agentic-stack
A portable memory-and-skills folder that keeps a coding agent's context and conventions consistent across more than a dozen harnesses, plus bounded, budgeted agentic loops and a local dashboard for agent activity and cost.
agentic-stack gives coding agents a single portable folder of memory, skills and protocols that plugs into Claude Code, Cursor, Windsurf, OpenCode, OpenClaw, GitHub Copilot CLI, Gemini CLI, Hermes, Codex, Antigravity and other harnesses, so switching tools does not reset how the agent works.
Bounded agentic loops, introduced in a recent release, run a maker, a deterministic verifier and an independent checker in sequence inside owned git worktrees, with finite attempt, runtime, output and token budgets, deny-path gates and resumable checkpoints; a supervisor bounds and audits the child processes without acting as an operating-system sandbox. A local data layer tracks harness activity, cron runs, active agents, token and cost estimates and generates dashboards, and approved redacted runs can be turned into local trace records, context cards, eval cases and training-ready data without sending telemetry or training a model.
It is written in Python and licensed under Apache-2.0.