July 15, 2026
Hook, line, and comment war
Show HN: Capn-hook for coding agents – don't grep the same mystery twice
This coding memory tool promises less repeat work, and the comments instantly got messy
TLDR: cap'n hook is a new tool that helps coding AIs remember where things are in a project, aiming to cut repeat work and save money. Commenters were split between curiosity and nitpicking—some questioned why old memories get deleted instead of updated, while others immediately derailed into cereal jokes.
A new tool called cap'n hook is pitching a very relatable dream: stop making coding assistants re-learn the same answer every single time. The idea is simple enough for non-coders too—if an AI helper spends ages figuring out where something lives in a project, cap'n hook saves that trail so next time it can jump straight there. The maker says this cut repeat waste by 77% in tests, and if the code changes, the saved answer gets tossed out automatically so the AI doesn’t cling to old info like a clueless ex.
But the real action? The comments, of course. One of the first reactions was basically, “Cool, but why delete the memory—why not just update it?” That’s the practical, slightly skeptical camp: people like the idea, but they’re already poking at whether the tool is being too dramatic about throwing old answers overboard. Another commenter went straight for the performance anxiety angle, asking about speed and how much local computing power this thing gulps down. Translation: sure, saving time sounds great, but not if your laptop starts wheezing.
And then came the internet’s favorite coping mechanism: jokes. One user instantly turned “capn” into breakfast comedy with “capn crunch,” while another spun up a finance pun about “chips n dips.” So the mood is classic tech-launch energy: part impressed, part suspicious, and part unable to resist turning the whole thing into a cereal meme. In other words, a very healthy launch.
Key Points
- •Capn-hook is a persistent-memory CLI for coding agents that stores file-based answers to codebase questions across sessions.
- •The article reports a 77% token reduction on repeat questions across 60 developer questions on 5 production codebases, with correct answers in both evaluated conditions.
- •Its workflow consists of asking for a saved answer before searching, charting newly discovered answers, and automatically deleting entries when referenced files change or disappear.
- •Integration is implemented through session-start hooks for Claude Code and Codex using a short context instruction rather than a wrapper or middleware.
- •The tool supports hybrid semantic recall through QMD by default, optional BM25 keyword search without embeddings, and installation via npm or Bun with Node.js fallback.