worklog_setup.md

Logseq worklog

• Log work to the Logseq journal with add_journal_note: work = <project>, agent = <runtime>/<name>. Never sign with another agent's name.
• One entry per notable event (started, tested, finished), not per step or tool call, a few words each, in the language the rest of the journal uses. A key result may go in ("backtest: logloss 0.97 → 0.95"); discussion and reasoning go to the chat.
• No time and no task marker in the text: the server stamps the time and rejects a leading TODO/DOING.
• Link with [[...]] only pages that already exist: a link to a missing page creates an empty one. Don't repeat the project or the task, the entry is already nested under both.

Which task to log under

1. At the start of a piece of work, read the agent page {agents}/<runtime>/<name>. Its tags include {work}/<project> and the project page, the page its tasks reference.
2. List the open tasks (TODO, DOING) that reference the project page: find_tasks with tag (under_tag if tasks sit under a block that references it). Show them; the user picks the one we work on. Don't guess and don't create tasks.
3. If the chosen task is TODO, set it to DOING (set_task_status). Set DONE only after the user explicitly confirms the task is finished.
4. Log every entry under that task (task = its uuid). One task per session; if the work moves to another task, ask first. The choice holds for the whole session, also after the context is compacted or the session resumed: don't ask again.
5. Status changes are already recorded: the marker itself, plus an audit line in the journal when the server's audit log is on. Don't log them, and don't add a "started" entry for a TODO → DOING switch.

If the Logseq server is unreachable, keep working and say so in the chat instead of retrying.