Bring the agents your team already built—or start from ours. Compile them into one governed model, render native files for Claude Code and GitHub Copilot, and enforce policy where the work happens.
Custom agents live in provider-specific folders. Claude remembers one chat. Copilot remembers another. Teams accumulate valuable behavior, but no portable model, review boundary, or consistent enforcement layer.
Most platforms ask you to start over. NEPOPSX comprehends the agents already living in your repos, turns them into a provider-neutral governed model, and lets your team review before anything is republished.
Read Claude Code and GitHub Copilot agents directly from the repository.
Preserve orchestration, flag fidelity loss, dedupe against the catalog, and quarantine threats.
Bless the canonical agent, emit native provider files, and ship local PreToolUse policy enforcement.
Dry run is local, loss-aware, and safe to run in an existing repository.
nepopsx import-agents --dry-runNEPOPSX ships specialized, role-based agents that share one governed memory — so every task starts with your team's context, not a blank slate.
From idea to shipped code — a pipeline of specialized sub-agents, each enforcing your team's approved standards as the work moves through it.
A personalized generalist that recalls before it builds — and verifies before it's done.
Keeps documentation honest — catches drift and captures decisions as they happen.
Plans, executes, and reflects — validating changes against your team's quality bar.
NEPOPSX sits between your team and your AI coding tools.
Native agent files generated for every provider — no manual translation.
Automatic — captured by session-end hooks, then typed and scoped to the right agent, service, or workspace. No one has to remember to write it down.
Confidence scoring and a security scan decide what auto-approves — everything risky queues for human review, with an append-only audit trail.
At task start, your agent queries the workspace's approved knowledge over MCP and pulls back only what's been reviewed. Not chat history — governed team knowledge.
Every captured learning is scored for confidence and scanned for secrets. High-confidence, low-risk ones auto-promote. Everything else waits for a human — and nothing is ever silently rewritten.
No new editor and no new model. Import or install, sync, and keep using Claude Code and GitHub Copilot—while the same governed model stays ready for whatever provider comes next.
Approved learnings automatically flow back into every developer's agents.
Not documentation. Not a wiki nobody reads. Behavior — enforced by the tools your team already uses.
Onboarding stops meaning "read the wiki." It means "start building, fully equipped."
Roll out one governed standard across every repo and every developer. New hires inherit it on day one.
Configure once. Learn continuously. Work everywhere.