The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/linkedin-connect.js explicitly disables global SSL/TLS certificate verification by setting NODE_TLS_REJECT_UNAUTHORIZED = "0". This is a dangerous security practice that leaves all network requests made by the process vulnerable to Man-In-The-Middle (MITM) attacks.- [DYNAMIC_EXECUTION]: Both automation scripts construct JavaScript code as strings and execute them within the browser context via Runtime.evaluate. These strings interpolate data retrieved from an external database (linkedin_leads), creating a risk of JavaScript injection if the database content is manipulated or contains malicious characters that bypass the provided escaping logic.- [COMMAND_EXECUTION]: The skill uses child_process.execSync in scripts/linkedin-connect.js to execute shell commands (starting the browser via clawdbot).- [DATA_EXFILTRATION]: The skill accesses sensitive environments, including a PostgreSQL database (via DATABASE_URL) and an active LinkedIn browser session. It has the capability to read personal profile data and perform automated social actions.- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by processing untrusted data from an external database and substituting it into messaging templates without rigorous sanitization or boundary markers.
Ingestion points: Database records fetched from the linkedin_leads table.
Boundary markers: None; data is directly interpolated into message strings.
Capability inventory: Database access (pg), shell command execution (execSync), and browser automation (Runtime.evaluate).
Sanitization: Minimal escaping is applied to strings before they are injected into JavaScript execution contexts.

linkedin-connect