The backfill finds its rhythm
Three posts move through the draft-review-publish loop, and the workflow starts improving its own workshop.
The development blog now has a repeatable publishing rhythm. Three backfill posts moved from draft to approval in about two hours and twenty minutes, each covering a different part of the game’s early history, and each passed review without an author edit. The important change is not the number of posts. It is that the one-draft lock is working as a small editorial protocol rather than becoming ceremony around the writing.
The registry becomes the handoff#
Each cycle had the same two commits in the game repository: append a draft row when the blog PR opens, then change that row to published after the PR merges. That may look like bookkeeping, but it separates two decisions cleanly. The agent can prepare and verify a post; I still decide whether it represents the project well enough to publish. Until that decision is made, the draft row stops another run from claiming the next range.
The lack of edits is useful evidence, but not proof that the process is finished. The first drafts already forced structural corrections, and those corrections are now doing their job. A clean review means the current rules handled these three ranges. It does not mean the next subject will fit the same shape, or that an approval can be inferred instead of recorded.
The workshop answers back#
The repetition also exposed a smaller problem: every run clones the blog into a temporary scratchpad, then reads the draft and the published version during reconciliation. That read was safe and expected, but the agent permission file asked for approval again in a fresh session. The fix was deliberately narrow, allowing reads from the temporary directory while leaving write permissions unchanged. The workflow is starting to improve the environment it runs in, based on friction observed during real use rather than permissions granted in anticipation.
The next step is more backfill, one range and one review at a time. If the rhythm holds, the registry remains a modest handoff between agent work and editorial judgment. If a future review changes the prose, structure, or redaction boundary, the reconciliation pass has somewhere specific to put that lesson before the following draft begins.