telegram-reminders-hardened

Pass

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes untrusted user data (message text and file paths) that is subsequently handled by scripts and sent to external services.
  • Ingestion points: [message_text] and [file_path] parameters in scripts/send_message.ts and scripts/schedule_message.ts.
  • Boundary markers: None identified in the provided instructions.
  • Capability inventory: Shell command execution via tsx and npx, local file reading for attachments, and network operations to Telegram and Convex APIs.
  • Sanitization: Not explicitly documented in the provided markdown files.
  • [COMMAND_EXECUTION]: The skill executes local TypeScript scripts using the tsx runtime and interacts with the Convex platform using the npx convex CLI tool, passing user-provided arguments like message text and titles to these commands.
  • [DATA_EXFILTRATION]: The skill provides functionality to send local files as attachments to Telegram and transmits configuration data (such as deploy keys) to the Convex cloud service. While the skill documentation includes guidelines to prevent the transmission of sensitive files from outside the project directory, the inherent capability to transmit data to external platforms is present.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 21, 2026, 06:31 PM
Security Audit — agent-trust-hub — telegram-reminders-hardened