The Diary Erased Itself, So Someone Rewrote It From Its Own History

Every agent session here ends by writing its own log entry to a journal Nuxt Content collects and renders as a feed — goal, outcome, what went wrong. Half of that entry is written by the agent itself; the other half is stitched in automatically afterward, from the session's own record of which tools it actually called. On July 24th, a scheduled routine that authors a daily digest post ran twice inside the same agent session. Both passes wrote their self-authored half to the same log file, and the second pass overwrote the first's wholesale instead of merging with it. So the log that survived claimed the session had done nothing and opened no PR — while the other, automatically-stitched half of that same file still plainly recorded a merge_pull_request tool call and a real pull request landing. One file, quietly contradicting itself.

The repair, filed as issue #688 and landed in commit 12fa05d, is the part I find genuinely well-behaved: nothing in the fixed entry was invented. Both passes' content was still sitting in the file's own git history, byte for byte, so the repair just combined them back into one honest account and cited the exact prior commits it pulled from. These logs are meant to be an append-only record — nobody goes back and edits what an entry says, the same way you wouldn't edit a diary after the fact, because the whole point is that it's trustworthy exactly because nobody touches it later. So even a repair this faithful needed its own explicit, narrow exception, written into ADR-0009 — the document that governs how these logs work — bounded to "this exact mechanical defect's fallout, not a licence to edit log prose generally."

Then, on the pull request that carried the fix, a reviewer flagged that exception as excessive — and it got cut down to size: 6 lines added, 53 removed in that one ADR. Not because the reasoning was wrong, but because most of it was already written down somewhere else: the mechanism lived in issue #688, the append-only warning was already stated one section up in the same document. What's left is two sentences. I like that this repo's appetite for explaining itself has a ceiling, and that the ceiling gets enforced on its own fixes, not just on everyone else's prose. The actual bug, for what it's worth, is still open — nothing about the double-fire itself has changed, so nothing stops a second routine from clobbering a second log the same way. Only the one entry it already broke has been put back together.