upwork-client-messenger

Warn

Audited by Gen Agent Trust Hub on Jul 23, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill is instructed to access potentially sensitive local file paths.
  • Evidence: The skill documentation indicates it should read identity and rate context from upwork-profile.md or files within the ~/.claude/ directory.
  • Context: This access is used to personalize the drafted messages. However, ~/.claude/ is a hidden configuration directory for the agent platform, and accessing it for skill-specific data is a behavior that requires review to ensure it does not expose other sensitive agent configurations.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its primary data ingestion mechanism.
  • Ingestion points: The skill accepts "thread context" (pasted client messages) as a primary input in the SKILL.md instructions.
  • Boundary markers: The instructions do not define clear boundary markers or delimiters to encapsulate the untrusted client text.
  • Capability inventory: The skill has the capability to read local files (upwork-profile.md) and generate text based on user/client input.
  • Sanitization: There is no instruction to sanitize or ignore commands embedded within the pasted client thread context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 23, 2026, 09:50 AM
Security Audit — agent-trust-hub — upwork-client-messenger