promoting-work-log-insightslisted

# promoting-work-log-insights The `/writing-work-logs` skill writes a durable narrative after each task, and its Step 4 asks the author to fold generalizable insights into the rules **in the same change set**. That step is discretionary and single-log: the same agent that just wrote one log decides, in the moment, whether an insight is worth promoting. Nothing ever looks _across_ logs. So an insight that shows up once and is skipped as "too specific" can recur three logs later and still never reach `.claude/rules/` — the loop leaks. This skill closes that leak. It reads the whole `docs/logs/` corpus at once, finds the insights that recur (recurrence is the strongest signal that an insight generalizes), and promotes each to the durable home where it will actually change future behavior. A promoted insight is stamped back in its source logs so the next run skips it — that provenance marker is how the loop stays closed instead of re-proposing the same thing forever. See [`docs/contributing/glossary.md`](../../../docs/contributing/glossary.md) (ADR-0099) for the strict definitions of **observation** (the raw, run-specific noticing) and **insight** (the generalized, actionable claim this skill promotes) this page assumes throughout. **Three evidence sources, not one** (ADR-0084). Work logs are the narrative record, but they are not the only place this project's lived experience accumulates: | Source | What it is good for