mark-read
Warn
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill accesses Telegram session files stored at
~/.local/share/gramjs/<profile>.session. These session files contain sensitive authentication tokens that provide full access to a user's Telegram account if compromised.\n- [INDIRECT_PROMPT_INJECTION]: The skill interpolates the user-providedchat_id_or_usernameparameter directly into a Bash script. Without proper sanitization, this creates an attack surface for command injection.\n - Ingestion points: The
chatparameter used in the Bash usage block inSKILL.md.\n - Boundary markers: None present; the parameter is placed directly into the command string.\n
- Capability inventory: Access to shell execution (
Bash), file system navigation (Glob), and file reading (Read).\n - Sanitization: No evidence of input escaping or validation logic.\n- [DYNAMIC_EXECUTION]: The "Self-Evolving Skill" directive instructs the agent to modify the
SKILL.mdfile immediately based on runtime results. This self-modification capability could be abused by an attacker who triggers specific errors to trick the agent into writing malicious instructions into its own definition, effectively achieving persistence.
Audit Metadata