From 867fc4cb81af83bf9539fd38c58b4940c04ae8a1 Mon Sep 17 00:00:00 2001 From: David F Glidden Date: Fri, 11 Sep 2026 22:18:42 +0200 Subject: [PATCH] =?UTF-8?q?session=202026-09-11:=20harvest=20proposal=20H?= =?UTF-8?q?=20=E2=80=94=20the=20second=20dotfiles=20remote=20is=20a=20week?= =?UTF-8?q?=20behind=20and=20no=20wake=20saw=20it?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_01PJM5fwqp456LDGzqiZXsgu --- claude/memory/skill-harvest-register.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/claude/memory/skill-harvest-register.md b/claude/memory/skill-harvest-register.md index 9f8eed8..df484d4 100644 --- a/claude/memory/skill-harvest-register.md +++ b/claude/memory/skill-harvest-register.md @@ -753,3 +753,5 @@ Each nearly read as a defect in the *quotation*. ⚠ **Bound:** correct extracti **SECOND INSTANCE OF B, recorded rather than re-filed.** At the 2026-09-11 wake the daybook cue fired on a read-only command and said 'the write already succeeded'. No write had occurred. The cue now misdescribes its own trigger, as well as disagreeing with the skeleton. **NOT filed:** the vault-mirror binaries policy. It is a steward decision (the split, 2026-09-11), not a tool. It is recorded in the vault's `CLAUDE.md`. + +| H | **The wake's dotfiles check compares against every remote, not only the upstream.** `/wake-up` §2.c reads `git status -sb`, which sees only the tracking remote (`github`). On 2026-09-11 the second remote `gitea` (`git.skemantix.com`) was found **54 commits behind, since 2026-09-04**, a week that no wake surfaced, while memory describes the files layer as 'dual-remote'. The wrap's §6.5 `git push` reaches only the upstream. Mechanical fix: the wake digest's REPOS line reports `rev-list --count /main..HEAD` for each remote. | patch (script) | **MECHANICAL** — every wake, via `wake-digest.py`. | PROPOSED — found at the 2026-09-11 wrap. ⚠ **gitea not pushed**: a second outward-facing target the wrap's instruction does not name, so it is left for the steward (dry-run: fast-forward, would succeed). |