Tacit reviews Claude Code plugin captures recurring code mistakes as memory

A developer found that Claude Code repeatedly made the same classes of mistakes that human reviewers had to flag, because tacit team knowledge never reached the model. A hand-maintained lessons file grew to 99 lines after 312 logged captures, yet the model still missed rules once it was deep into a task, since timing rather than memory was the failure point. The developer built Tacit, a Claude Code plugin, which triggers on fetched PR review comments and prompts the model to find the root cause and generalise it. Lessons are stored in two tiers, with general rules and repo-specific ones kept separate, and injected at session start and at moments like pushes and file edits. Triggers are limited to concrete events to avoid the reminders being ignored as background noise.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in