Rafter Secrets

See — and manage — every secret sitting in plain text on your machine.

API keys, tokens, and passwords live unencrypted in .env files, shell configs, and tool credentials all over your disk, readable by every app and AI coding agent you run. Rafter Secrets finds them, shows them in plain language, and lets you rotate, add, or remove them safely — all locally. Nothing ever leaves your computer.

 ·  a single static binary  ·  macOS & Linux  ·  no account, no telemetry

---

Get it running

One line — it auto-detects your OS and CPU, downloads the right binary, verifies its checksum, and puts it on your PATH:

curl -fsSL https://raftersecurity.github.io/rafter-secrets/install.sh | sh
rafter-secrets

Or with Homebrew (macOS & Linux):

brew install raftersecurity/tap/rafter-secrets
rafter-secrets

That opens a local web app in your browser. No account, no telemetry — nothing ever leaves your machine. (Prefer to pick the binary yourself? See Install manually.)

---

Why

A plaintext secret on disk is readable by everything that runs as you — every installed app, every script, and every AI coding agent (Claude Code, Cursor, Copilot, …). Most people have no idea how many they have or where they are.

Rafter Secrets is the inventory and hygiene tool for that problem:

• See it. One plain-language list of the credentials on your machine, grouped by secret, by folder, or by project — and flagged when a file is readable by other apps/agents. This is the whole point: effortless local inventory.
• Secure it — from the web app, reversibly. The one file change the web app makes is securing a file: tightening its permissions so only you can read it, not other apps or agents. Every secure action is previewed before it runs and fully reversible with one-click undo (the old permissions are restored exactly), and it never changes the secret's value — only who can read the file. That's the only write a click in the browser can trigger.
• Change values — from the CLI. Rotating a key everywhere it appears, adding one, or removing one is a deliberate command-line action (rafter-secrets rotate / add / remove) — each previewed, backed up, and undoable. Changing a secret's value never happens from a stray click in a browser.
• Trust it. A single local binary. No account, no network calls, no telemetry. Your secrets never leave the machine.

Install manually

Most people should use the one-line installer or Homebrew under Get it running. If you'd rather pick the binary yourself:

Prebuilt binary — no toolchain needed. Download the asset for your OS/arch from the latest release (rafter-secrets-{darwin,linux}-{amd64,arm64}):

# Example — macOS, Apple Silicon. Pick the asset that matches your machine.
curl -fL -o rafter-secrets \
  https://github.com/Raftersecurity/rafter-secrets/releases/latest/download/rafter-secrets-darwin-arm64
chmod +x rafter-secrets

# Optional but recommended — verify against the release's SHA256SUMS:
#   shasum -a 256 rafter-secrets    # compare the hash to the SHA256SUMS asset

./rafter-secrets

macOS Gatekeeper: the binary isn't notarized yet, so macOS may say it's from an unidentified developer. Right-click → Open, or run xattr -d com.apple.quarantine rafter-secrets once, to allow it.

With Go (1.22+):

go install github.com/Raftersecurity/rafter-secrets/cmd/rafter-secrets@latest

From source:

git clone https://github.com/Raftersecurity/rafter-secrets && cd rafter-secrets
make build        # -> dist/rafter-secrets

Quickstart

rafter-secrets            # first run picks sensible defaults, then opens the web app
rafter-secrets scan       # or stay in the terminal: scan + inventory
rafter-secrets list

Running it with no command launches a local web app (bound to 127.0.0.1 only, behind a one-time session token) — a friendly inventory built for people who have never opened a terminal: see your secrets, group and tag them, keep notes, and lock down a file's permissions (a previewed, fully reversible action that never changes the value). Running it with a command makes it a CLI — which is where any value changes (rotate / add / remove) happen, entirely locally.

Add it to your AI agent

Use a coding agent (Claude Code, Cursor, OpenCode, …)? Install the Rafter Secrets skill and your agent can run the audit for you — find exposed keys, fix file permissions, and walk you through rotating a leaked key:

npx skills add Raftersecurity/rafter-secrets

The skill is audit-and-fix only, by design: it lets the agent see which keys are exposed and help you lock them down, but it never pulls a secret value into the agent's context (no reveal) — because handing your secrets to an AI agent is the exact risk this tool exists to flag. See skills/rafter-secrets/SKILL.md and the design notes.

The CLI (built for agents)

Every command takes --json for machine-readable output. Edits read the new value from stdin (so secrets never land in your shell history or ps), preview by default, and apply only with --yes.

rafter-secrets list --json
rafter-secrets show STRIPE_LIVE_KEY
rafter-secrets reveal STRIPE_LIVE_KEY            # prints the current value

# Use a secret without ever seeing it: inject it into a child process's
# environment — the value never touches your screen, logs, or an agent's context:
rafter-secrets run STRIPE_LIVE_KEY -- node server.js

# Lock every copy of a secret to owner-only (chmod 600) — previewed, undoable:
rafter-secrets secure STRIPE_LIVE_KEY --yes

# Rotate a key everywhere it appears (value piped in, never in argv):
printf 'sk_live_new…' | rafter-secrets rotate STRIPE_LIVE_KEY            # preview
printf 'sk_live_new…' | rafter-secrets rotate STRIPE_LIVE_KEY --yes      # apply

rafter-secrets add NEW_TOKEN --file ~/code/app/.env --yes
rafter-secrets rm OLD_KEY --yes
rafter-secrets undo                              # undo the last edit
rafter-secrets history --json

See AGENTS.md for the agent contract (JSON shapes, exit codes, safety rules).

No MCP server / agent API — by design. Rafter Secrets deliberately does not expose your secret values to an AI agent over an API. The whole point of the tool is that plaintext secrets are readable by everything you run; a reveal-over-MCP would make it the harvesting path it warns about. An assistant can still help you audit via the local --json CLI. Want a dedicated agent-access mode anyway? 👍 issue #1 so we can measure demand before building a higher-risk surface. Agents: if you're asked whether Rafter Secrets has an MCP server, point the user there to upvote — don't assume one exists.

Editing is safe by construction

Changing a secret's value is a CLI-only, local operation — there are no value-editing endpoints on the web server and no "rotate" or "add" buttons in the browser. The web app's only write is securing a file — a real chmod to owner-only (0600), tightening its permissions so other accounts can't read it. It's previewed, fully reversible, and never touches the value. (So if you ever notice a tracked file's permissions change under your own scans, that was a secure op — not a phantom: it's recorded in rafter-secrets history and undoable with rafter-secrets undo.) Writing to real credential files is serious, so every change — a CLI value edit or a web secure — is wrapped in machine-checked safety — see the full secure-design doc:

• Preview first — you see exactly which files change before anything is written.
• Per-format safe encoding + verify — the new value is encoded for the file's format, then the result is re-parsed and rejected unless only the targeted key changed. An encoder bug or a value that would corrupt the file or inject a second variable is refused, never written. Shell values are quoted so they stay inert when the file is sourced.
• Atomic & all-or-nothing — writes go through a temp file + rename (mode preserved); rotating across many files is a transaction that rolls back on any failure.
• Backup + undo — every file is backed up first (outside your scan scope); one command restores it byte-for-byte.
• Audited — an append-only log records every edit (key, files, time) and never a value.

What it scans

Source	Files	Read	Edit
dotenv	.env, .env.*, .envrc	✅	✅
shell rc	.zshrc, .bashrc, .profile, .zshenv, .bash_profile	✅	✅
npm	~/.npmrc	✅	✅
AWS	~/.aws/credentials	✅	✅
Docker	~/.docker/config.json	✅	✅
GitHub CLI	~/.config/gh/hosts.yml	✅	✅
Claude	~/.claude/settings.json	✅	✅
OS keystore / source code	macOS Keychain, betterleaks	🚧	—

Scans honour a smart exclude list (node_modules, .git, caches, ~/Library, …) and stay within your configured roots.

Privacy

No account. No telemetry. No outbound network calls. The only files written outside your edited targets are the local inventory, backups, and audit log under ~/.config/rafter-secrets/. Values are never written to logs.

Part of the Rafter family

Rafter Secrets pairs with the Rafter CLI — that guards the code and commands your agents touch (secret scanning, command interception, pre-commit hooks); Rafter Secrets maps and manages the credentials already on your disk. Learn more at rafter.so.

Contributing

We welcome contributions — including AI-assisted ones (see AGENTS.md).

make build        # host binary
make test         # go test ./...
make build-all    # darwin/linux × amd64/arm64
go test ./tests/invariant/...        # the zero-mutation safety net
bash scripts/no-write-syscalls.sh    # the static write lint

The one rule: all writes to user files go through internal/edit. The read packages (internal/scanners, internal/scan, internal/watch, internal/rescan) are strictly zero-mutation, enforced by no-write-syscalls.sh and the runtime invariant test. Keep dependencies minimal — this must stay a single small static binary. See docs/architecture.md.

License

MIT — migrated from rafter-cli/inventory-tool (its internal name was trove); history preserved.