Staff MCP Server

Tools that read sensitive data ([read_file, list_dir]) and tools that can send data out ([manage_mcp_session, execute_command]) are exposed together. An agent can move private data to the sink.

Evidence: sources [read_file, list_dir] → sinks [manage_mcp_session, execute_command]

Fix: Keep secret-reading and egress capabilities on separate, separately-approved servers.

Location: flow read_file → manage_mcp_session

Tool "manage_mcp_session" appears to run shell commands or evaluate code (parameter "command"). Arbitrary execution driven by model input is one of the most dangerous MCP capabilities; combined with any untrusted input it becomes RCE.

Fix: Sandbox execution, allowlist commands/arguments, and never pass model output to a shell unescaped.

Location: tool manage_mcp_session

Tool "execute_command" appears to run shell commands or evaluate code (keyword "execute command" in tool name, parameter "command"). Arbitrary execution driven by model input is one of the most dangerous MCP capabilities; combined with any untrusted input it becomes RCE.

Fix: Sandbox execution, allowlist commands/arguments, and never pass model output to a shell unescaped.

Location: tool execute_command

In the server's implementation (`dist/src/docker-proxy.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: rt { spawn } from "node:child_process"; import { randomBytes } from "node:crypto"; import * as fs from "node: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/src/docker-proxy.js

In the server's implementation (`dist/src/lsp/manager.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: spawn, execSync } from "child_process"; import * as path from "path"; import * as fs from "fs"; import { pathToFileURL,

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/src/lsp/manager.js

In the server's implementation (`dist/src/tools/file-tools.js:3`): 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 { exec } from "child_process"; import { promisify } from "util"; import * as readline from "readline"; 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/src/tools/file-tools.js

In the server's implementation (`dist/src/tools/mcp-client-tools.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: import { spawn } from "child_process"; import { Client } from "@modelcontextprotocol/sdk/client/index.js"; import { Std

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/src/tools/mcp-client-tools.js

In the server's implementation (`dist/src/tools/shell-tools.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: spawn, execSync } from "child_process"; import * as os from "os"; import * as fs from "fs"; import { z } from "zod"; imp

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/src/tools/shell-tools.js

In the server's implementation (`dist/src/utils/tool-utils.js:3`): 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: t { exec, spawn } from "child_process"; import { promisify } from "util"; import { createRequire } from "node:module"; i

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/src/utils/tool-utils.js

In the server's implementation (`dist/src/lsp/manager.js:250`): A shell/process command assembled from concatenated or interpolated values is command injection when any part is attacker-influenced — the OWASP canonical RCE flow. Verify what reaches the interpolated value. 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(`${platformNpm} ${args.join(" ")}`, { cwd: STAFF_TOOLS_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/src/lsp/manager.js

In the server's implementation (`dist/src/tools/shell-tools.js:173`): A shell/process command assembled from concatenated or interpolated values is command injection when any part is attacker-influenced — the OWASP canonical RCE flow. Verify what reaches the interpolated value. 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 { execSync(`taskkill /pid ${child.pid} /T /F`, { stdio: "ignore" }); } catch { /*

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/src/tools/shell-tools.js

Tool "edit_file_by_replace" can write, overwrite or delete files (keyword "edit_file" in tool name). Verify it is scoped to a safe directory.

Fix: Constrain file operations to an explicit, non-sensitive root; reject path traversal.

Location: tool edit_file_by_replace

Tool "write_file" can write, overwrite or delete files (keyword "write_file" in tool name). Verify it is scoped to a safe directory.

Fix: Constrain file operations to an explicit, non-sensitive root; reject path traversal.

Location: tool write_file

Tool "delete_file" can write, overwrite or delete files (keyword "delete_file" in tool name). Verify it is scoped to a safe directory.

Fix: Constrain file operations to an explicit, non-sensitive root; reject path traversal.

Location: tool delete_file

Tool "manage_mcp_session" takes a command-shaped parameter "command" with no enum/pattern constraint. Free-form, model- or attacker-controlled arguments reaching a shell is the command-injection precondition.

Fix: Constrain the parameter (enum/pattern), or build the command from a fixed template with escaped args.

Location: tool manage_mcp_session · inputSchema.properties.command

Tool "execute_command" takes a command-shaped parameter "command" with no enum/pattern constraint. Free-form, model- or attacker-controlled arguments reaching a shell is the command-injection precondition.

Fix: Constrain the parameter (enum/pattern), or build the command from a fixed template with escaped args.

Location: tool execute_command · inputSchema.properties.command

Tool "edit_file_by_replace" can mutate/egress but declares no destructiveHint. Clients that don't default to spec-safe behavior may not prompt before running it.

Fix: Declare accurate annotations, and gate destructive tools on user confirmation regardless.

Location: tool edit_file_by_replace

Tool "write_file" can mutate/egress but declares no destructiveHint. Clients that don't default to spec-safe behavior may not prompt before running it.

Fix: Declare accurate annotations, and gate destructive tools on user confirmation regardless.

Location: tool write_file

Tool "delete_file" can mutate/egress but declares no destructiveHint. Clients that don't default to spec-safe behavior may not prompt before running it.

Fix: Declare accurate annotations, and gate destructive tools on user confirmation regardless.

Location: tool delete_file

Tool "manage_mcp_session" can mutate/egress but declares no destructiveHint. Clients that don't default to spec-safe behavior may not prompt before running it.

Fix: Declare accurate annotations, and gate destructive tools on user confirmation regardless.

Location: tool manage_mcp_session

Tool "execute_command" can mutate/egress but declares no destructiveHint. Clients that don't default to spec-safe behavior may not prompt before running it.

Fix: Declare accurate annotations, and gate destructive tools on user confirmation regardless.

Location: tool execute_command

Tool "read_file" takes a path parameter "path" with no constraint. Without a canonicalize-and-contain check (not visible statically), this permits ../ traversal outside the intended root.

Fix: Resolve and verify the path stays within an allowed root; reject traversal sequences.

Location: tool read_file · inputSchema.properties.path

Tool "edit_file_by_replace" takes a path parameter "path" with no constraint. Without a canonicalize-and-contain check (not visible statically), this permits ../ traversal outside the intended root.

Fix: Resolve and verify the path stays within an allowed root; reject traversal sequences.

Location: tool edit_file_by_replace · inputSchema.properties.path

Tool "write_file" takes a path parameter "path" with no constraint. Without a canonicalize-and-contain check (not visible statically), this permits ../ traversal outside the intended root.

Fix: Resolve and verify the path stays within an allowed root; reject traversal sequences.

Location: tool write_file · inputSchema.properties.path

Tool "delete_file" takes a path parameter "path" with no constraint. Without a canonicalize-and-contain check (not visible statically), this permits ../ traversal outside the intended root.

Fix: Resolve and verify the path stays within an allowed root; reject traversal sequences.

Location: tool delete_file · inputSchema.properties.path

Tool "list_dir" takes a path parameter "path" with no constraint. Without a canonicalize-and-contain check (not visible statically), this permits ../ traversal outside the intended root.

Fix: Resolve and verify the path stays within an allowed root; reject traversal sequences.

Location: tool list_dir · inputSchema.properties.path

In a packaging/dev/install script (shipped, but not the server runtime) (`dist/tests/package-info.test.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: { execFile } from "node:child_process"; import * as fs from "node:fs"; import * as path from "node:path"; import { pathT

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/tests/package-info.test.js

In a packaging/dev/install script (shipped, but not the server runtime) (`dist/tests/shell-tools.test.js:70`): 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: st { spawn } = require('child_process'); const g = spawn(process.execPath, ['-e', 'setInterval(() => {}, 1000)']); conso

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/tests/shell-tools.test.js