linkedin-engage
Warn
Audited by Gen Agent Trust Hub on Apr 8, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The script explicitly disables TLS certificate validation globally by setting
process.env.NODE_TLS_REJECT_UNAUTHORIZED = "0". This makes all network requests made by the script vulnerable to Man-in-the-Middle (MITM) attacks. - [CREDENTIALS_UNSAFE]: The database connection configuration in
scripts/linkedin-engage.jsusesssl: { rejectUnauthorized: false }, which disables SSL certificate verification when connecting to the PostgreSQL database specified inDATABASE_URL. - [COMMAND_EXECUTION]: The script uses
execSyncto execute the shell commandclawdbot browser start --profile clawd. While this appears intended to bootstrap the required browser environment, executing shell commands carries risks if inputs are not properly sanitized. - [REMOTE_CODE_EXECUTION]: The skill uses the
Runtime.evaluatemethod to execute JavaScript code within the browser session. Several instances involve interpolating data scraped from LinkedIn posts directly into the code strings executed in the browser. Although basic escaping and truncation are applied, this pattern of dynamic code generation from untrusted external content represents a code injection surface. - [DATA_EXFILTRATION]: The skill automates a logged-in LinkedIn session to perform actions. This level of browser control allows for the potential extraction of sensitive personal data or the performance of unauthorized actions on behalf of the user.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its processing of external LinkedIn post content.
- Ingestion points:
scripts/linkedin-engage.js(scrapes post content from LinkedIn activity pages). - Boundary markers: None; external content is processed without clear delimiters or 'ignore' instructions.
- Capability inventory: Full browser automation (navigation, clicking, typing) via CDP and persistent database storage access.
- Sanitization: Scraped content is truncated to 50 characters and escaped for JavaScript string literals, but remains part of executable browser scripts.
Audit Metadata