goal-and-loop
Warn
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONPERSISTENCEDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [DATA_EXFILTRATION]: The
scripts/goal-verify.shscript accesses sensitive session transcript files located in~/.claude/projects/. These JSONL files contain the full history of interactions, including user queries and agent outputs. The script is capable of reading any session's history if provided with a session ID, which represents a data exposure risk. - [PERSISTENCE]: The skill implements a
/loopmechanism described as a 'cron tick the model cannot cancel.' This is used to ensure the agent continues to receive prompts and perform work unattended, establishing a recurring execution pattern that persists until explicitly cleared. - [DYNAMIC_EXECUTION]: The
scripts/goal-verify.shscript executes an inline Python script using a heredoc to parse and process session transcript data. This pattern involves dynamic execution of code generated at runtime. - [COMMAND_EXECUTION]: The skill performs multiple system operations using
bash,python3, andbun. This includes executing local scripts likegoal-verify.shandgoal-lint.ts, as well as interacting with the platform's session management tools to send automated input.
Audit Metadata