Hi Product Hunt 👋 I use Claude Code every day across several client and product repositories, and two things kept costing me time. First, every session starts from zero: the reason I rejected a library last month, the bug that took an afternoon, the convention a client asked for, all gone unless I paste it in again. Second, the agent's defaults are not a careful engineer's habits: it commits without running the right checks, says "fixed" without evidence, reviews its own code and installs packages unasked. developer-os is the plugin I built to fix that for myself: Memory in an Obsidian vault. Before a task the agent searches my notes and cites the note behind every claim. While it works it captures decisions into an inbox. Nothing becomes knowledge until I approve it, and nothing is ever deleted, only archived. Rules in every session, kept under 8 KB: tests before each commit, a plan before large changes, a different agent reviews agent-written code, every task ends with Done / Not done / Needs you. Skills for the daily work: specs, code review, repo audits into a backlog, PR review comments, bug triage, deploy checklists, end-of-session wrap-up. Hooks that ask, never block, before dependency installs, force pushes and git-hook bypasses. What it deliberately is not: there is no server, no database, no account and no telemetry. It is Markdown plus three small Python scripts (standard library, no network), MIT licensed. Install in Claude Code: /plugin marketplace add msolecki/developer-os /plugin install developer-os@developer-os I would love your feedback on two things: which skill you would cut, and what your agent still forgets that a vault note would fix. I will be here all day answering questions.

Developer OS

Long-term memory and engineering rules for Claude Code

Long-term memory and engineering rules for Claude Code