adithyan-ak

Agentmask — AI skill for Claude Code

AI community

Mask your secrets from AI coding agents.

How to install Agentmask

This entry records only its repository, not the path inside it, so there is no exact command to give. Open adithyan-ak/agentmask and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agentmask does

Mask your secrets from AI coding agents. Full context. Zero exposure.

Alternatives in AI

Empryo — Empryo issue tracker + SoulForge (v2) 1.1k ★
Clawcodex — Token efficient Claude Code full Python rebuild 875 ★
Memtrace Public — Structural memory for AI coding agents 466 ★

README

agentmask

Secrets never enter context. AI never misses a beat.

agentmask demo

agentmask prevents Claude Code, Cursor, and other AI coding assistants from reading, leaking, or committing your secrets. It works through three reinforcing layers:

Block — Hooks that prevent secret files from being read and secret values from being written
Redirect — An MCP server that provides redacted file access so the agent can still work
Instruct — Behavioral rules that teach the agent to prefer safe alternatives

Quickstart

npm install -g agentmask
cd your-project
agentmask init

`init` scans your entire repository for secrets, builds a blocklist of every file containing them, and installs hooks + MCP server + behavioral rules for all detected IDEs. Secrets are blocked before they ever enter the AI's context.

agentmask auto-detects which IDEs are present (Claude Code, Cursor) and configures both. Use `--claude` or `--cursor` to target a specific IDE.

What It Does

Scenario	What Happens
Claude tries to `Read .env`	Blocked. Static pattern match. Redirected to `safe_read`.
Claude tries to read `src/config.ts` (has hardcoded AWS key)	Blocked. Found by init scan, in blocklist. Redirected to `safe_read`.
Claude writes `sk_live_...` into source code	Blocked. Content scan catches it. Told to use env var instead.
Claude runs `cat .env` via Bash	Blocked. Command pattern match.
Claude runs `git commit` with secrets in staged files	Blocked. Pre-commit scan. Shown file:line of each secret.
Claude reads a new file with a secret (not yet in blocklist)	Warned + auto-blocklisted. First read leaks, every subsequent read is blocked.
Claude does normal coding (90%+ of operations)	No effect. Sub-50ms hook, completely invisible