AI coding agents read plaintext .env files and silently expose API keys and credentials to LLM context
Claude Code, Cursor, and Copilot automatically read project files including .env during sessions, loading API keys, tokens, and passwords into LLM context. GitGuardian's 2026 report: 28.65M secrets leaked on GitHub in 2025, AI-service leaks up 81%. Infisical launched Agent Vault (156 HN pts, April 2026) to proxy credentials so agents never see raw secrets. Developers are actively looking for CLI-native solutions.
A CLI proxy that redacts .env secrets before AI coding agents read project files, so credentials never enter LLM context
2.0k ▲Score Breakdown
Social Proof 3 sources
Existing Solutions 3 competitors
Open-source HTTP credential proxy and vault for AI agents. Agents never see raw secrets; vault proxies outbound API calls.
Encrypted .env files using public-key cryptography (ECIES, AES-256). Commit encrypted secrets to git; agents cannot read without private key.
Pre-commit hook that detects secrets before they land in repos. Extended to intercept AI coding tool context in 2026.
Gap Assessment
Agent Vault is early-stage research preview. dotenvx (5.5k stars) covers encrypted .env but not agent-context interception. No CLI-native zero-trust secrets proxy owns this yet