Kastell MCP Server

In the server's implementation (`dist/core/audit/checks/accounts.js:83`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: NETRC, name: "No .netrc Files", severity: "warning", check: (output) => { const hasN

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/core/audit/checks/accounts.js

In the server's implementation (`dist/core/audit/checks/secrets.js:41`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: tch(/^(\d{3,4})\s+(.*(?:id_rsa|id_ed25519|id_ecdsa|id_dsa|.*\.pem).*)/); if (match) {

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/core/audit/checks/secrets.js

In the server's implementation (`dist/core/audit/commands.js:235`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: -la /root/.rhosts /root/.netrc /root/.forward /etc/hosts.equiv 2>/dev/null || echo 'NONE'`, `awk -F: '($3 < 100

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/core/audit/commands.js

In the server's implementation (`dist/core/audit/compliance/categories/index.js:286`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: ", "Ensure no users have .netrc files", "full")], [CHECK_IDS.ACCOUNTS.ACCT_NO_FORWARD]: [cis("6.2.10", "Ensure no us

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/core/audit/compliance/categories/index.js

In the server's implementation (`dist/core/deploy.js:26`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: H key generated (~/.ssh/id_ed25519)"); } else { logger.warning("Could not generate SSH key —

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/core/deploy.js

In the server's implementation (`dist/core/provision.js:204`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: H key generated (~/.ssh/id_ed25519)\n"); } try { const keyId = await provider.uploadSshKey(getSshKeyName

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/core/provision.js

In the server's implementation (`dist/utils/sshKey.js:7`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: const SSH_KEY_FILES = ["id_ed25519.pub", "id_rsa.pub", "id_ecdsa.pub"]; export function findLocalSshKey() { const ss

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/utils/sshKey.js

In the server's implementation (`dist/core/doctor-local.js:1`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: ort { spawnSync } from "child_process"; import { existsSync, accessSync, constants } from "fs"; import axios from "axios

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/core/doctor-local.js

In the server's implementation (`dist/core/plugin.js:1`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: import { spawn } from "child_process"; import { PLUGIN_STATUS_LOADED } from "../plugin/registry.js"; import { existsSyn

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/core/plugin.js

In the server's implementation (`dist/core/scheduleManager.js:1`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: ort { spawnSync } from "child_process"; import { appendFileSync, mkdirSync, readdirSync, statSync } from "fs"; import {

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/core/scheduleManager.js

In the server's implementation (`dist/utils/encryption.js:4`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: ort { spawnSync } from "child_process"; import { platform } from "os"; import { join } from "path"; import { KASTELL_DIR

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/utils/encryption.js

In the server's implementation (`dist/utils/openBrowser.js:1`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: port { execFile } from "child_process"; import { isWindows } from "./platform.js"; import { sanitizedEnv } from "./ssh.j

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/utils/openBrowser.js

In the server's implementation (`dist/utils/scp.js:1`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: import { spawn } from "child_process"; import { assertValidIp, sanitizedEnv, resolveScpPath } from "./ssh.js"; import {

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/utils/scp.js

In the server's implementation (`dist/utils/secureWrite.js:2`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: ort { spawnSync } from "child_process"; import { resolve, sep } from "path"; import { tmpdir } from "os"; import { isWin

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/utils/secureWrite.js

In the server's implementation (`dist/utils/ssh.js:1`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: pawn, spawnSync } from "child_process"; import { existsSync } from "fs"; import { isWindows } from "./platform.js"; impo

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/utils/ssh.js

In the server's implementation (`dist/utils/sshKey.js:2`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: ort { spawnSync } from "child_process"; import { homedir } from "os"; import { join } from "path"; import { sanitizedEnv

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/utils/sshKey.js

In the server's implementation (`kastell-plugin/hooks/stop-quality-check.cjs:6`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: { execSync } = require('child_process'); // MANDATORY stdin guard — exit silently if stdin unavailable (e.g. after /cle

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server kastell-plugin/hooks/stop-quality-check.cjs

In the server's implementation (`dist/adapters/shared.js:20`): A hardcoded outbound call to a fixed external host inside server code is a classic exfiltration/telemetry channel — especially paired with reads of local data. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: try { await axios.get(`https://${domain}`, { timeout: HTTP_TIMEOUT_MS, valid

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/adapters/shared.js

In the server's implementation (`dist/utils/healthCheck.js:13`): A hardcoded outbound call to a fixed external host inside server code is a classic exfiltration/telemetry channel — especially paired with reads of local data. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: try { await axios.get(`http://${ip}:${port}`, { timeout: HTTP_TIMEOUT_MS, va

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/utils/healthCheck.js

In the server's implementation (`dist/utils/updateCheck.js:59`): A hardcoded outbound call to a fixed external host inside server code is a classic exfiltration/telemetry channel — especially paired with reads of local data. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: const response = await axios.get("https://registry.npmjs.org/kastell/latest", { timeout: REQUEST_TIMEOUT_MS

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/utils/updateCheck.js

In the server's implementation (`dist/plugin/activeProbeLoader.js:15`): Loading a module chosen at runtime (from a variable) can pull in and run attacker-influenced code paths. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: th, importer = (url) => import(url)) { const root = realpathSync(pluginDir); const candidate = resolve(root, han

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/plugin/activeProbeLoader.js

In the server's implementation (`dist/plugin/handlerResolver.js:6`): Loading a module chosen at runtime (from a variable) can pull in and run attacker-influenced code paths. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: const mod = (await import(url)); const handler = (typeof mod.default === "function" ? mod.default : mod.default

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/plugin/handlerResolver.js

In the server's implementation (`dist/plugin/loader.js:13`): Loading a module chosen at runtime (from a variable) can pull in and run attacker-influenced code paths. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: ns?.importer ?? ((p) => import(p)); const loaded = []; const errors = []; clearPluginRegistry(); if (!ex

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/plugin/loader.js