context-continuity
Fail
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: HIGHPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PERSISTENCE]: The skill includes scripts that facilitate background persistence on macOS by installing a LaunchAgent ('scripts/install_auto_cycle_launchagent.sh') and repository-level persistence via git hooks ('scripts/install_git_hooks.py'). These methods allow for automatic, repeated execution of skill scripts across different sessions and terminal activities.
- [INDIRECT_PROMPT_INJECTION]: The core rehydration functionality in 'scripts/rehydrate.py' collects data from various ingestion points such as git history, project files, and event logs. This data is then formatted for direct inclusion into an LLM prompt. The process lacks adequate sanitization or robust boundary markers to prevent instructions contained within these external data sources from being interpreted by the agent.
- [COMMAND_EXECUTION]: Multiple utility scripts ('scripts/auto_cycle.py', 'scripts/snapshot.py', 'scripts/memory_store.py') utilize the 'subprocess' module to perform git operations. This establishes a broad command execution capability that could be exploited if malicious data were to influence the commands or arguments processed by these tools.
Recommendations
- AI detected serious security threats
Audit Metadata