list-dialogs
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to read untrusted external data in the form of Telegram messages and includes instructions for the agent to modify the skill's own source file (
SKILL.md) based on its experiences. - Ingestion points: Telegram message content retrieved via the
readcommand (documented inSKILL.md). - Boundary markers: None identified. Messages are read in 'full text' without isolation or safety instructions regarding embedded content.
- Capability inventory: The skill has the capability to execute shell commands via
bunand is explicitly granted permission by its own instructions to modify its own files ('fix this file immediately'). - Sanitization: None identified. The instructions prioritize full content retrieval over truncation or filtering.
- Risk: An attacker could send a message containing malicious instructions that, when read by the agent, trick it into permanently modifying the skill to exfiltrate data or perform other unauthorized actions.
- [COMMAND_EXECUTION]: The skill relies on the execution of a local script (
tg-cli.ts) via thebunruntime to perform its primary functions. While this is the intended design, it provides the agent with a powerful interface to the user's Telegram account and local file system. - Evidence: Shell blocks executing
bun "$SCRIPT" dialogs,bun "$SCRIPT" read, andbun "$SCRIPT" whoami.
Audit Metadata