Sealgate — AI skill for Claude Code
Protect sensitive prompt spans with trusted LLM detection and local authenticated encryption before using Claude Code.
How to install Sealgate
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open madeye/sealgate and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Sealgate does
Protect sensitive prompt spans with trusted LLM detection and local authenticated encryption before using Claude Code.
Alternatives in AI
- 09 Query Engine — Prompt 09: Get the QueryEngine (Core LLM Loop) Functional 2.3k ★
- DontFeedTheAI — Transparent anonymization proxy for AI-assisted pentesting 654 ★
- Claude Fable 5 System Prompt Clean — the optimized, token-efficient version of the leaked Claude Fable 5 / Mythos 5 system prompt 475 ★
README
sealgate
**[Read the website and getting-started guide →](https://madeye.github.io/sealgate/)**
Encrypt sensitive text before Claude Code sends it remotely. `sealgate claude` keeps Claude's native terminal interface and subscription login, with a local gateway that inspects complete model requests and an OS sandbox (the macOS sandbox, or Docker on Linux). Linux tools have direct network access outside SEALGATE; macOS blocks other network traffic by default. Your configured **trusted detection provider**, such as local vLLM, identifies sensitive spans; Node encrypts them with a local AES-256-GCM key.
For narrower workflows, `sealgate protect` prints a protected prompt for pasting and `sealgate chat` offers a prompt-only terminal wrapper. The companion plugin supplies a reminder; it cannot intercept requests by itself.
This is conventional authenticated encryption, **not homomorphic inference**. Claude can use the surrounding text but cannot understand encrypted values.
flowchart LR
U[Native Claude and local tools] -->|Isolated local relay| G[SEALGATE gateway]
G -->|Original request text| P[Trusted LLM or local vLLM]
P -->|Sensitive spans| G
K[Local encryption key] --> G
G -->|Protected request and subscription OAuth| A[Anthropic]
A -->|Streamed response| G
G --> U
See the [gateway guide](docs/gateway.md) for the native launcher and network boundary, [how it works](docs/how-it-works.md) for the detection, encryption, and decryption steps, and [the security model](docs/security.md) for trust boundaries and limitations. The instructions below cover installation and configuration.
Install
Requires Node.js 22 or later. The enforced `sealgate claude` launcher additionally requires the native `claude` binary on PATH and either macOS (ARM64 or x86-64, using the built-in `sandbox-exec` sandbox) or Linux ARM64/x86-64 with a local Docker daemon. Manual preprocessing and the older chat wrapper also work on WSL. The project
Related Skills
AI Bridge
Local bridge letting an AI coding agent drive your real, logged-in Chrome — trusted input, CSP-proof eval, aut
Fireraven Agent Hooks
FireGuard hooks for AI agents: protect Cursor, Claude Code, Windsurf, Devin, and Copilot Studio from secret le
Prep Multipov
Verify that the multipov.ai MCP server is configured, authenticated, and responsive in Claude Code. Run before
Trustline
Trusted baseline + auto-rollback for AI agent config files (.claude/, AGENTS.md, .mcp.json). Detects prompt-in
Prismor
Self-hosted runtime control plane for AI agents. Observe or HITL approve or Block rogue tool calls before it e
Aigate
AI Prompt Secret Scanner: local proxy and Claude Code hook that blocks secrets before they reach AI APIs
Related Agents
LLM Orchestrator
Use this agent for LLM integration work — prompt engineering, multi-provider abstraction (OpenRouter, Gemini,
Bh Admin Alice
Beamhall IT operator "Alice" driving the appliance over her own authenticated MCP channel (beamhall-it role).
Security Architecture
Boucle implements a defense-in-depth security model to protect against prompt injection attacks and maintain t