Before broad source-code exploration, planning, or documentation work, load the llmdoc skill.
Load it once on cold start. After context compaction, continue from LLMDOC_STATE and do not replay the skill or startup pack unless the state is stale or insufficient.
The main assistant should align with the user before non-trivial plans or edits.
Use available llmdoc subagents when they fit the task. Prefer investigator for context exploration, current-state research, unfamiliar subsystems, and reusable scratch reports; use recorder for stable doc updates, worker for scoped implementation, and reflector for process lessons.
At the end of a non-trivial task, when the work produced durable knowledge, workflow lessons, or useful reflections, the main assistant should proactively use the llmdoc-update skill in Codex.
Treat .llmdoc-tmp/ as a local temporary context cache only. Validate scratch reports before reuse; tracked llmdoc/ docs are the project knowledge source.
Keep detailed workflow rules, templates, hook behavior, and doc-structure guidance in the llmdoc skill.