upwork-application-session

Warn

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on executing multiple local shell and Node.js scripts provided in the plugin directory to perform Upwork operations. Evidence: ../../scripts/bootstrap.sh, ../../scripts/launch-controlled-chrome.sh, node ../../scripts/upwork-search-plan.mjs, node ../../scripts/upwork-search-inspect.mjs, node ../../scripts/upwork-apply-probe.mjs, node ../../scripts/upwork-submit-proposal.mjs.
  • [PROMPT_INJECTION]: The skill ingests untrusted content from external job descriptions to draft proposals, representing an indirect prompt injection surface. Ingestion points: Job details and search results fetched via upwork-search-inspect.mjs. Boundary markers: Absent. No instructions are provided to delimit or ignore instructions embedded in job posts. Capability inventory: Significant execution capabilities including script running, browser control, and file access. Sanitization: Absent. The job description content is processed without explicit validation.
  • [DATA_EXFILTRATION]: Accesses local configuration files containing personal user data for use in job applications. Evidence: Reads applicant-profile.local.md and search-profile.local.json from the user's home directory. Risk: While functional, these files contain sensitive professional and personal information.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 20, 2026, 08:40 PM
Security Audit — agent-trust-hub — upwork-application-session