registry / AI-agent capability abuse / @compr/opscontext-mcp / 2.8.3

@compr/opscontext-mcp@2.8.3

OpsContext for AI Agents — read-only fleet visibility (PM2/nginx/Docker/git/cron) + tamper-evident audit log + policy-as-code hooks. The ops + compliance layer Claude Code can't grow natively.

AI Security Review

scanned 12d ago · by lpm-firewall-ai

The explicit Claude hook setup is a concrete agent configuration capability, so this warrants a warning. The package has no npm install lifecycle trigger for that setup, and the inspected sources do not establish a publish-blocking attack.

Static reason

No blocking static signals were detected.; previous stored version diff introduced dangerous source

Trigger

A user invokes the install-claude-hook CLI command.

Impact

The hooks can observe Claude session prompts and tool events and run a session gate.

Mechanism

The command copies a hook script into the user's Claude hooks directory and adds hook entries to settings.json.

AI rationale

Evidence

dist/cli.jsdist/install-claude-hook.jspackage.json

Decision evidence

public snapshot

AI called this Suspicious at 84.0% confidence as Dangerous Capability with medium false-positive risk.

Evidence for warning

The manifest defines only a prepublishOnly lifecycle script; the flagged setup is not wired to npm install.
The CLI exposes Claude hook installation as an explicit command.
That command copies a hook script into the user's Claude hooks directory and updates settings.json.
The installer adds hooks for prompt submission, tool use, and session start.
The activation source contains a comment instructing reviewers not to re-audit it, which is reviewer-directed manipulation.

Evidence against

The hook setup runs through an explicit CLI command rather than an npm install hook.
The installer backs up settings.json and deduplicates existing hook registrations.

Behavioral surface

Source

ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemNetworkShell

Supply chain

HighEntropyStringsPossibleObfuscationUrlStrings

Source & flagged code

7 flagged

dist/adapters.jsView file

Lines 26-66javascript

26}

27/**

28 * Load a single adapter from its module path.

29 * Supports:

30 * - Local files: "./adapters/foo.js" (relative to config file)

31 * - npm packages: "@compr/contextengine-jira"

32 * - Named exports: "my-package#myAdapter"

33 */

34async function loadAdapterModule(modulePath, config) {

35 let moduleSpecifier = modulePath;

36 let exportName = null;

37 // Support "module#export" syntax

38 if (modulePath.includes("#")) {

39 const [mod, exp] = modulePath.split("#", 2);

40 moduleSpecifier = mod;

41 exportName = exp;

42 }

43 // Resolve relative paths from CWD

44 if (moduleSpecifier.startsWith(".") || moduleSpecifier.startsWith("/")) {

45 moduleSpecifier = resolve(process.cwd(), moduleSpecifier);

46 }

47 const mod = await import(moduleSpecifier);

48 // Check for factory function

Medium

Dynamic Require

Package source references dynamic require/import behavior.

dist/adapters.jsView on unpkg · L46

49 if (exportName && typeof mod[exportName] === "function") {

50 return await mod[exportName](config);

51 }

52 if (typeof mod.createAdapter === "function") {

53 return await mod.createAdapter(config);

54 }

55 // Check for default export

56 if (mod.default) {

57 if (typeof mod.default === "function") {

58 return await mod.default(config);

59 }

60 if (typeof mod.default.collect === "function") {

61 return mod.default;

62 }

63 }

64 // Check for named adapter export

65 if (typeof mod.adapter === "object" && typeof mod.adapter.collect === "function") {

66 return mod.adapter;

dist/install-autostart.jsView file

Lines 1-33javascript

1// 🔒 LOCKED [AUTOSTART-INSTALL] — 2026-06-23

2// ⛔ NEVER bootstrap into a `system/` domain (would need root + run as root).

3// Use `gui/$UID` — per-user agent, started at user login, runs as the user.

4// ⛔ NEVER write the plist before checking if a server is already listening on

5// the port. A pre-existing process means we'd race with the launchd-managed

6// one for port 7842.

7// ⛔ NEVER ship a plist that calls `npx -y @latest` — every restart would

8// fetch the registry, eating ~3s and breaking offline. Pin a specific

9// node path + a specific dist path.

dist/install-claude-hook.jsView file

•Manifest-trigger-reachable source links an external AI-agent control path to a behavior-bearing write operation. import { existssync, readfilesync, writefilesync, copyfilesync, chmodsync, mkdirsync } from "fs"; import { join, dirname } from "path"; import { homedir } from "os"; import { fileurltopath } from "url"; import { execsync } from "child_process"; const __dirname_esm = dirname(fileurltopath(import.meta.url)); const claude_dir = jo

High

Trigger Reachable External Ai Agent Control Surface Mutation

Manifest-trigger-reachable source writes behavior-bearing configuration into a user or project AI-agent control surface.

dist/install-claude-hook.jsView on unpkg

defaults/claude-code-hook.shView file

•path = defaults/claude-code-hook.sh kind = build_helper sizeBytes = 3337 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

defaults/claude-code-hook.shView on unpkg

dist/activation.js#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 6

High

Semantic Analysis Limited

A bounded semantic-analysis stage reached its safety limit; remaining detectors completed, but this package requires AI review.

dist/activation.js#virtual:normalized:round1View on unpkg

dist/agents.jsView file

•matchType = previous_version_dangerous_delta matchedPackage = @compr/opscontext-mcp@2.8.2 matchedIdentity = npm:QGNvbXByL29wc2NvbnRleHQtbWNw:2.8.2 similarity = 0.881 summary = stored previous version shares package body but lacks this dangerous source file

High

Previous Version Dangerous Delta

This package version adds a dangerous source file absent from the previous stored version; route for source-aware review.

dist/agents.jsView on unpkg

Source & flagged code

7 flagged · loading source

dist/adapters.jsView file

46} L47: const mod = await import(moduleSpecifier); L48: // Check for factory function

Medium

Dynamic Require

Package source references dynamic require/import behavior.

dist/adapters.jsView on unpkg · L46

dist/install-autostart.jsView file

13Trigger-reachable persistence chain: manifest.bin -> dist/cli.js -> dist/install-autostart.js L13: // forever — defeating the whole point. L14: // FIX: To add platform support beyond macOS, branch on process.platform and L15: // add equivalent systemd / NSSM logic. Keep `gui/$UID` and KeepAlive ... L19: import { homedir, platform } from "os"; L20: import { execSync } from "child_process"; L21: import { createRequire } from "module"; ... L34: const __filename_esm = fileURLToPath(import.meta.url); L35: const __dirname_esm = dirname(__filename_esm); L36: const requireFromHere = createRequire(import.meta.url); ... L101: // was itself run with it (an explicit choice, not a default). L102: export function buildPlist(nodePath, entryPath, nodeBinDir, env = process.env) { L103: // OPSCON

High

Trigger Reachable Persistence

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

dist/install-autostart.jsView on unpkg · L13

Findings

4 High6 Medium5 Low

HighTrigger Reachable External Ai Agent Control Surface Mutationdist/install-claude-hook.js

HighTrigger Reachable Persistencedist/install-autostart.js

HighSemantic Analysis Limiteddist/activation.js#virtual:normalized:round1

HighPrevious Version Dangerous Deltadist/agents.js

MediumDynamic Requiredist/adapters.js

MediumNetwork

MediumEnvironment Vars

MediumInstall Persistencedist/install-autostart.js

MediumShips Build Helperdefaults/claude-code-hook.sh

MediumStructural Risk Force Deep Review

LowNon Install Lifecycle Scripts

LowScripts Present

LowFilesystem

Affected versions and remediation

This report applies to @compr/opscontext-mcp@2.8.3.

See all versions of @compr/opscontext-mcp for other recorded verdicts.

Review the evidence and your use of @compr/opscontext-mcp@2.8.3 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: 2026-10-11 07:04:43.442Z (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