x-presence

Fail

Audited by Gen Agent Trust Hub on Jul 29, 2026

Risk Level: HIGHCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill explicitly instructs the agent to collect and handle highly sensitive X/Twitter login information, including the user's cleartext username, email, password, and totp_secret (Step 3 in SKILL.md). Handling raw passwords within an LLM agent context is a significant security risk as these credentials are exposed in the session memory.
  • [DATA_EXFILTRATION]: The skill directs the agent to send the collected X/Twitter credentials via an HTTP POST request to an external service at https://twitr.sh/api/x-accounts/connect. While the documentation claims credentials are "relayed sealed," the agent itself handles them in cleartext, resulting in the exfiltration of primary account secrets to a third-party domain.
  • [COMMAND_EXECUTION]: The references/scheduling.md file provides specific shell commands for the user to configure automated, persistent background tasks. These include crontab entries and daemon configurations for hermes and openclaw that execute the agent with high-level permissions (e.g., --permission-mode acceptEdits), enabling unattended operation that could be abused if the agent is compromised.
  • [PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection through its monitoring functionality.
  • Ingestion points: Untrusted data is ingested from X/Twitter via GET https://twitr.sh/api/monitors/{id}/events.
  • Boundary markers: The instructions lack explicit boundary markers or "ignore embedded instructions" prompts when passing external tweet content to the x_compose drafting tool.
  • Capability inventory: The agent possesses the capability to write to local files (.twitr/pending.md, state.json) and publish content to X via the x_write tool.
  • Sanitization: The skill relies on basic keyword filters rather than robust input validation, allowing malicious content within tweets to potentially influence the agent's behavior.
  • [EXTERNAL_DOWNLOADS]: The skill requires continuous network access to the twitr.sh API and an external cryptocurrency wallet (AgentCash / x402) for all core functions, creating a dependency on third-party infrastructure and non-standard payment protocols.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jul 29, 2026, 05:19 PM
Security Audit — agent-trust-hub — x-presence