Eighth in the series is reporecall, a plugin from proofofwork-agency built around a problem anyone using Claude Code on a large codebase will recognise: every new session starts from zero, re-reading files and rebuilding context that a previous session already had.
How it builds context
Reporecall indexes the codebase using tree-sitter AST parsing across 22 languages, combined with call-graph traversal so it understands which functions call which. On top of that it runs a hybrid search – keyword matching plus vector embeddings – so a query can match on exact function names as well as on conceptual similarity.
The part that makes it feel different from a plain search index is the timing: it injects relevant context in roughly 5 milliseconds before Claude starts reasoning about a task, rather than waiting for Claude to go looking for it. It’s available both as Claude Code hooks and as an MCP server, so it can slot into whichever integration point fits your setup.
Local codebase memory, not project management
Worth being precise about scope here: this is memory of the code itself – structure, call graphs, relevant snippets – not a task tracker or a record of past conversations. It’s a complement to something like Align (post four) rather than a replacement for it.
Installing it
It ships as an npm package:
npm install -g @proofofwork-agency/reporecallFrom there it can be wired in as either a Claude Code hook or an MCP server, depending on which you’d rather manage.
Next Sunday: private-journal-mcp, obra’s semantic-search journaling server – a different kind of memory, this time for what the agent itself has noticed rather than what’s in the code.
Leave a Reply
You must be logged in to post a comment.