Citadel

Why we included this project

Citadel is a plugin for Claude Code and OpenAI Codex that keeps project state inside the repository, so a long agent run can resume where an earlier session left off. Instead of rebuilding context with every fresh prompt, the agent stores its context, decisions, evidence, and handoffs in the repo itself; Citadel routes natural-language requests into repeatable workflows, applies safety hooks and approval boundaries, and records cost telemetry plus reviewable evidence of what changed. It also coordinates parallel agent fleets. Developers and small teams already using an agentic coding tool daily will recognize the problem it solves: single-shot prompts are too fragile, and a bounded, removable layer beats wiring up a heavier orchestration platform. Installation runs through each runtime's built-in plugin market, needs no npm package, and activates one project at a time, so it is easy to trial on a single repository before adopting it widely.