self-reflection

Warn

Audited by Gen Agent Trust Hub on Feb 28, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: Accesses sensitive files located in ~/.config/alma/chats/, which contain private 1-on-1 chat history, and ~/.config/alma/SOUL.md, which stores the agent's personality traits.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes raw logs from group and private chats. 1. Ingestion points: Log files in ~/.config/alma/groups/ and ~/.config/alma/chats/. 2. Boundary markers: Minimal headers are used to separate logs, but no specific delimiters or instructions exist to prevent the model from following commands found within those logs. 3. Capability inventory: The skill has access to the Bash and Write tools, allowing it to modify files or execute system commands based on the reflection process. 4. Sanitization: No content filtering or validation is performed on the ingested log data.
  • [COMMAND_EXECUTION]: Uses Bash to iterate through and read multiple log files, which could lead to unintended file access if path variables are compromised.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 28, 2026, 09:32 AM