@solongate/proxy 0.57.0 → 0.59.0

package/README.md

@@ -1,68 +1,78 @@
-# @solongate/proxy
-
-> **Don't `npm i` this.** There's nothing to install. Run the two commands below to pair your machine:
-> ```bash
-> npx -y @solongate/proxy@latest
-> npx -y @solongate/proxy@latest login
-> ```
-
-**Security for AI agents.** SolonGate enforces a security policy on every action an AI agent takes — every tool call is allowed, blocked, or logged before it runs. No code changes.
-
-```
-AI agent ──(tool call)──> SolonGate guard ──> Tool runs
-                               │
-                          [policy eval]   ← OPA-WASM, fail-closed
-                          [prompt-injection scan]
-                          [allow / block / audit]
-```
-
-SolonGate is fully managed — nothing to host. Pair your machine with one command and manage policies + audit logs at [dashboard.solongate.com](https://dashboard.solongate.com).
-
----
-
-## Quick start
-
-**You need:** a free [SolonGate account](https://auth.solongate.com), Node.js 18+ on the machine you want to protect, and an AI tool that makes tool calls (Claude Code; Gemini CLI is also supported).
-
-Pair the machine and turn on the guard — no API keys to copy. First fetch SolonGate, then log in:
-
-```bash
-npx -y @solongate/proxy@latest
-npx -y @solongate/proxy@latest login
-```
-
-It opens your browser to authorize the device. Approve it from your signed-in dashboard and SolonGate installs a **global guard hook** that intercepts every tool call from every AI session on the machine and evaluates it against your active policy.
-
-> Start a **new** terminal session afterwards — hooks load when a session starts, so already-open terminals aren't guarded yet.
-
-## Write a policy
-
-Open **Policies** in the dashboard and create one. A policy is a set of rules; each rule targets a tool plus a constraint (path / command / filename / URL) and is either ALLOW or DENY. Two modes:
-
-- **Denylist** — everything is allowed except what you block (e.g. deny any file named `*.env`, or commands matching `*rm -rf*`).
-- **Whitelist** — nothing is allowed except what you permit.
-
-Click **Activate** to make a policy the one the guard enforces. Rules compile to an OPA-WASM bundle and take effect on the next tool call.
-
-## Review the audit log
-
-Every decision shows up under **Audit** — what was allowed, what was blocked, and which rule matched. When a legitimate action is blocked, open the entry and click **Whitelist this**: SolonGate adds a narrow ALLOW exception for exactly that action and recompiles. Revoke any exception from **Granted exceptions**.
-
-## Uninstall
-
-```bash
-npx @solongate/proxy logout
-```
-
----
-
-## Why SolonGate?
-
-AI agents get direct access to your system — shell, file system, databases, network. A single prompt-injection attack can turn your assistant into an attacker. SolonGate sits between the agent and the action, enforcing your policy on every tool call before it runs, and recording every decision for audit.
-
-- **Docs:** [solongate.com/docs](https://solongate.com/docs)
-- **Dashboard:** [dashboard.solongate.com](https://dashboard.solongate.com)
-
-## License
-
-MIT
+# @solongate/proxy
+
+> **Don't `npm i` this.** There's nothing to import. Run the two commands below to pair your machine:
+> ```bash
+> npx -y @solongate/proxy@latest
+> npx -y @solongate/proxy@latest login
+> ```
+
+**Security for AI agents.** SolonGate checks every action an AI agent takes. Each tool call (shell command, file read or write, network request) is allowed, blocked, or logged before it runs, based on a policy you control. No code changes.
+
+```
+AI agent  ──(tool call)──>  SolonGate guard  ──>  Tool runs
+                                  │
+                             [ policy check ]
+                             [ allow / block / log ]
+```
+
+SolonGate is fully managed. There is nothing to host: pair your machine with one command, then manage policies and review logs at [dashboard.solongate.com](https://dashboard.solongate.com).
+
+## Quick start
+
+**You need:** a free [SolonGate account](https://auth.solongate.com), Node.js 18+ on the machine you want to protect, and an AI tool that makes tool calls (Claude Code; Gemini CLI is also supported).
+
+Pair the machine and turn on the guard. There are no API keys to copy. First fetch SolonGate, then log in:
+
+```bash
+npx -y @solongate/proxy@latest
+npx -y @solongate/proxy@latest login
+```
+
+This opens your browser to authorize the device. Approve it from your signed-in dashboard and SolonGate installs a global guard hook that intercepts every tool call from every AI session on the machine and checks it against your active policy.
+
+> Start a **new** terminal session afterwards. Hooks load when a session starts, so already-open terminals aren't guarded yet.
+
+## Write a policy
+
+Open **Policies** in the dashboard and create one. A policy is a set of rules. Each rule targets a tool plus a constraint (path, command, filename, or URL) and is either ALLOW or DENY. Two modes:
+
+- **Denylist:** everything is allowed except what you block (for example, deny any file named `*.env`, or commands matching `*rm -rf*`).
+- **Whitelist:** nothing is allowed except what you permit.
+
+Click **Activate** to make a policy the one the guard enforces. It takes effect on the next tool call.
+
+## What SolonGate can enforce
+
+- **Policy rules:** allow or block tool calls by path, command, filename, or URL.
+- **Ghost paths:** make chosen files and folders invisible to the agent. They don't show up in listings and can't be read.
+- **Data loss prevention (DLP):** if a call contains a secret (API key, token, private key), block it, or hide the secret in the output the model sees.
+- **Rate limiting:** cap how many tool calls an agent can make per minute, hour, or day.
+
+## Review the audit log
+
+Every decision shows up under **Audit**: what was allowed, what was blocked, and which rule matched. Each agent run is also grouped as a **session** with its own timeline. When a legitimate action is blocked, open the entry and click **Whitelist this**: SolonGate adds a narrow ALLOW exception for exactly that action. Revoke any exception from **Granted exceptions**.
+
+## Keep logs on your own machine
+
+Prefer to keep everything local? In **Settings**, turn on **Local log storage** and pick a folder. Every decision is written there as a JSON line, and nothing is sent to SolonGate cloud.
+
+## Get alerted
+
+Set a threshold in **Settings** and SolonGate notifies you by Telegram, email, or webhook when blocks spike.
+
+## Uninstall
+
+```bash
+npx @solongate/proxy logout
+```
+
+## Why SolonGate?
+
+AI agents get direct access to your system: shell, file system, databases, network. SolonGate sits between the agent and the action, enforcing your policy on every tool call before it runs, and recording every decision for audit.
+
+- **Docs:** [solongate.com/docs](https://solongate.com/docs)
+- **Dashboard:** [dashboard.solongate.com](https://dashboard.solongate.com)
+
+## License
+
+MIT

package/dist/global-install.js

@@ -232,7 +232,7 @@ async function runGlobalInstall(opts = {}) {
   writeFileSync(join(p.hooksDir, "stop.mjs"), readHook("stop.mjs"));
   writeFileSync(join(p.hooksDir, "shield.mjs"), readHook("shield.mjs"));
   console.log(`  Installed hooks \u2192 ${p.hooksDir}`);
-  installClaudeShim(join(p.hooksDir, "shield.mjs"));
+  removeClaudeShim();
   writeFileSync(p.configPath, JSON.stringify({ apiKey, apiUrl }, null, 2) + "\n");
   console.log(`  Wrote ${p.configPath}`);
   let existing = {};