Everything Claude Code ships battle-tested agent workflows

This is less a repo than an operating system for agentic development. Its popularity makes sense: once the model is good enough, the bottleneck becomes orchestration, memory, guardrails, and repeatable process.

• –Continuous learning and instinct export/import are the most interesting ideas here because they try to turn one-off sessions into durable team knowledge.
• –Security review via AgentShield gives the project a production bias that most agent stacks never reach.
• –The GitHub App is the strongest moat: turning repo history into team-specific skills and defaults is a much better learning loop than shipping static prompts.
• –Selective install and cross-harness support make it portable across Claude Code, Codex, Cursor, and OpenCode, but teams still need curation so everything does not become noise.