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.sh script 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 /loop mechanism 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.sh script 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, and bun. This includes executing local scripts like goal-verify.sh and goal-lint.ts, as well as interacting with the platform's session management tools to send automated input.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 7, 2026, 05:55 AM
Security Audit — agent-trust-hub — goal-and-loop