lockjaw open-sources egress credential injection so AI agents never hold real secrets
samgoodwin89 · x · 2026-10-25
joelhooks open-sourced lockjaw (MIT), an egress credential injection system for agents: agents hold placeholders like {{secret:example/key}} instead of real keys. A Durable Object cell authenticates the caller, verifies the approved host and header placement, decrypts the credential, fills the header, sends the HTTPS request, and redacts credential echoes from upstream responses. It's an explicit gateway, not a transparent proxy. It pairs with Rat King, an AT proto agent network for identity/trust; both are Durable Object native and built on alchemyrun and EffectTS. 13 stars on GitHub.
More from coding & agent
- Claude Code patches Flight Simulator 95/98 to run on 64-bit Windows with an 86-byte vectored exception handler fix — Torturephile · 2026-10-25
- Retake: open-source MCP tool lets coding agents record website video walkthroughs — Guilty_Lingonberry · 2026-10-25
- pgrust's rigorous testing has already surfaced multiple real Postgres bugs — cedric_chee · 2026-10-25
- Buddy: open-source Android app runs Claude Code/Codex agents on your phone via local MCP — 3amae404 · 2026-10-25
- Beam 1.0 501B (23B active) nine days in: dev reports strong backend coding — MeinDruckerSpinnt · 2026-10-25
- Dev builds The Waya, an AI agent communication platform prototype, entirely from a phone — Glittering-Produce19 · 2026-10-25