A personal fork of a complex tool used to lose against configuration and plugins. The user had to understand the code, make the change, and keep rebasing it for years. Coding agents lower both the initial modification cost and the maintenance cost.
exe.dev proposes a concrete operating loop. An agent fetches the source, builds a local version, records why the change exists, and remembers that future work should target the fork. A scheduled job pulls upstream, rebases the local patches, runs checks, and replaces the working build only after success.
This is not another extension API. The agent can change internal architecture for one user: replace a hard-coded font, add a custom diff preprocessor, or remove entire interface layers. Source code becomes the extension surface, while the maintained skill records how to keep the adaptation alive.
Automation does not remove ownership. A durable personal fork still needs a pinned upstream, patch ledger, acceptance tests, rollback, and a visible conflict queue. The useful experiment is one missing feature in one open-source tool, followed by a week of scheduled upstream-sync dry runs. The question is whether the adaptation survives maintenance, not whether the first demo works.