
What happened
A developer's task.md grew to 1,630 lines, about 45% records of finished work, and Claude reported 63 unfinished articles when all were done, because one item's status lived in four places.
Why it matters
Claude is only as reliable as the note it reads, so a handover file with stale flags can make a competent assistant report finished work as pending, and the fix is one status source plus deleting finished rows.
Summaries like this, every morning.
The failure the author describes is not a model limitation but a bookkeeping one. Because a single item's status lived in a heading emoji, a table of contents, the body text, and an outside ledger, updating one copy left the others stale. Claude read an older flag and reported finished work as remaining — three times in one conversation. The correction was structural: split the index (MEMORY.md), the in-progress handover (HANDOVER.md), and stable knowledge into separate files, so changing state never means editing knowledge and vice versa.
The author's later refinements follow the same logic. Finished rows are deleted rather than archived in place, decisions get their own ledger with a column for whether they may be proposed again, and status moves to OPEN.md once a project reaches dozens of items. The table of contents was dropped because it duplicated status. The rule that a number should be looked up at its source rather than copied treats every figure in a note as something that starts aging the moment it is written.
Reading is enforced rather than assumed. After a conversation opened in a different folder began work without reading the project notes and broke a rule, the author paired a SessionStart hook that injects a reading list with a PreToolUse check that blocks the first edit or push if the log shows no sign of reading it. The author notes the same split of index, handover, and individual knowledge files is packaged, with personal values removed, in a paid template called AI organization kit, which he says is not an official Anthropic product.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Ziva.sh tested GPT-6 Luna on 10 beginner Godot prompts, with 'Godot 4' specified, no version, and web search;…

A developer's AI-operated cloud browser was blocked with HTTP 403 three times in one day when posting YouTube…

In a 2026-06-19 commit (b5af2be32), Shelly added a flow where a natural-language request is parsed by parseAge…

An FA engineer's Claude Desktop skill post-processes pptx output to remove per-character font direct specifica…

Google Antigravity lead Varun Mohan posted on X that Gemini usage limits were reset for all users and a promis…

Requests from an iOS app or LINE go to Muse, a Meta personal agent on a cloud VM, which routes jobs through a…
