Migrate Custom GPTs
migrate-cgpt backs up a user-owned Custom GPT privately, then converts it into a skill folder, Git repository, or supported ChatGPT plugin. It preserves required behavior and reports anything the destination cannot yet do.
Install and invoke
Section titled “Install and invoke”npx skills add thatguysam/skills --skill migrate-cgptUse $migrate-cgpt to convert these GPT instructions and reference files into a local Codex skill.Use $migrate-cgpt to migrate this authorized GPT configuration into my GitHub skills repository.Claude Code uses /migrate-cgpt for a standalone installation and /sam:migrate-cgpt in the collection bundle.
Read backup and installation steps for the original-source snapshot, canonical skill schema, ChatGPT Web options, and a hand-test checklist.
Behavior
Section titled “Behavior”The skill creates and verifies a private source snapshot, inventories the source, preserves the task’s output and permission requirements, maps dependencies, writes the replacement, and checks the actual destination. It keeps the original GPT intact during portable conversion. An integration schema does not count as a working Action.
The authoring method uses Zach Prompting’s requirement-preservation rules and Matt Pocock’s writing-for-agents guidance. It removes behavior-neutral instructions and puts conditional details behind clear reference links.
Inputs & outputs
Section titled “Inputs & outputs”Supply instructions or authorized editor access, available knowledge files, relevant Action configuration without credentials, and a target runtime or repository. Familiar requests and good outputs make comparison stronger.
The result contains SKILL.md and only the references, assets, metadata, or scripts the workflow needs. The migration report gives the destination, component mapping, actual checks, unresolved dependencies, and an invocation example.
Verification has two parts: compare the converted files with the source requirements, then run the generated skill with representative inputs. Keep expected requirements separate from the runners and retain their actual responses. A correctly reported missing integration can pass this evaluation while the replacement remains partial.
States & edge cases
Section titled “States & edge cases”ready: required behavior and delivery verified.partial: useful conversion delivered with named missing dependencies.awaiting-evals: packaging checked, required execution evidence still missing.blocked: source access or another essential input prevents a useful conversion.
A public GPT URL does not expose its configuration. A missing roster or knowledge file must remain missing, not be reconstructed from guesses. Authentication must be configured separately. A cloud file is not an installation on your computer.
Built-in ChatGPT migration is a separate operation. Check the official retirement FAQ for account-specific availability and effects before using it.
Data shape
Section titled “Data shape”Each source component has a destination and a status: preserved, adapted, missing, or intentionally excluded with a reason. Each evaluation records its input, expected behavior, observed result, and pass/fail/blocked outcome. The optional JSON backup manifest records captured metadata, file hashes, and missing components. It is a project capture record, separate from standard Agent Skills frontmatter; it is not a GPT restore format.
Verification
Section titled “Verification”See recorded checks and limits. Portable conversion does not establish identical output from another model or restore an absent API connection.