v2.1.0 — Safety & Trust · Jun 10 → Jun 30
This is the "stop scaring users" release. All four items are independent so you can ship them in any order.
--dry-run and --backup first — they're the most urgent. Dry-run prints a unified diff to stdout. Backup simply does shutil.copy(path, path + ".bak") before any in-place write. Maybe 2 days each.
Config loader next. Add a config.py module that walks up the directory tree looking for .pycodecommenter.yaml and loads it with ruamel.yaml. Wire the values to the existing hard-coded defaults. A week's work max.
Smarter parameter inference is also v2.1 — see the dedicated audit prompt below.
Test rewrite is ongoing across all phases but the baseline pass (assertions on critical paths) should land here.

v2.2.0 — Accuracy & Output · Jul 1 → Aug 15
This is the "be correct and be composable" release. Decorator awareness cleans up the false-positive noise. JSON output unlocks every downstream integration — make it a flag on both validate and coverage. The Sphinx :raises: fix is a targeted one-liner in validator.py.

v3.0.0 — Safe, Configurable, Extensible · Aug 16 → Nov 1
The big one. The merge algorithm is the hardest item in the entire roadmap — budget 3–4 weeks for it alone. The approach: parse the existing docstring into sections, generate fresh sections, then do a section-by-section merge where any section with human content that doesn't match the auto-generated template is preserved. Multi-style support travels alongside — once the generator is style-aware, NumPy and Sphinx become config flags.

v3.1.0 — Integrations · Nov 2, 2026 → Mar 1, 2027
VS Code extension, GitHub Action for PR automation, SARIF output. These all depend on the v3.0 foundation being solid. Don't start them until merge and JSON output are stable.