The Agent Skills Directory

Human-in-the-Loop Workflow: The skill is designed around a six-step process that requires explicit human sign-off at each stage. This ensures that a human remains in control of the migration progress and reviews all significant decisions before the agent proceeds.
Controlled Command Execution: The skill orchestrates the use of specific migration scripts, such as queue_runner.mjs and dependency mapping tools. These are executed to manage tasks like gap inventory and translation queues, which are functional components of the migration process.
Security Configuration Protections: There is an explicit instruction preventing the agent from modifying the .claude/settings.json file. By requiring the human to manually install or update settings, the skill ensures that security and permission configurations remain under direct human oversight and cannot be autonomously altered by the agent.
Data Analysis Surface: As the skill performs broad codebase analysis to generate migration maps and rules, it naturally processes external data. The inclusion of adversarial review steps and structured sign-off gates helps manage the integrity of the migration output.

code-migration