agentic-stack makes `.agent/` memory portable across tools
agentic-stack packages shared memory, skills, and protocols into a portable `.agent/` folder that can be dropped into different coding harnesses and carry project conventions with it. The repo positions itself as a “one brain, many harnesses” layer for Claude Code, Cursor, Windsurf, OpenCode, OpenClaw, Hermes, Pi Coding Agent, and a DIY Python loop, with an onboarding wizard that seeds preferences and feature toggles into the project. It is notable less as a new model or framework and more as an interoperability play for preserving context, review rules, and workflow norms across tools.
Hot take: this is a standards move disguised as a utility repo. If it works as advertised, the value is not the folder itself but the fact that it makes agent behavior portable across otherwise incompatible harnesses.
- –The strongest idea is portability: the same memory and conventions can follow you between tools instead of being trapped in one vendor’s config.
- –The repo is opinionated about memory structure, with separate working, episodic, semantic, and personal layers, plus a review protocol for promoting lessons.
- –The onboarding flow matters because it turns “context” into explicit project policy, not ad hoc chat history.
- –The project is most compelling for teams who already bounce between multiple AI coding environments and want consistency in reviews, commit style, and test strategy.
- –Risk: portability standards are only useful if enough tools adopt them, so this lives or dies on ecosystem support and how easy the adapters are to maintain.
DISCOVERED
3h ago
2026-04-20
PUBLISHED
3h ago
2026-04-20
RELEVANCE
AUTHOR
Github Awesome