Safe Bifrost MCP Server

This server (without client built-ins) exposes a complete data-exfiltration chain: get_task_progress → list_workspace → create_task. Untrusted input is ingested, private data is read, and it can be sent to an external sink via the agent composing the tools (→). Static analysis proves the primitive exists, not that a specific run will occur.

Fix: Remove one leg of the trifecta: isolate untrusted-input tools from secret-reading tools and from egress tools, or require human approval between them.

Location: flow get_task_progress → list_workspace → create_task

In the server's implementation (`dist/doctor.js:159`): 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: ePaths = [".env", ".ssh/id_rsa", "token.json", "credentials"]; for (const sp of sensitivePaths) { const bloc

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/doctor.js

In the server's implementation (`dist/security/planGuard.js:6`): 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: |token\b|credentials?\b|id_rsa\b|id_ed25519\b|ssh.{0,12}private.{0,8}key|private.{0,8}key|密钥|令牌|凭据|私钥)/is, sugge

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/security/planGuard.js

In the server's implementation (`dist/security/sensitiveGuard.js:10`): 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: ..+$/i, /(?:^|[\\/])id_rsa$/i, /(?:^|[\\/])id_ed25519$/i, /(?:^|[\\/])id_ecdsa$/i, /(?:^|[\\/])\.ssh[\\/

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/security/sensitiveGuard.js

In the server's implementation (`dist/smoke-test.js:217`): 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: [ ".env", ".ssh/id_rsa", "secrets/token.json", "keys/private.key", "cookies.sqlite", ".git-crede

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/smoke-test.js

In the server's implementation (`src/doctor.ts:173`): 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: ePaths = [".env", ".ssh/id_rsa", "token.json", "credentials"]; for (const sp of sensitivePaths) { const blocked = is

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 src/doctor.ts

In the server's implementation (`src/security/planGuard.ts:14`): 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: |token\b|credentials?\b|id_rsa\b|id_ed25519\b|ssh.{0,12}private.{0,8}key|private.{0,8}key|密钥|令牌|凭据|私钥)/is, suggesti

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 src/security/planGuard.ts

In the server's implementation (`src/security/sensitiveGuard.ts:11`): 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: \..+$/i, /(?:^|[\\/])id_rsa$/i, /(?:^|[\\/])id_ed25519$/i, /(?:^|[\\/])id_ecdsa$/i, /(?:^|[\\/])\.ssh[\\/]/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 src/security/sensitiveGuard.ts

In the server's implementation (`src/smoke-test.ts:243`): 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: = [ ".env", ".ssh/id_rsa", "secrets/token.json", "keys/private.key", "cookies.sqlite", ".git-credentia

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 src/smoke-test.ts

Tool "create_task" 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 create_task

In the server's implementation (`dist/doctor.js:12`): 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 } from "node:child_process"; import { createServer } from "node:net"; import { getConfig } from "./config.js"

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/doctor.js

In the server's implementation (`dist/runner/changeCapture.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: spawnSync } from "node:child_process"; import { isSensitivePath } from "../security/sensitiveGuard.js"; const MAX_HASH_

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/runner/changeCapture.js

In the server's implementation (`dist/runner/runTask.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: spawnSync } from "node:child_process"; import { createWriteStream, existsSync, readFileSync, writeFileSync, } from "nod

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/runner/runTask.js

In the server's implementation (`dist/smoke-test.js:19`): 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: spawnSync } from "node:child_process"; import { loadConfig, getConfig } from "./config.js"; import { savePlan } from ".

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/smoke-test.js

In the server's implementation (`src/doctor.ts:13`): 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 } from "node:child_process"; import { createServer } from "node:net"; import { loadConfig, getConfig } from

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 src/doctor.ts

Tool "create_task" takes a command-shaped parameter "test_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 create_task · inputSchema.properties.test_command

Tool "create_task" 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 create_task

Tool "list_workspace" 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_workspace · inputSchema.properties.path

In a packaging/dev/install script (shipped, but not the server runtime) (`scripts/doctor-smoke.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: spawnSync } from "node:child_process"; import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs"; impor

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 scripts/doctor-smoke.js

In a packaging/dev/install script (shipped, but not the server runtime) (`scripts/http-mcp-smoke.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: rt { spawn } from "node:child_process"; import { existsSync, mkdirSync, mkdtempSync, rmSync, writeFileSync, }

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 scripts/http-mcp-smoke.js

In a packaging/dev/install script (shipped, but not the server runtime) (`scripts/lifecycle-smoke.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: spawnSync } from "node:child_process"; import { existsSync, mkdirSync, mkdtempSync, readFileSync, rmSyn

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 scripts/lifecycle-smoke.js

In a packaging/dev/install script (shipped, but not the server runtime) (`scripts/mcp-smoke.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: spawnSync } from "node:child_process"; import { existsSync, mkdirSync, mkdtempSync, readFileSync, rmSyn

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 scripts/mcp-smoke.js

In a packaging/dev/install script (shipped, but not the server runtime) (`scripts/pack-clean.js:10`): 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: ecFileSync } from "node:child_process"; import { join, relative, resolve } from "node:path"; import { fileURLToPath }

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 scripts/pack-clean.js