dsh-codevault

The most frustrating moment in reading open-source code is not the initial confusion but the total amnesia three days later. dsh-codevault closes that forgetting loop: after you understand a mechanism during a chat session, a single natural-language prompt triggers a record, and the plugin archives it as a knowledge card indexed by repository, file, and symbol. Re-reading the same object appends to that card's timeline rather than spawning a new entry, so the card grows organically with each encounter and the top-level readCount increments while earlier points are deduplicated and merged.

Cards are written as Markdown plus JSONL with auto-generated [[wikilinks]], so they drop straight into an Obsidian vault and the graph view works out of the box. Once you point the plugin at a vault directory, the model can read-only search and surface related notes you already have; after your confirmation, structured links are woven into the card to connect code understanding with project documentation. Because the underlying log is strictly append-only and the card is a generated view, read_history can replay any prior version precisely, and read_delete lets you remove an event and let the card rebuild correctly with no extra versioning storage. Each record can carry two or three revisit questions that prompt active recall the next time you return to the card.

Everything runs locally: the plugin never reads your repos, calls external APIs, or touches the network. It does not score code, does not invent fields it has not verified, and every destructive operation requires explicit confirmation. It is built for developers who read open-source code closely and want a lightweight, linked, replayable personal archive of what they understood, especially those already managing knowledge in Obsidian and looking to weave code-reading notes into their existing documentation.