agentic-stack

Why we included this project

Anyone who runs coding agents in more than one tool has hit the same wall: Claude Code, Cursor, Windsurf, and the rest each keep their own memory and skills, so switching tools means your agent starts over. agentic-stack fixes that with a portable .agent/ folder that holds memory, skills, and loop protocols, and any supported harness can read it, so the knowledge your agent accumulates follows it between tools. It also includes a local data layer that pulls harness activity, cron runs, token and cost estimates, and KPI summaries into one dashboard, plus a flywheel that turns approved, redacted runs into trace records, eval cases, and training-ready JSONL without sending telemetry. If you manage several coding agents and want them to learn once and behave consistently, this gives you a single layer to standardize on.