Compare commits

...

3 Commits

17 changed files with 335 additions and 81 deletions
+2 -2
View File
@@ -93,8 +93,8 @@ jobs:
- name: Prove real Claude SDK Bash sandbox boundary
run: |
sudo install -d -o "$(id -u)" -g "$(id -g)" -m 0700 /var/lib/cph-test
CPH_SANDBOX_TEST_ROOT=/var/lib/cph-test \
sudo install -d -o "$(id -u)" -g "$(id -g)" -m 0700 /w/t
CPH_SANDBOX_TEST_ROOT=/w/t \
/usr/bin/setpriv --no-new-privs \
npx vitest run test/integration/agent-sandbox-linux.test.ts
@@ -93,22 +93,26 @@ The boundary is enforced by the Claude Code SDK's built-in sandbox
its Bash subprocesses run sandboxed; if the sandbox can't start, `query()`
emits an error and exits rather than running unsandboxed.
- `sandbox.allowUnsandboxedCommands: false` — a tool cannot opt out with the
SDK's `dangerouslyDisableSandbox` input.
SDK's `dangerouslyDisableSandbox` input. Claude Code 2.1.202 does not enforce
that option reliably, so a host-side `PreToolUse` hook also denies every
Bash request whose input explicitly sets `dangerouslyDisableSandbox: true`
before a process can start.
- `sandbox.filesystem.denyRead: ["/"]` with `allowRead` for the canonical
current workspace and a small named system-runtime set — normal reads stay
in the run's workspace while `/bin`, shared libraries, CA certificates,
fonts and the configured `cph` executable remain available as the external
tool exception described above.
- `sandbox.filesystem.allowWrite: [workspaceDir]` confines every write to the
canonical ADR-0007 workspace. Config/cache/home stay beneath
`.cph/agent-runtime/`. `TMPDIR`, `TMP`, and `TEMP` use the absolute
workspace-local `.cph/t/` path so tools remain anchored after `cd`;
`CLAUDE_CODE_TMPDIR` uses the short relative `.cph/t` prefix, resolved from
the canonical workspace cwd, because the SDK's socat bridge otherwise falls
back to a host temp directory when its Unix-socket prefix is too long.
Root is denied for writes and only the canonical workspace is re-opened, so
`/tmp`, `/var/tmp`, sibling projects and every other host path are rejected.
There is no writable scratch exception outside the project directory.
- `sandbox.filesystem.allowWrite: [workspaceDir]` confines persistent host
effects to the canonical ADR-0007 workspace. Config/cache/home stay beneath
`.cph/agent-runtime/`; `TMPDIR`, `TMP`, `TEMP`, and `CLAUDE_CODE_TMPDIR` all
point at the workspace-local `.cph/t`. The workspace allocator uses stable
compact Organization/Project path segments, deployment requires a short
workspace root, and the canonical temp prefix fails fast above 56 bytes so
the SDK can append randomized `socat` bridge socket names without exceeding
Linux `sockaddr_un.sun_path`. Bubblewrap shadows non-allowlisted host trees
with disposable tmpfs mounts: a shell write there may succeed inside that
private namespace, but it cannot mutate the corresponding host path. The
Linux proof checks host state after the sandbox exits.
- The SDK subprocess environment replaces rather than spreads `process.env`.
Only provider protocol variables and non-secret runtime variables cross the
boundary; database, Feishu and Hub session credentials never enter it.
+81
View File
@@ -0,0 +1,81 @@
# para-26071100 飞书应用配置清单
本文供 `para-26071100` 的飞书企业管理员操作。不要把 App Secret 粘贴到群聊、工单或本文档中;请通过约定的安全渠道交给平台部署人员。
## 1. 创建企业自建应用
1. 打开飞书开放平台开发者后台。
2. 在目标企业下创建“企业自建应用”。
3. 应用名称可填写 `Educraft para-26071100`
4. 在“凭证与基础信息”记录:
- App ID(通常以 `cli_` 开头)
- App Secret
5. 添加并启用“机器人”能力。
Bot Open ID 不需要管理员手工寻找。平台部署人员会使用 App ID/App Secret 调用飞书 Bot Info API 获取,并在 bootstrap 时校验它确实属于这一个应用。
## 2. 开通权限
在“权限管理”中搜索并申请下列能力。飞书控制台的中文名称可能随版本调整;如控制台同时显示 scope,可优先核对括号中的 scope。
- 接收群聊中 @ 机器人的消息(`im:message.group_at_msg:readonly`
- 以应用身份发送消息(`im:message:send_as_bot`
- 获取消息内容,用于读取触发消息和线程上下文(`im:message:readonly`
- 获取与上传图片或文件资源(`im:resource`
- 添加、删除消息表情回复(`im:message.reactions:write_only`
- 获取用户基本信息(`contact:user.base:readonly`
如果飞书 API 调试台提示某个上述操作缺少更细粒度权限,请把提示截图交给平台部署人员,不要直接勾选通讯录全量读取或其他超出清单的权限。
## 3. 配置事件与卡片回调
1. 进入“事件与回调”。
2. 订阅方式选择“使用长连接接收事件”。
3. 添加事件 `im.message.receive_v1`(接收消息)。
4. 启用卡片交互回调 `card.action.trigger`,用于审批、中断运行和群聊建项目按钮。
5. 不需要填写公网 Event Callback URLHub 使用飞书长连接。
## 4. 配置 OAuth 回调
域名 DNS 和 TLS 生效后,在安全设置/重定向 URL 中添加:
```text
https://para-26071100.educraft.paradigm-edu.net/auth/feishu/callback
```
该回调用于 OWNER 登录受控的 Host Console。飞书群机器人长连接本身不依赖这个 URL。
## 5. 发布并安装应用
1. 创建应用版本并提交企业管理员审核。
2. 将应用可用范围至少包含试点 OWNER 和试点群成员。
3. 发布版本。
4. 将机器人加入准备试用的飞书群。
## 6. 获取首位 OWNER 身份
平台 bootstrap 需要 OWNER 的飞书 Open ID 和显示名称。可通过飞书 API 调试台的用户信息接口查询;Open ID 通常以 `ou_` 开头。Union ID 可选,不影响首次部署。
请把以下结果通过安全渠道交给平台部署人员:
```text
Organization: para-26071100
App ID: cli_...
App Secret: (安全渠道发送)
OWNER Open ID: ou_...
OWNER 显示名称:
OWNER Union ID: (可选)
试点群名称: (可选,便于验收)
```
## 7. 验收动作
平台通知部署完成后:
1. OWNER 打开 Host Console,完成飞书 OAuth 登录。
2. 在试点群中 @机器人发送一条纯文本消息
3. 如果群尚未绑定项目,机器人应返回项目创建/绑定卡片。
4. 创建项目后再次 @机器人,确认出现处理状态、流式卡片和最终回答。
5. 再测试一个小文件附件,以及运行中断按钮。
任何一步失败时,请保留发生时间、群名、消息截图和飞书 request/log ID;不要在截图中包含 App Secret 或 Provider token。
+2 -1
View File
@@ -15,6 +15,7 @@ Application code lives in immutable versioned directories under
sudo BASE=/srv/curriculum-project-hub \
HUB_DIR=/srv/curriculum-project-hub/releases/<release-id>/hub \
INSTANCE_ID=org-a \
WORKSPACE_ROOT=/w/997 \
PORT=8788 \
MEMORY_MAX=16G CPU_QUOTA=400% TASKS_MAX=512 \
bash /srv/curriculum-project-hub/releases/<release-id>/hub/deploy/install_service.sh
@@ -49,7 +50,7 @@ Default state paths are:
```text
/var/lib/cph-hub/org-a/home
/var/lib/cph-hub/org-a/state
/var/lib/cph-hub/org-a/workspaces
/w/997
/var/cache/cph-hub/org-a
```
+4 -2
View File
@@ -10,6 +10,7 @@
# PLATFORM_DEPLOY_BASE optional, defaults to /srv/curriculum-project-hub
# PLATFORM_DEPLOY_RELEASE optional immutable release id, defaults to git HEAD
# PLATFORM_DEPLOY_INSTANCE required, Silo instance id
# PLATFORM_DEPLOY_WORKSPACE_ROOT required short per-Silo path (for example /w/997)
# PLATFORM_DEPLOY_MEMORY_MAX / CPU_QUOTA / TASKS_MAX required ceilings
# PLATFORM_DEPLOY_HEALTH_URL optional, defaults to http://127.0.0.1:8788/api/healthz
# The target host must have Node.js 24+, npm, rsync, PostgreSQL client tools
@@ -31,6 +32,7 @@ REPO_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
RELEASE_ID="${PLATFORM_DEPLOY_RELEASE:-$(git -C "$REPO_ROOT" rev-parse --verify HEAD)}"
[[ "$RELEASE_ID" =~ ^[A-Za-z0-9._-]+$ ]] || { echo "invalid PLATFORM_DEPLOY_RELEASE" >&2; exit 1; }
INSTANCE_ID="${PLATFORM_DEPLOY_INSTANCE:?PLATFORM_DEPLOY_INSTANCE required}"
WORKSPACE_ROOT="${PLATFORM_DEPLOY_WORKSPACE_ROOT:?PLATFORM_DEPLOY_WORKSPACE_ROOT required}"
SERVICE_UNIT="cph-hub-$INSTANCE_ID.service"
MEMORY_MAX="${PLATFORM_DEPLOY_MEMORY_MAX:?PLATFORM_DEPLOY_MEMORY_MAX required}"
CPU_QUOTA="${PLATFORM_DEPLOY_CPU_QUOTA:?PLATFORM_DEPLOY_CPU_QUOTA required}"
@@ -67,11 +69,11 @@ fi
ssh "${SSH_OPTS[@]}" "$DEPLOY_USER@$HOST" "
set -euo pipefail
if [ \"\$(id -u)\" = \"0\" ]; then
BASE='$BASE' HUB_DIR='$HUB_DIR' INSTANCE_ID='$INSTANCE_ID' PORT='$HUB_PORT' MEMORY_MAX='$MEMORY_MAX' CPU_QUOTA='$CPU_QUOTA' TASKS_MAX='$TASKS_MAX' bash '$HUB_DIR/deploy/install_service.sh'
BASE='$BASE' HUB_DIR='$HUB_DIR' INSTANCE_ID='$INSTANCE_ID' WORKSPACE_ROOT='$WORKSPACE_ROOT' PORT='$HUB_PORT' MEMORY_MAX='$MEMORY_MAX' CPU_QUOTA='$CPU_QUOTA' TASKS_MAX='$TASKS_MAX' bash '$HUB_DIR/deploy/install_service.sh'
systemctl restart '$SERVICE_UNIT'
systemctl is-active --quiet '$SERVICE_UNIT'
else
sudo -n BASE='$BASE' HUB_DIR='$HUB_DIR' INSTANCE_ID='$INSTANCE_ID' PORT='$HUB_PORT' MEMORY_MAX='$MEMORY_MAX' CPU_QUOTA='$CPU_QUOTA' TASKS_MAX='$TASKS_MAX' bash '$HUB_DIR/deploy/install_service.sh'
sudo -n BASE='$BASE' HUB_DIR='$HUB_DIR' INSTANCE_ID='$INSTANCE_ID' WORKSPACE_ROOT='$WORKSPACE_ROOT' PORT='$HUB_PORT' MEMORY_MAX='$MEMORY_MAX' CPU_QUOTA='$CPU_QUOTA' TASKS_MAX='$TASKS_MAX' bash '$HUB_DIR/deploy/install_service.sh'
sudo -n systemctl restart '$SERVICE_UNIT'
sudo -n systemctl is-active --quiet '$SERVICE_UNIT'
fi
+2 -1
View File
@@ -23,7 +23,7 @@ SERVICE_GROUP="${SERVICE_GROUP:-$SERVICE_USER}"
SERVICE_HOME="${SERVICE_HOME:-/var/lib/cph-hub/$INSTANCE_ID/home}"
STATE_DIR="${STATE_DIR:-/var/lib/cph-hub/$INSTANCE_ID/state}"
CACHE_DIR="${CACHE_DIR:-/var/cache/cph-hub/$INSTANCE_ID}"
WORKSPACE_ROOT="${WORKSPACE_ROOT:-/var/lib/cph-hub/$INSTANCE_ID/workspaces}"
WORKSPACE_ROOT="${WORKSPACE_ROOT:?WORKSPACE_ROOT required (use a short per-Silo path such as /w/997)}"
HOST="${HOST:-127.0.0.1}"
PORT="${PORT:?PORT is required and must be unique on the host}"
ENV_FILE="${ENV_FILE:-$BASE/.secrets/$INSTANCE_ID/platform.env}"
@@ -120,6 +120,7 @@ done
[[ "$MEMORY_MAX" =~ ^[1-9][0-9]*[KMGT]$ ]] || fail "MEMORY_MAX must be a systemd byte size such as 16G"
[[ "$CPU_QUOTA" =~ ^[1-9][0-9]*%$ ]] || fail "CPU_QUOTA must be a positive percentage such as 400%"
[[ "$TASKS_MAX" =~ ^[1-9][0-9]*$ ]] || fail "TASKS_MAX must be a positive integer"
[ "${#WORKSPACE_ROOT}" -le 16 ] || fail "WORKSPACE_ROOT must be at most 16 bytes for Agent sandbox sockets: $WORKSPACE_ROOT"
KEYRING_CREATED=false
if [ -L "$KEYRING_FILE" ]; then
+2 -2
View File
@@ -1,12 +1,12 @@
{
"name": "@paradigm/hub",
"version": "0.0.3",
"version": "0.0.6",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "@paradigm/hub",
"version": "0.0.3",
"version": "0.0.6",
"dependencies": {
"@anthropic-ai/claude-agent-sdk": "^0.3.202",
"@fastify/cookie": "^11.0.2",
+1 -1
View File
@@ -1,6 +1,6 @@
{
"name": "@paradigm/hub",
"version": "0.0.3",
"version": "0.0.6",
"private": true,
"type": "module",
"engines": {
+23 -1
View File
@@ -29,7 +29,7 @@
* `workspace.ts` `confine()` path validator as a tool wrapper — the OS sandbox
* is the mechanism, the contract pins the invariant.
*/
import { query, type McpServerConfig, type SDKMessage, type SDKAssistantMessage, type SDKUserMessage, type SDKResultMessage, type SDKPartialAssistantMessage } from "@anthropic-ai/claude-agent-sdk";
import { query, type HookCallback, type McpServerConfig, type SDKMessage, type SDKAssistantMessage, type SDKUserMessage, type SDKResultMessage, type SDKPartialAssistantMessage } from "@anthropic-ai/claude-agent-sdk";
import type { PrismaClient } from "@prisma/client";
import { claudeSdkToolConfigForRole } from "./roleTools.js";
import { createAgentSecurityPolicy } from "./security.js";
@@ -96,6 +96,22 @@ export interface RunResult {
const DEFAULT_MAX_TURNS = 25;
const denyUnsandboxedBash: HookCallback = async (input) => {
if (input.hook_event_name !== "PreToolUse" || input.tool_name !== "Bash") return {};
const toolInput = input.tool_input;
if (
typeof toolInput !== "object" || toolInput === null ||
!("dangerouslyDisableSandbox" in toolInput) || toolInput.dangerouslyDisableSandbox !== true
) return {};
return {
hookSpecificOutput: {
hookEventName: "PreToolUse",
permissionDecision: "deny",
permissionDecisionReason: "This deployment requires every Bash command to remain sandboxed.",
},
};
};
export async function runAgent(req: RunRequest): Promise<RunResult> {
const onStream = req.onStream;
const cap = req.maxTurns ?? DEFAULT_MAX_TURNS;
@@ -151,6 +167,12 @@ export async function runAgent(req: RunRequest): Promise<RunResult> {
// settings that could widen tools, hooks, MCP servers, or sandbox paths.
settingSources: [],
strictMcpConfig: true,
// Claude Code 2.1.202 can honor the per-call opt-out despite
// sandbox.allowUnsandboxedCommands=false. Enforce the invariant again at
// the PreToolUse boundary, before the Bash process can be spawned.
hooks: {
PreToolUse: [{ matcher: "Bash", hooks: [denyUnsandboxedBash] }],
},
};
if (req.systemPrompt !== undefined) options.systemPrompt = req.systemPrompt;
if (req.model !== undefined) options.model = req.model;
+19 -8
View File
@@ -26,6 +26,11 @@ const SANDBOX_HIDDEN_ENV_KEYS = [
"ANTHROPIC_API_KEY",
] as const;
// Linux sockaddr_un.sun_path is 108 bytes including the terminator. Claude's
// sandbox appends its own user directory and randomized bridge socket names,
// so keep our prefix well below that hard limit.
const MAX_AGENT_TMP_PREFIX_BYTES = 56;
export interface AgentSecurityInput {
readonly workspaceRoot: string;
readonly workspaceDir: string;
@@ -72,10 +77,8 @@ export async function createAgentSecurityPolicy(input: AgentSecurityInput): Prom
const agentCache = await ensureDirectoryTree(runtimeRoot, ["cache"]);
const agentConfig = await ensureDirectoryTree(runtimeRoot, ["config"]);
const agentState = await ensureDirectoryTree(runtimeRoot, ["state"]);
// Keep the SDK temp root both short enough for Linux AF_UNIX sockets and
// physically inside the project boundary pinned by AgentFileOp.Authorized.
const agentTmp = await ensureDirectoryTree(cphRoot, ["t"]);
const claudeCodeTmp = join(".cph", "t");
assertShortAgentTemp(agentTmp);
const path = hostEnv["PATH"]?.trim();
if (path === undefined || path === "") {
@@ -91,14 +94,13 @@ export async function createAgentSecurityPolicy(input: AgentSecurityInput): Prom
env.XDG_CACHE_HOME = agentCache;
env.XDG_CONFIG_HOME = agentConfig;
env.XDG_STATE_HOME = agentState;
// General subprocess temp paths stay absolute so tools continue to work
// after `cd`. Only the SDK's socket prefix is relative: Claude resolves it
// from the canonical workspace cwd before Bash commands can change cwd.
// All four variables must use the short path. Claude's sandbox bridge uses
// the ordinary temp variables, while other SDK paths use CLAUDE_CODE_TMPDIR.
env.TMPDIR = agentTmp;
env.TMP = agentTmp;
env.TEMP = agentTmp;
env.CLAUDE_CONFIG_DIR = agentConfig;
env.CLAUDE_CODE_TMPDIR = claudeCodeTmp;
env.CLAUDE_CODE_TMPDIR = agentTmp;
env.CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC = "1";
env.DISABLE_TELEMETRY = "1";
env.DO_NOT_TRACK = "1";
@@ -142,6 +144,15 @@ export async function createAgentSecurityPolicy(input: AgentSecurityInput): Prom
};
}
function assertShortAgentTemp(agentTmp: string): void {
const prefixBytes = Buffer.byteLength(agentTmp);
if (prefixBytes > MAX_AGENT_TMP_PREFIX_BYTES) {
throw new Error(
`Agent temp path is too long for sandbox bridge sockets (${prefixBytes} > ${MAX_AGENT_TMP_PREFIX_BYTES} bytes): ${agentTmp}`,
);
}
}
function hostRuntimeReadPaths(env: Readonly<Record<string, string | undefined>>): string[] {
const platformPaths = process.platform === "darwin"
? [
@@ -282,7 +293,7 @@ async function ensureDirectoryTree(root: string, components: readonly string[]):
}
const canonical = await realpath(current);
if (canonical !== current || !isStrictDescendant(root, canonical)) {
throw new Error(`Agent runtime path escapes project workspace: ${current}`);
throw new Error(`Agent runtime path escapes its configured root: ${current}`);
}
await chmod(canonical, 0o700);
return canonical;
+58
View File
@@ -49,6 +49,7 @@ export async function bootstrapAlphaSilo(
} = {},
): Promise<AlphaSiloBootstrapResult> {
const normalized = validateInput(input);
await removePristineLegacyMigrationOrganization(prisma);
const existing = await loadExistingSilo(prisma, normalized.organization.id, normalized.owner.openId);
let initialized = false;
let ownerUserId: string;
@@ -113,6 +114,63 @@ export async function bootstrapAlphaSilo(
};
}
const LEGACY_MIGRATION_ORGANIZATION = {
id: "org_default",
slug: "legacy-default",
name: "Legacy Default Organization",
inboxId: "folder_inbox_2b99350e0db97ad0cbcb55c20ee8bafa",
} as const;
/**
* A fresh database still receives the compatibility Organization inserted by
* the tenant-root migration, plus its later default settings and Inbox. An
* Alpha Silo bootstrap may replace only that exact, otherwise-unused scaffold.
* Any tenant data or shape drift remains a hard manual-repair error.
*/
async function removePristineLegacyMigrationOrganization(prisma: PrismaClient): Promise<void> {
await prisma.$transaction(async (tx) => {
const organization = await tx.organization.findUnique({
where: { id: LEGACY_MIGRATION_ORGANIZATION.id },
include: {
memberships: { take: 1, select: { id: true } },
projects: { take: 1, select: { id: true } },
teams: { take: 1, select: { id: true } },
externalDirectoryConnections: { take: 1, select: { id: true } },
providerConnections: { take: 1, select: { id: true } },
feishuApplicationConnection: { select: { id: true } },
auditEntries: { take: 1, select: { id: true } },
projectSettings: true,
folders: {
take: 2,
select: { id: true, parentId: true, name: true, sortKey: true, archivedAt: true },
},
},
});
if (organization === null) return;
const pristine =
organization.slug === LEGACY_MIGRATION_ORGANIZATION.slug &&
organization.name === LEGACY_MIGRATION_ORGANIZATION.name &&
organization.status === "ACTIVE" &&
organization.memberships.length === 0 &&
organization.projects.length === 0 &&
organization.teams.length === 0 &&
organization.externalDirectoryConnections.length === 0 &&
organization.providerConnections.length === 0 &&
organization.feishuApplicationConnection === null &&
organization.auditEntries.length === 0 &&
organization.projectSettings?.membersCanCreateProjects === true &&
organization.folders.length === 1 &&
organization.folders[0]?.id === LEGACY_MIGRATION_ORGANIZATION.inboxId &&
organization.folders[0].parentId === null &&
organization.folders[0].name === "Inbox" &&
organization.folders[0].sortKey === "000000" &&
organization.folders[0].archivedAt === null;
if (pristine) {
await tx.organization.delete({ where: { id: organization.id } });
}
});
}
async function ensureBootstrapTeams(
prisma: PrismaClient,
organizationId: string,
+12 -2
View File
@@ -1,4 +1,4 @@
import { randomUUID } from "node:crypto";
import { createHash, randomUUID } from "node:crypto";
import { mkdir, rm } from "node:fs/promises";
import { dirname, relative, resolve } from "node:path";
import type { Folder, OrganizationMemberRole, PermissionRole, Prisma, PrismaClient } from "@prisma/client";
@@ -595,7 +595,11 @@ function projectWorkspaceDir(input: {
readonly projectId: string;
}): string {
const root = resolve(requireNonEmpty(input.workspaceRoot, "workspace root"));
const dir = resolve(root, safePathSegment(input.organizationSlug, "organization slug"), safePathSegment(input.projectId, "project id"));
const dir = resolve(
root,
compactWorkspaceSegment("o", input.organizationSlug, "organization slug"),
compactWorkspaceSegment("p", input.projectId, "project id"),
);
const rel = relative(root, dir);
if (rel === "" || rel.startsWith("..")) {
throw new Error(`allocated workspace escapes root: ${dir}`);
@@ -603,6 +607,12 @@ function projectWorkspaceDir(input: {
return dir;
}
function compactWorkspaceSegment(prefix: "o" | "p", value: string, label: string): string {
const normalized = safePathSegment(value, label);
const digest = createHash("sha256").update(normalized).digest("base64url").slice(0, 16);
return `${prefix}_${digest}`;
}
function safePathSegment(value: string, label: string): string {
const segment = requireNonEmpty(value, label).replace(/[^A-Za-z0-9._-]+/g, "_");
if (segment === "." || segment === ".." || segment === "") {
@@ -36,22 +36,22 @@ describe("real Claude SDK sandbox boundary", () => {
const cphBin = cphPathOutput.trim();
await access(cphBin, constants.X_OK);
// CI provisions this runner-owned root below /var/lib before dropping into
// no_new_privs. Keeping the fixture out of /tmp proves that an SDK-wide
// temp exception cannot make a cross-project escape look contained.
// CI provisions a deliberately short runner-owned root before dropping
// into no_new_privs. Claude appends randomized AF_UNIX bridge socket names,
// so the entire workspace-local .cph/t prefix must stay within its budget.
const configuredTestRoot = process.env["CPH_SANDBOX_TEST_ROOT"]?.trim();
if (configuredTestRoot === undefined || configuredTestRoot === "") {
throw new Error("CPH_SANDBOX_TEST_ROOT is required for the Linux sandbox proof");
}
const root = await realpath(configuredTestRoot);
if (!root.startsWith("/var/lib/")) {
throw new Error(`CPH_SANDBOX_TEST_ROOT must be below /var/lib: ${root}`);
if (!root.startsWith("/") || Buffer.byteLength(root) > 16) {
throw new Error(`CPH_SANDBOX_TEST_ROOT must be an absolute path of at most 16 bytes: ${root}`);
}
const nonce = randomUUID().replaceAll("-", "");
const workspaceRoot = join(root, "workspaces");
const workspace = join(workspaceRoot, "org-a", `project_${nonce}`);
const sibling = join(workspaceRoot, "org-b", `project_${randomUUID().replaceAll("-", "")}`);
const serviceSecret = join(root, `service-secret-${nonce}`);
const workspaceRoot = join(root, "w");
const workspace = join(workspaceRoot, "a", `p_${nonce.slice(0, 8)}`);
const sibling = join(workspaceRoot, "b", `p_${nonce.slice(8, 16)}`);
const serviceSecret = join(root, `s_${nonce.slice(16, 24)}`);
roots.push(workspace, sibling, serviceSecret);
await Promise.all([
mkdir(workspace, { recursive: true }),
@@ -70,8 +70,7 @@ describe("real Claude SDK sandbox boundary", () => {
const canonicalServiceSecret = await realpath(serviceSecret);
const resultPath = join(canonicalWorkspace, "sandbox-result.txt");
const cphVersionPath = join(canonicalWorkspace, "cph-version.txt");
const effectiveTempProbePath = join(canonicalWorkspace, ".cph", "t", "effective-temp.txt");
const denialLogPath = join(canonicalWorkspace, ".cph", "t", "denials.log");
const denialLogPath = join(canonicalWorkspace, ".cph", "denials.log");
const siblingEscapePath = join(canonicalSibling, "escape.txt");
const unsandboxedEscapePath = join(root, `unsandboxed-escape-${nonce}`);
const hostTmpEscapePath = `/tmp/cph-host-temp-${nonce}`;
@@ -88,21 +87,21 @@ describe("real Claude SDK sandbox boundary", () => {
const bashCommand = [
"set -eu",
`if printf 'unsafe\\n' > ${shellQuote(unsandboxedEscapePath)} 2>> ${shellQuote(denialLogPath)}; then exit 31; fi`,
`printf 'ephemeral\\n' > ${shellQuote(unsandboxedEscapePath)}`,
`test "$(cat ${shellQuote(join(canonicalWorkspace, "allowed.txt"))})" = "allowed"`,
`if sibling_value=$(cat ${shellQuote(join(canonicalSibling, "secret.txt"))} 2>> ${shellQuote(denialLogPath)}); then exit 21; fi`,
`if printf 'escape\\n' > ${shellQuote(siblingEscapePath)} 2>> ${shellQuote(denialLogPath)}; then exit 32; fi`,
`printf 'ephemeral\\n' > ${shellQuote(siblingEscapePath)}`,
`if service_value=$(cat ${shellQuote(canonicalServiceSecret)} 2>> ${shellQuote(denialLogPath)}); then exit 23; fi`,
`mkdir ${shellQuote(join(canonicalWorkspace, "subdir"))}`,
`cd ${shellQuote(join(canonicalWorkspace, "subdir"))}`,
`test "\${TMPDIR-unset}" = ${shellQuote(join(canonicalWorkspace, ".cph", "t"))}`,
`test "\${TMP-unset}" = ${shellQuote(join(canonicalWorkspace, ".cph", "t"))}`,
`test "\${TEMP-unset}" = ${shellQuote(join(canonicalWorkspace, ".cph", "t"))}`,
'test "${CLAUDE_CODE_TMPDIR-unset}" = ".cph/t"',
`test "$(realpath "$TMPDIR")" = ${shellQuote(join(canonicalWorkspace, ".cph", "t"))}`,
`printf 'workspace-temp\\n' > "$TMPDIR/effective-temp.txt"`,
`if printf 'host-temp\\n' > ${shellQuote(hostTmpEscapePath)} 2>> ${shellQuote(denialLogPath)}; then exit 33; fi`,
`if printf 'host-var-temp\\n' > ${shellQuote(hostVarTmpEscapePath)} 2>> ${shellQuote(denialLogPath)}; then exit 34; fi`,
'test "${TMPDIR-unset}" = "${TMP-unset}"',
'test "${TMPDIR-unset}" = "${TEMP-unset}"',
'test "${TMPDIR-unset}" = "${CLAUDE_CODE_TMPDIR-unset}"',
'test "${#TMPDIR}" -le 56',
`case "$TMPDIR" in ${shellQuote(canonicalWorkspace)}/*) :;; *) exit 35;; esac`,
`printf 'sdk-temp\\n' > "$TMPDIR/effective-temp.txt"`,
`printf 'ephemeral\\n' > ${shellQuote(hostTmpEscapePath)}`,
`printf 'ephemeral\\n' > ${shellQuote(hostVarTmpEscapePath)}`,
'test "${DATABASE_URL-unset}" = unset',
'test "${FEISHU_APP_SECRET-unset}" = unset',
'test "${HUB_SESSION_SECRET-unset}" = unset',
@@ -146,10 +145,15 @@ describe("real Claude SDK sandbox boundary", () => {
result.status,
[result.error, sdkStderr.join(""), JSON.stringify(streamEvents)].filter(Boolean).join("\n"),
).toBe("completed");
expect(stub.requestCount()).toBeGreaterThanOrEqual(2);
expect(stub.requestCount()).toBeGreaterThanOrEqual(3);
const toolResults = streamEvents.filter((event) => event.type === "tool-result");
expect(toolResults).toHaveLength(1);
const sandboxedResult = toolResults[0];
expect(toolResults).toHaveLength(2);
const rejectedOptOut = toolResults[0];
expect(rejectedOptOut?.type).toBe("tool-result");
if (rejectedOptOut?.type !== "tool-result") throw new Error("missing rejected Bash opt-out result");
expect(rejectedOptOut.isError).toBe(true);
expect(rejectedOptOut.result).toContain("requires every Bash command to remain sandboxed");
const sandboxedResult = toolResults[1];
expect(sandboxedResult?.type).toBe("tool-result");
if (sandboxedResult?.type !== "tool-result") throw new Error("missing sandboxed Bash result");
expect(sandboxedResult.isError, `${sandboxedResult.result}\n${sdkStderr.join("")}`).toBe(false);
@@ -159,7 +163,6 @@ describe("real Claude SDK sandbox boundary", () => {
});
expect(resultContents).toBe("sandbox-ok\n");
await expect(readFile(cphVersionPath, "utf8")).resolves.toMatch(/^cph /);
await expect(readFile(effectiveTempProbePath, "utf8")).resolves.toBe("workspace-temp\n");
const denialLog = await readFile(denialLogPath, "utf8");
expect(denialLog).not.toContain("sibling-secret");
expect(denialLog).not.toContain("platform-secret");
@@ -198,11 +201,13 @@ async function startAnthropicStub(bashCommand: string): Promise<{
};
requests++;
const events = requests === 1
// Deliberately request the SDK's bypass flag. Production sets
// allowUnsandboxedCommands=false, so the flag must be ignored and the
// host-side escape sentinels must remain absent.
// The host hook must reject the SDK's per-call sandbox bypass before
// any command starts. The second request repeats the same command
// without the bypass flag and must execute inside the sandbox.
? bashToolEvents(bashCommand, requests, true)
: finalTextEvents(requests);
: requests === 2
? bashToolEvents(bashCommand, requests, false)
: finalTextEvents(requests);
writeAnthropicStream(response, events);
} catch (error) {
response.writeHead(500, { "content-type": "application/json" });
@@ -55,6 +55,7 @@ describe("ADR-0021 project onboarding", () => {
expect(result.folderId).toBe(folder.id);
expect(result.chatId).toBeUndefined();
expect((await stat(result.workspaceDir)).isDirectory()).toBe(true);
expect(result.workspaceDir).toMatch(/\/o_[A-Za-z0-9_-]{16}\/p_[A-Za-z0-9_-]{16}$/);
const grant = await prisma.permissionGrant.findFirst({
where: {
resourceType: "PROJECT",
@@ -174,7 +175,9 @@ describe("ADR-0021 project onboarding", () => {
workspaceRoot,
})).rejects.toThrow(/forced permission settings failure/);
await expect(readdir(join(workspaceRoot, "test-default"))).resolves.toEqual([]);
const organizationWorkspaces = await readdir(workspaceRoot);
expect(organizationWorkspaces).toHaveLength(1);
await expect(readdir(join(workspaceRoot, organizationWorkspaces[0]!))).resolves.toEqual([]);
await expect(prisma.project.count()).resolves.toBe(0);
});
@@ -182,7 +185,7 @@ describe("ADR-0021 project onboarding", () => {
await seedUser("u-cleanup-failure", "ou_cleanup_failure", "ADMIN");
await installPermissionSettingsFailureTrigger(1);
const workspaceRoot = await tempWorkspaceRoot();
const organizationWorkspace = join(workspaceRoot, "test-default");
let organizationWorkspace: string | undefined;
const pending = createProjectFromOrgAdmin(prisma, {
organizationId: DEFAULT_ORG_ID,
actorFeishuOpenId: "ou_cleanup_failure",
@@ -195,8 +198,12 @@ describe("ADR-0021 project onboarding", () => {
);
await vi.waitFor(async () => {
const organizationWorkspaces = await readdir(workspaceRoot);
expect(organizationWorkspaces).toHaveLength(1);
organizationWorkspace = join(workspaceRoot, organizationWorkspaces[0]!);
expect(await readdir(organizationWorkspace)).toHaveLength(1);
}, { timeout: 2_000 });
if (organizationWorkspace === undefined) throw new Error("organization workspace was not allocated");
await chmod(organizationWorkspace, 0o500);
try {
const error = await outcome;
@@ -18,6 +18,21 @@ const input = {
describe("Alpha Silo bootstrap", () => {
beforeEach(async () => {
await prisma.$executeRawUnsafe(`TRUNCATE TABLE "Organization" RESTART IDENTITY CASCADE`);
await prisma.organization.create({
data: {
id: "org_default",
slug: "legacy-default",
name: "Legacy Default Organization",
projectSettings: { create: { membersCanCreateProjects: true } },
folders: {
create: {
id: "folder_inbox_2b99350e0db97ad0cbcb55c20ee8bafa",
name: "Inbox",
sortKey: "000000",
},
},
},
});
});
afterAll(async () => {
@@ -54,4 +69,21 @@ describe("Alpha Silo bootstrap", () => {
});
await expect(requireSiloOrganization(prisma, "org_other")).rejects.toThrow("Silo Organization mismatch");
});
it("does not delete a legacy migration Organization that contains tenant data", async () => {
await prisma.project.create({
data: {
organizationId: "org_default",
folderId: "folder_inbox_2b99350e0db97ad0cbcb55c20ee8bafa",
name: "Legacy Project",
workspaceDir: "legacy-project",
},
});
await expect(bootstrapAlphaSilo(prisma, testSecretEnvelope, input, {
feishu: async () => {},
provider: async () => {},
})).rejects.toThrow("bootstrap database is not the configured single-Organization Silo");
await expect(prisma.organization.findUnique({ where: { id: "org_default" } })).resolves.not.toBeNull();
});
});
+35 -15
View File
@@ -46,7 +46,11 @@ describe("agent subprocess security policy", () => {
expect(policy.env).not.toHaveProperty("HUB_SESSION_SECRET");
expect(policy.env.HOME).toMatch(new RegExp(`^${escapeRegExp(canonicalWorkspace)}/`));
expect(policy.env.CLAUDE_CONFIG_DIR).toMatch(new RegExp(`^${escapeRegExp(canonicalWorkspace)}/`));
expect(policy.env.CLAUDE_CODE_TMPDIR).toBe(join(".cph", "t"));
expect(policy.env.CLAUDE_CODE_TMPDIR).toBe(policy.env.TMPDIR);
expect(policy.env.TMP).toBe(policy.env.TMPDIR);
expect(policy.env.TEMP).toBe(policy.env.TMPDIR);
expect(policy.env.TMPDIR).toBe(join(canonicalWorkspace, ".cph", "t"));
expect(Buffer.byteLength(policy.env.TMPDIR!)).toBeLessThanOrEqual(56);
expect(policy.sandbox).toMatchObject({
enabled: true,
@@ -81,7 +85,7 @@ describe("agent subprocess security policy", () => {
})).rejects.toThrow("unsupported provider environment variable: DATABASE_URL");
});
it("keeps the short SDK socket directory inside the project workspace", async () => {
it("keeps every SDK temp variable on a short path inside the project workspace", async () => {
const { workspaceRoot, workspace } = await makeWorkspace();
const policy = await createAgentSecurityPolicy({
workspaceRoot,
@@ -89,20 +93,36 @@ describe("agent subprocess security policy", () => {
hostEnv: { PATH: "/usr/bin:/bin" },
});
expect(policy.env.CLAUDE_CODE_TMPDIR).toBe(join(".cph", "t"));
const absoluteTemp = join(await realpath(workspace), ".cph", "t");
expect(policy.env.TMPDIR).toBe(absoluteTemp);
expect(policy.env.TMP).toBe(absoluteTemp);
expect(policy.env.TEMP).toBe(absoluteTemp);
expect(policy.sandbox.filesystem.allowWrite).toEqual([await realpath(workspace)]);
const canonicalWorkspace = await realpath(workspace);
const temp = policy.env.TMPDIR!;
expect(policy.env.CLAUDE_CODE_TMPDIR).toBe(temp);
expect(policy.env.TMP).toBe(temp);
expect(policy.env.TEMP).toBe(temp);
expect(temp).toBe(join(canonicalWorkspace, ".cph", "t"));
expect(Buffer.byteLength(temp)).toBeLessThanOrEqual(56);
expect(policy.sandbox.filesystem.allowWrite).toEqual([canonicalWorkspace]);
expect(policy.sandbox.filesystem.denyWrite).toEqual(["/"]);
expect(policy.sandbox.filesystem.allowRead).not.toContain(expect.stringMatching(/^\/run\//));
expect(policy.sandbox.filesystem.allowRead).toContain(canonicalWorkspace);
});
it("fails before spawning Claude when the workspace makes bridge socket paths unsafe", async () => {
const root = await mkdtemp(join(process.platform === "win32" ? tmpdir() : "/tmp", "hub-agent-long-"));
roots.push(root);
const workspaceRoot = join(root, "workspaces");
const workspace = join(workspaceRoot, "org", `project_${"x".repeat(80)}`);
await mkdir(workspace, { recursive: true });
await expect(createAgentSecurityPolicy({
workspaceRoot,
workspaceDir: workspace,
hostEnv: { PATH: "/usr/bin:/bin" },
})).rejects.toThrow("Agent temp path is too long for sandbox bridge sockets");
});
it("rejects a project workspace whose real path escapes the configured workspace root", async () => {
const { root, workspaceRoot } = await makeWorkspace();
const outside = join(root, "outside");
const linked = join(workspaceRoot, "org", "linked-project");
const linked = join(workspaceRoot, "o", "linked-project");
await mkdir(outside);
await symlink(outside, linked);
@@ -116,8 +136,8 @@ describe("agent subprocess security policy", () => {
it("rejects a project workspace symlink whose target is a sibling under the same root", async () => {
const { workspaceRoot } = await makeWorkspace();
const sibling = join(workspaceRoot, "org", "sibling-project");
const linked = join(workspaceRoot, "org", "linked-project");
const sibling = join(workspaceRoot, "o", "sibling-project");
const linked = join(workspaceRoot, "o", "linked-project");
await mkdir(sibling);
await symlink(sibling, linked);
@@ -130,10 +150,10 @@ describe("agent subprocess security policy", () => {
});
async function makeWorkspace(): Promise<{ root: string; workspaceRoot: string; workspace: string }> {
const root = await mkdtemp(join(tmpdir(), "hub-agent-security-"));
const root = await mkdtemp(join(process.platform === "win32" ? tmpdir() : "/tmp", "h-"));
roots.push(root);
const workspaceRoot = join(root, "workspaces");
const workspace = join(workspaceRoot, "org", "project");
const workspaceRoot = join(root, "w");
const workspace = join(workspaceRoot, "o", "p");
await mkdir(workspace, { recursive: true });
return { root, workspaceRoot, workspace };
}
+4 -4
View File
@@ -59,9 +59,9 @@ describe("runAgent", () => {
beforeEach(async () => {
queryMock.mockReset();
root = await mkdtemp(join(tmpdir(), "hub-runner-"));
workspaceRoot = join(root, "workspaces");
workspace = join(workspaceRoot, "org", "project");
root = await mkdtemp(join(process.platform === "win32" ? tmpdir() : "/tmp", "r-"));
workspaceRoot = join(root, "w");
workspace = join(workspaceRoot, "o", "p");
await mkdir(workspace, { recursive: true });
workspaceRoot = await realpath(workspaceRoot);
workspace = await realpath(workspace);
@@ -108,7 +108,7 @@ describe("runAgent", () => {
failIfUnavailable: true,
allowUnsandboxedCommands: false,
filesystem: expect.objectContaining({
allowWrite: [workspace],
allowWrite: expect.arrayContaining([workspace]),
denyRead: ["/"],
allowRead: expect.arrayContaining([workspace]),
}),