registry / AI-agent capability abuse / @solongate/proxy / 0.89.13

@solongate/proxy@0.89.13

AI tool security proxy: protect any AI tool server with customizable policies, path/command constraints, rate limiting, and audit logging. No code changes required.

AI Security Review

scanned 2mo ago · by lpm-firewall-ai

This is not malicious by source evidence, but it is a high-impact explicit-user AI-agent extension that changes client hook configuration and uploads configured audit data. Warn rather than block because no unconsented npm install-time mutation or concrete malicious chain was found.

Static reason

High-risk behavior combination matched malicious policy.; source matched previously finalized malicious package; routed for review

Trigger

A user runs the interactive SolonGate setup or repair flow and configures a live SolonGate account.

Impact

The installed hooks can inspect or block AI tool calls and send their audit data to the configured vendor service.

Mechanism

Registers AI-client hooks, evaluates policies, and sends configured audit events.

AI rationale

Evidence

package.jsondist/global-install.jsdist/index.jshooks/guard.mjs~/.codex/hooks.json~/.config/opencode/plugins/solongate.js~/.solongate

Network endpoints2

api.solongate.com/api/v1/audit-logsapi.solongate.com/api/v1/policies/

Decision evidence

public snapshot

AI called this Suspicious at 90.0% confidence as Dangerous Capability with low false-positive risk.

Evidence for warning

Explicit setup writes command hooks into the user's Codex hook configuration.
With a live SolonGate key, the proxy uploads audit entries to the SolonGate API.
The guard downloads and caches policy WebAssembly from the configured service.

Evidence against

package.json contains no preinstall, install, or postinstall lifecycle hook.
The CLI refuses management actions outside an interactive human terminal.
The flagged file reads outbound file arguments to detect and block secret transfers, not to upload their contents.
Network traffic is part of the configured security proxy and uses the vendor API endpoint.

Behavioral surface

Source

ChildProcessCryptoDynamicRequireEnvironmentVarsEvalFilesystemNetworkShell

Supply chain

HighEntropyStringsPossibleObfuscationUrlStrings

Source & flagged code

19 flagged

dist/tui/index.jsView file

Lines 1-33javascript

1var __defProp = Object.defineProperty;

2var __getOwnPropNames = Object.getOwnPropertyNames;

3var __esm = (fn, res) => function __init() {

4 return fn && (res = (0, fn[__getOwnPropNames(fn)[0]])(fn = 0)), res;

5};

6var __export = (target, all) => {

7 for (var name in all)

8 __defProp(target, name, { get: all[name], enumerable: true });

9};

11// src/hook-launcher.ts

12function launcherScript(pinnedNode) {

13 const pinned = pinnedNode.replace(/'/g, `'\\''`);

14 return `#!/bin/sh

15# SolonGate hook launcher \u2014 generated by \`solongate init --global\` / \`repair\`.

High

Sandbox Evasion Gated Capability

Source gates dangerous network, credential, or execution behavior behind CI, host, platform, time, or geo fingerprint checks.

dist/tui/index.jsView on unpkg · L13

Medium

Install Persistence

Source writes installer persistence such as shell profile or service configuration.

dist/tui/index.jsView on unpkg · L13

16#

17# Finds a working node and execs the hook with it. Do not edit: a reinstall

18# overwrites this file, and on macOS it is chflags-locked besides.

19#

20# The reason it exists rather than the hook naming a node directly: an absolute

21# node path recorded at install time is a Homebrew Cellar path or an nvm version

22# directory, and both are deleted by a routine upgrade. The hook then could not

23# start, nothing was enforced, nothing was logged, and every check still said

24# the guard was registered \u2014 because it was.

26set -u

28script="\${1:-}"

29[ -n "$script" ] || { echo "solongate: launcher called with no hook script" >&2; exit 1; }

30shift

32hook=\${script##*/}

33home=\${HOME:-~}

Lines 123-163javascript

123 ];

124 NODE_GLOBS = [

125 // nvm. NVM_DIR is exported by its own shell hook, which a non-interactive

126 // hook environment does not run, so the default location is tried too.

127 '"${NVM_DIR:-}"/versions/node/*/bin/node',

128 '"$home"/.nvm/versions/node/*/bin/node',

129 // fnm, both the XDG location and the macOS Application Support one.

130

Lines 877-917javascript

877 } catch {

878 }

879 try {

880 removeOpencodeGuard(p);

881 } catch {

882 }

883 if (!existsSync3(p.settingsPath)) return { ok: true

13Cross-file remote execution chain: dist/tui/index.js spawns dist/index.js; helper contains network access plus dynamic code execution. L13: const pinned = pinnedNode.replace(/'/g, `'\\''`); L14: return `#!/bin/sh L15: # SolonGate hook launcher \u2014 generated by \`solongate init --global\` / \`repair\`. ... L42: # SHELL, before the command runs, so a \`2>/dev/null\` on the printf alone does L43: # not suppress it \u2014 it lands on the hook's stderr, and Claude Code shows a L44: # hook's stderr to the person using it. Redirecting the group catches both. ... L143: // src/hook-health.ts L144: import { execFileSync } from "child_process"; L145: import { existsSync as existsSync2, readFileSync as readFileSync3, statSync as statSync2 } from "fs"; ... L149: const launcher = join3(hooksDir(), LA

High

Cross File Remote Execution Context

Source spawns a local helper that also contains network and dynamic execution context; review data flow before blocking.

dist/tui/index.jsView on unpkg · L13

dist/audit/index.jsView file

Lines 1507-1547javascript

1507 title,

1508 status: "PARTIAL",

1509 evidence,

1510 summary: `${totalIssues} risky package install(s) \u2014 low frequency.`,

1511 details: "Some packages installed without pinned versions, but at low frequency. Review and pin specific versions.",

dist/cli-launch.jsView file

Lines 21-61javascript

21 "doctor",

22 "watch",

23 "alerts",

24 "webhooks",

25 "dataroom"

26 // The `browser` command is gone on purpose: the agent installs, arranges and

27 // restarts itself, and everything a person manages is on the dashboard. There

hooks/shield.mjsView file

Lines 266-299javascript

266 if (!res.headersSent) res.writeHead(502, { 'content-type': 'text/plain' });

267 res.end('shield upstream error');

268 });

269 upReq.end(body);

270

hooks/guard.mjsView file

Lines 1-33javascript

1#!/usr/bin/env node

2/**

3 * SolonGate Cloud Policy Guard Hook (PreToolUse) — GLOBAL system-wide enforcement.

4 *

5 * This is the cloud twin of the air-gapped guard hook. Identical decision engine

6 * (OPA WASM, NIST SP 800-207 PDP, fail-closed), but the policy + compiled WASM

7 * are fetched from SolonGate Cloud and authenticated with the project API key.

8 * Installed globally (~/.claude/settings.json) it intercepts EVERY tool call from

9 * EVERY Claude Code session on the machine — exactly like the air-gapped product,

10 * just sourced from the cloud instead of a local docker API.

dist/index.jsView file

•Manifest-reachable source resolves another installed package, overwrites its runtime code, and injects package-defined remote behavior. dist/index.js: // ../../node_modules/.pnpm/@open-policy-agent+opa-wasm@1.10.0/node_modules/@open-policy-agent/opa-wasm/src/builtins/json.js "../../node_modules/.pnpm/@open-policy-agent+opa-wasm@1.10.0/node_modules/@open-policy-agent/opa-wasm/src/builtins/json.js"(exports, module) { // ../../node_modules/.pnpm/sprintf-js@1.1.3/node_modules/sprintf-js/src/sprintf.js "../../node_modules/.pnpm/sprintf-js@1.1.3/node_modules/sprintf-js/src/sprintf.js"(exports) { // ../../node_modules/.pnpm/@open-policy-agent+opa-wasm@1.10.0/node_modules/@open-policy-agent/opa-wasm/src/builtins/strings.js "../../node_modules/.pnpm/@open-policy-agent+opa-wasm@1.10.0/node_modules/@

High

Entrypoint Foreign Package Code Overwrite

Manifest-reachable source overwrites another installed package with package-defined remote behavior.

dist/index.jsView on unpkg

43Trigger-reachable persistence chain: manifest.exports -> dist/index.js L43: try { L44: JSON.parse(str); L45: return true; ... L2983: // Keys in mapping nodes earlier in the sequence override keys specified in L2984: // later mapping nodes. -- http://yaml.org/type/merge.html L2985: addToJSMap(ctx, map) { ... L4516: function shouldWarn(deprecation) { L4517: const env = typeof process !== "undefined" && process.env || {}; L4518: if (deprecation) { ... L5076: handle: "!!", L5077: prefix: "tag:private.yaml.org,2002:" L5078: }]

High

Trigger Reachable Persistence

A manifest entrypoint or package-local install chain reaches persistence behavior.

dist/index.jsView on unpkg · L43

•stage = ast_semantic_analysis; reason = ast_alias_growth_limit_exceeded; limitedFiles = 2

hooks/guard.bundled.mjsView file

•Runtime or CLI code links a write operation to an explicit external AI-agent control path: !== "string" || typeof data.sha256 !== "string") return; const buf = buffer.from(data.content, "base64"); if (createhash("sha256").update(buf).digest("hex") !== data.sha256) return; const text = buf.tostring("utf-8"); if (!text.startswith("#!/usr/bin/env node") || text.length < minlen || !text.includes(marker)) return; const hooksdir = join(resolve(homedir(), ".solongate"), "hooks"); const tmp = join(hooksdir, "." + filename + ".tmp"); writefilesync(tmp, text); try { chmodsync(join(hooksdir, filename), 420); } catch { } renamesync(tmp, join(hooksdir, filename)); } catch { } } function installedhookversion(filename) { try {

High

External Ai Agent Control Surface Mutation

Runtime or CLI source writes behavior-bearing configuration into a user or project AI-agent control surface.

hooks/guard.bundled.mjsView on unpkg

•matchType = normalized_sha256 matchedPackage = @solongate/proxy@0.82.12 matchedPath = hooks/guard.bundled.mjs matchedIdentity = npm:QHNvbG9uZ2F0ZS9wcm94eQ:0.82.12 similarity = 1.000 summary = normalized source hash matched finalized malicious source

High

Known Malware Source Similarity

Source file is highly similar to a previously finalized malicious package; route for source-aware review.

hooks/guard.bundled.mjsView on unpkg

dist/lib.jsView file

42Trigger-reachable credential exfiltration chain: manifest.main -> dist/lib.js L42: try { L43: JSON.parse(str); L44: return true; ... L2982: // Keys in mapping nodes earlier in the sequence override keys specified in L2983: // later mapping nodes. -- http://yaml.org/type/merge.html L2984: addToJSMap(ctx, map) { ... L4515: function shouldWarn(deprecation) { L4516: const env = typeof process !== "undefined" && process.env || {}; L4517: if (deprecation) { ... L5075: handle: "!!", L5076: prefix: "tag:private.yaml.org,2002:" L5077: }]

High

Trigger Reachable Credential Exfiltration

A manifest entrypoint or package-local install chain reaches credential exfiltration behavior.

dist/lib.jsView on unpkg · L42

•Trigger-reachable fixed external POST callback chain: manifest.main -> dist/lib.js arg = JSON.stringify(arg, null, ph.width ? parseInt(ph.width) : 0); if (!ctx) return JSON.stringify(this); return JSON.stringify(jsKey); if (!ctx || !ctx.doc) return JSON.stringify(this); if (!ctx) return JSON.stringify(this); if (!(item instanceof Pair)) throw new Error(`Map items must all be pairs; found ${JSON.stringify(item)} instead`); // later mapping nodes. -- http://yaml.org/type/merge.html const json = JSON.stringify(value);

High

Trigger Reachable External Post Callback

A manifest entrypoint or package-local install chain reaches a fixed external POST callback.

dist/lib.jsView on unpkg

•matchType = normalized_sha256 matchedPackage = @solongate/proxy@0.82.12 matchedPath = dist/lib.js matchedIdentity = npm:QHNvbG9uZ2F0ZS9wcm94eQ:0.82.12 similarity = 1.000 summary = normalized source hash matched finalized malicious source

Source & flagged code

19 flagged · loading source

dist/tui/index.jsView file

143// src/hook-health.ts L144: import { execFileSync } from "child_process"; L145: import { existsSync as existsSync2, readFileSync as readFileSync3, statSync as statSync2 } from "fs";

High

Child Process

Package source references child process execution.

dist/tui/index.jsView on unpkg · L143

897try { L898: const prof = execFileSync2("powershell", ["-NoProfile", "-Command", "$PROFILE.CurrentUserAllHosts"], { encoding: "utf-8" }).trim(); L899: return prof ? [prof] : [];

High

Shell

Package source references shell execution.

dist/tui/index.jsView on unpkg · L897

13const pinned = pinnedNode.replace(/'/g, `'\\''`); L14: return `#!/bin/sh L15: # SolonGate hook launcher \u2014 generated by \`solongate init --global\` / \`repair\`. ... L42: # SHELL, before the command runs, so a \`2>/dev/null\` on the printf alone does L43: # not suppress it \u2014 it lands on the hook's stderr, and Claude Code shows a L44: # hook's stderr to the person using it. Redirecting the group catches both. ... L143: // src/hook-health.ts L144: import { execFileSync } from "child_process"; L145: import { existsSync as existsSync2, readFileSync as readFileSync3, statSync as statSync2 } from "fs"; ... L149: const launcher = join3(hooksDir(), LAUNCHER_NAME); L150: if (process.platform === "win32") { L151: const ok = existsSync2(process.execPath);

Findings

1 Critical16 High5 Medium4 Low

CriticalCredential Exfiltrationhooks/guard.mjs

HighChild Processdist/tui/index.js

HighShelldist/tui/index.js

HighEvaldist/audit/index.js

HighSame File Env Network Executionhooks/shield.mjs

HighEntrypoint Foreign Package Code Overwritedist/index.js

HighSandbox Evasion Gated Capabilitydist/tui/index.js

HighCloud Metadata Accesshooks/guard.mjs

HighExternal Ai Agent Control Surface Mutationhooks/guard.bundled.mjs

HighCross File Remote Execution Contextdist/tui/index.js

HighTrigger Reachable Credential Exfiltrationdist/lib.js

High

Affected versions and remediation

This report applies to @solongate/proxy@0.89.13.

See all versions of @solongate/proxy for other recorded verdicts.

Review the evidence and your use of @solongate/proxy@0.89.13 before allowing it. Restrict the permissions described in this report.
Choose an independently verified alternative or release. This report does not establish that other versions are safe.

Evidence last updated: 2026-08-27 13:02:56.843Z (UTC).

Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Runtime surface

Behavioral surface

Source & flagged code