# treg
**treg is a credential and skill vault for teams where some of the members are agents.** Keys live in the vault and never land on anyone's machine; a proxy injects the real credential server-side when a call goes out.
## How it works
1. **Scan and upload** — sweep your projects for keys and skills, encrypt, store centrally
2. **Call** — a teammate or an agent runs `treg call` holding only their own personal token
3. **Inject** — the proxy swaps that token for the org's real credential at the boundary
4. **Audit** — every call is attributed to a specific person or agent, never masked behind a shared key
Per-user tokens are revocable individually. Rotation happens in the vault with no downtime and no chasing `.env` files across laptops.
## Why it matters for agents specifically
The reason to care isn't key hygiene in general, it's that giving an agent a live API key is the standard way to sandbox nothing. You isolate the filesystem and the process, then hand the container a working credential because otherwise it can't do its job. Prompt injection or a bad dependency and the key walks out.
treg is the same shape of answer as [[Hermes Agent]]'s credential firewall and [[varlock]]: **stand-in token inside, real secret swapped in at the network edge.** Three independent projects converging on that pattern in the same year is a decent signal it's the right one.
The distinctive bit here is scope. Hermes solves it for its own sandboxes; treg solves it for a whole team, and treats *skills* as a shared, credentialed resource alongside APIs — so you can hand a teammate's agent a capability without handing it the secret behind that capability.
Free tier to start.
## Related
- [[varlock]] — same credential-proxy pattern
- [[Hermes Agent]] — credential firewall for Docker sandboxes
- [[OpenSandbox Credential Vault]]
- [[Secret Operations (SOPS)]]
- [[scan-for-secrets]]
- [[AI Agents]]
- [[AI Agent Skills]]
## References
- Website: https://treg.superdesign.dev/