baskduf/FableCodex — reverse-engineered prompt

Build me a Codex plugin called FableCodex that gives Codex a more careful, evidence first workflow, kind of like Fable habits but still fully native to Codex.

I want it so when I call something like @codex fable5 in a prompt, the agent inspects the workspace before changing things, classifies the task, uses real files and tools instead of guessing, tracks multi step work in a simple goal ledger, records review findings that have to be closed before final completion, and always verifies with tests or other clear proof before saying the job is done. It should also support a lighter review only mode and an analysis only mode where no files are edited.

Please include the local state files for goals and findings, helper commands to create progress checkpoints, resolve findings, check status, and update the plugin, plus examples and tests so the workflow is easy to understand and trust. Keep the scope to workflow guidance and verification discipline, not changing model behavior. If you need to, look up the current Codex plugin docs online.