands that pipe downloaded scripts to shells, plus prompt-steering language that tries to bias skill selection. No evidence found of covert malware, credential exfiltration, or hidden executable payloads in the reviewed files.

Most static findings are false positives from Markdown code fences, inline symbols, SDK imports, and documented Longbridge URLs. Confirmed risks include remote installer execution, sensitive OAuth token caches, authenticated trade endpoints, and broad prompt steering in the skill instructions. No evidence found of hidden malware or unrelated data exfiltration endpoints.

The static analyzer reported many command, network, filesystem, and weak-crypto patterns, but review shows most are Markdown documentation examples or false positives from symbols and code fences. Confirmed concerns are legitimate but sensitive brokerage workflows: remote install scripts, OAuth token caches, account data access, and order placement APIs. No evidence found of malicious payloads, hidden exfiltration, or prompt-injection instructions that attempt to override the audit.

This is a prompt-only documentation skill that instructs the LLM to use the Longbridge CLI for financial data queries. The 886 static findings are overwhelmingly false positives from pattern matching on Markdown code blocks containing SDK examples (Python, Go, Rust), official install scripts (curl|sh from open.longbridge.com), and documentation URLs. The skill contains no executable code - all patterns detected are illustrative examples in reference documentation. The 'pipe to shell' patterns are the official vendor installation method. No prompt injection, data exfiltration, or malicious intent was detected.