better-work

Warn

Audited by Gen Agent Trust Hub on May 12, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill's initialization workflow creates a global state directory in the home folder (~/.better-work/) and performs shell-level operations including initializing git repositories and creating symbolic links between project directories and the home directory.
  • [COMMAND_EXECUTION]: The skill modifies persistent agent configuration by injecting behavioral instructions into CLAUDE.md files at both project and global (~/.claude/CLAUDE.md) levels. The documentation states that global injection is the default behavior and is performed without a separate prompt.
  • [EXTERNAL_DOWNLOADS]: The skill references external GitHub repositories (e.g., github.com/d-wwei/better-code) and instructs the agent to delegate tasks to these external subskills, potentially involving the installation or execution of external components.
  • [COMMAND_EXECUTION]: The protocol uses the Agent tool to dynamically spawn sub-agents for 'Adversarial Reviews', which involves task delegation to separate agent instances for architectural and logic verification.
  • [COMMAND_EXECUTION]: The skill maintains a global project registry in ~/.better-work/registry.yml, involving cross-session reading and writing of project metadata such as file paths and initialization timestamps.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 12, 2026, 08:48 PM