[Eli Pinkerton (wallstop)]

DISCLOSURE: LLM-GENERATED TEXT

Why: A late export write failure leaves documentation images from different batches.

What:

• Stage changed images before replacing published WebPs.
• Restore previous files and timestamps after publication failures.
• Retain recovery files when restoration is blocked.
• Cover staging, replacement, interruption, and rollback failures.

Refs #114.

---

Note

Low Risk
Local docs export tooling and tests only; no runtime, auth, or CI image pipeline behavior beyond safer atomic writes.

Overview
The docs WebP exporter now publishes the entire pending batch atomically instead of replacing files one at a time, so a mid-run failure cannot leave docs/images/ with a mix of old and new captures.

_publish_batch stages encoded files under .docs-webp-*, backs up existing outputs, then os.replaces them. On failure it rolls back using filesystem state (restoring prior bytes/timestamps or removing new files) and rejects non-regular output paths before any write. If rollback cannot complete, it keeps the recovery directory and surfaces paths in the error.

Troubleshooting documents the staging, restore, and manual recovery steps. Tests cover staging/publication/interruption failures, incomplete rollback, and directory outputs.

^{Reviewed by Cursor Bugbot for commit 5bc42c3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4e7503a. Configure here.}