Files
PaperClipAI/server/src/secrets/local-encrypted-provider.ts
T
DottaandPaperclip 198218bf55 fix(config): require shared storage for managed cloud (#18350)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work.
> - Managed Cloud instances must retain shared objects and secrets
across server replacement.
> - Implicit storage defaults can select local disk or an unintended
bucket.
> - Cloud runtime settings now come from a canonical database snapshot.
> - This pull request validates bootstrap prerequisites before database
setup and effective S3 settings after snapshot resolution.
> - The benefit is an actionable refusal before local signing state or
HTTP setup, while valid canonical Cloud settings work.

## Linked Issues or Issue Description

**What happened?**

A managed Cloud instance could start with local disk storage, an
implicit bucket, or a local key file. After canonical Cloud
configuration landed, checking only environment or file values would
also reject a valid database bucket. Shared schema defaults could make
an omitted raw bucket appear explicit.

**Expected behavior**

Managed Cloud requires external PostgreSQL, a stable environment master
key, and effective S3 settings with a nonblank bucket. The canonical
snapshot remains the configuration authority. Invalid configuration
refuses startup before signing or HTTP effects. Self-hosted defaults
remain compatible.

**Steps to reproduce**

1. Select managed Cloud with its tenant token or managed configuration
document.
2. Leave storage as local disk, or omit the S3 bucket from a raw
canonical snapshot.
3. Start the server. Startup must refuse. An explicit valid bucket in
the canonical snapshot must work without a file or environment bucket.

**Paperclip version or commit**

Current base: `4edd10804e2b42941aa2038e90c362c13a0b98dc`.

**Deployment mode**

Managed Cloud. Ordinary self-hosted defaults remain unchanged.

Related PRs: #18335 owns canonical Cloud runtime configuration. #18327
owns S3 run logs. #18352 owns credential persistence. #18347 owns
cutover checks. This PR adds configuration prerequisites to those
existing boundaries.

## What Changed

- Validate external PostgreSQL and a stable environment master key at
the Cloud bootstrap boundary, before database setup.
- Validate effective S3 settings after canonical snapshot publication
and before Cloud signing, runtime consumers, or HTTP setup.
- Require a nonblank bucket in the Cloud snapshot schema. Reject omitted
or blank buckets in every legacy Cloud file before one-time import
normalization, even when an environment bucket is supplied. Keep shared
self-hosted schema defaults.
- Preserve typed bucket provenance at the existing file-read boundary.
Accept explicit canonical bucket names, including `paperclip`, and
environment settings when the snapshot is null.
- Validate effective legacy storage with the existing resolver before
seeding. A rejected candidate leaves no canonical rows, so a corrected
source file can be imported on retry. Environment overlays are checked
but never persisted.
- Add real PostgreSQL snapshot and migration-retry tests, startup
refusal tests, and environment/file compatibility controls. Document
that Cloud always requires its stable environment key, even with an
external default secrets provider.

## Verification

- Latest focused acceptance passed 199/199 in six files (54.29 seconds),
using the command below plus
`server/src/__tests__/cloud-configuration-migration-postgres.test.ts`.
Real PostgreSQL proves rejected storage leaves all three migration
authority rows absent, then a corrected legacy file imports on retry. An
external-default-secrets control proves the deliberate unconditional
Cloud master-key contract.
- Before the peer correction, the frozen product diff passed 178 tests
in five files: managed-cloud-prerequisites, cloud-runtime-config,
server-startup-feedback-export, config-file, and
cloud-configuration-readers. Command: `pnpm exec vitest run
server/src/__tests__/{managed-cloud-prerequisites,cloud-runtime-config,server-startup-feedback-export,config-file,cloud-configuration-readers}.test.ts
--maxWorkers=1 --no-file-parallelism`. Real PostgreSQL used a qualified
POSIX test-only PG18.6 override because the host cannot supply the stock
runtime's SysV semaphore allocation.
- Coverage includes raw database and legacy omitted/blank bucket
rejection, valid database and environment buckets, null snapshots,
explicit default-named buckets, self-hosted compatibility, and refusal
before local signing mutation.
- The real AWS SDK configured-provider control uses a simulated loopback
S3 service. It writes, reads from a fresh provider, and deletes an
object. It also reads an encrypted credential with the stable
environment key. This does not qualify production cloud permissions.
- `git diff --check` passed. Final merges add unrelated master changes
from #18241, #18363, and #18370; the focused storage product diff is
unchanged.
- Peer review reproduced a fresh legacy local-disk import that gained
the default bucket before an S3 environment overlay. The importer now
requires a raw explicit bucket for every legacy Cloud file. Six controls
cover omission/blank values with and without an environment bucket; the
full focused rerun passed 184/184 across all five files in 194.89
seconds.
- Initial revised startup testing passed 167/168: one fixture omitted
the database URL and failed before its intended key assertion. The
corrected final run passed 178/178. An initial direct server typecheck
failed because Runner declarations were not built; this is a setup
failure, not a compile pass. Integration verification rebuilds
dependencies before full checks.
- Prior head `fae1e4aa` passed complete CI with 53 successes, two
intended skips, and Greptile 5/5.
`de71345f379cecb71dbfb219f35e14f942c10a3f` then passed complete CI (55
successes, three workflow skips, and a neutral external review), with
Greptile 5/5 and zero unresolved threads. The external review identified
the preseed rejection problem; its correction is committed at
`75e04d0d77e19dcef0a8cb070b6b518f0f173b14`. The final merge head is
`19b809ea2823a6a4e5ceb48ef2c2bac4721733d6`; [Fresh complete
CI](https://github.com/paperclipai/paperclip/actions/runs/41127309762)
passed on that head: 55 successful checks, three workflow skips, and one
neutral external review. Greptile reviewed this exact head at 5/5 with
no actionable issues. Both earlier external review threads are resolved
with the tested fixes. Earlier infrastructure retries remain documented
in [the first
diagnosis](https://github.com/paperclipai/paperclip/pull/18350#issuecomment-6529964690)
and [the runner shutdown
evidence](https://github.com/paperclipai/paperclip/pull/18350#issuecomment-6530364769).
Earlier missing SDK build and fixture metadata failures were repaired
before those passes.
- Current frozen-head CI first attempt failed one unchanged Antigravity
authentication success case in Runner Vitest 1/2 (`outcome=failure`,
1994 tests passed). The same file passed 11/11 locally in 2.33 seconds,
and identical Runner source passed the prior complete CI run. Attempt 2
retried only failed jobs (the Runner shard and dependent aggregate).
Both passed without a source change. No second retry was used. The exact
timing cause is not established. [Failed
job](https://github.com/paperclipai/paperclip/actions/runs/41127309762/job/124309181778).
- No production data, buckets, or live credentials were used. Complete
local repository suites are not duplicated in this worktree;
current-head required CI supplies the full suite gate.

## Risks

- Managed deployments with local disk, omitted buckets, file-only keys,
or missing external database URLs now refuse startup. Operators must
supply valid prerequisites before rollout.
- Required deployment environment overlays must remain stable across
replacements. For example, a saved local-disk provider with an explicit
bucket can use an environment S3 provider; removing that required
overlay correctly refuses startup until the environment or canonical
configuration is corrected. Environment settings are never silently
copied into the canonical row.
- Effective legacy storage must also pass before first seeding. A bad
candidate does not create authority rows. Existing saved rows remain
authoritative and can require environment overrides or a canonical
configuration change if they are invalid.
- A present legacy Cloud file needs an explicit raw bucket even if its
saved provider is local disk or its environment supplies a bucket.
Correct the file before one-time import, or use the existing
absent-file/null-snapshot environment-only setup. Environment overlays
are not persisted as operator choices.
- A previously stored canonical nonblank `paperclip` bucket is
authoritative. Prior normalization erased whether that value was
originally explicit. This change cannot recover that provenance and does
not reject existing nonblank canonical values.
- Validation proves configuration shape, not cloud permissions, database
reachability, or key compatibility with existing encrypted data. Keep
the master key stable across replacements.
- Cloud snapshot JSON parsing is stricter; omitted or blank raw S3
bucket values need correction. No database migration or provider
operation changes are added. These prerequisites do not prove that every
server persistence path is disposable.

> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.

## Model Used

OpenAI Codex, exact model ID `gpt-6.1-sol`. Context window: unknown. The
model used reasoning, repository tools, code execution, and test
execution.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

---------

Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-10-30 16:21:46 -05:00

336 lines
11 KiB
TypeScript

import { createCipheriv, createDecipheriv, createHash, hkdfSync, randomBytes } from "node:crypto";
import { chmodSync, existsSync, linkSync, mkdirSync, readFileSync, statSync, unlinkSync, writeFileSync } from "node:fs";
import path from "node:path";
import { resolveDefaultSecretsKeyFilePath } from "../home-paths.js";
import type {
PreparedSecretVersion,
SecretProviderHealthCheck,
SecretProviderModule,
SecretProviderValidationResult,
StoredSecretVersionMaterial,
} from "./types.js";
import { badRequest } from "../errors.js";
import { SecretProviderClientError } from "./types.js";
interface LocalEncryptedMaterial extends StoredSecretVersionMaterial {
scheme: "local_encrypted_v1";
iv: string;
tag: string;
ciphertext: string;
}
export type LocalEncryptedKeyEnv = Partial<Record<
"PAPERCLIP_SECRETS_MASTER_KEY" | "PAPERCLIP_SECRETS_MASTER_KEY_FILE", string | undefined
>>;
function resolveMasterKeyFilePath(env: LocalEncryptedKeyEnv = process.env) {
const fromEnv = env.PAPERCLIP_SECRETS_MASTER_KEY_FILE;
if (fromEnv && fromEnv.trim().length > 0) return path.resolve(fromEnv.trim());
return resolveDefaultSecretsKeyFilePath();
}
function decodeMasterKey(raw: string): Buffer | null {
const trimmed = raw.trim();
if (!trimmed) return null;
if (/^[A-Fa-f0-9]{64}$/.test(trimmed)) {
return Buffer.from(trimmed, "hex");
}
try {
const decoded = Buffer.from(trimmed, "base64");
if (decoded.length === 32) return decoded;
} catch {
// ignored
}
if (Buffer.byteLength(trimmed, "utf8") === 32) {
return Buffer.from(trimmed, "utf8");
}
return null;
}
/** Check the configured key without reading, creating, or changing key files. */
export function isValidLocalEncryptedMasterKey(raw: string): boolean {
return decodeMasterKey(raw) !== null;
}
function loadOrCreateMasterKey(allowCreate = true, env: LocalEncryptedKeyEnv = process.env): Buffer {
const envKeyRaw = env.PAPERCLIP_SECRETS_MASTER_KEY;
if (envKeyRaw && envKeyRaw.trim().length > 0) {
const fromEnv = decodeMasterKey(envKeyRaw);
if (!fromEnv) {
throw badRequest(
"Invalid PAPERCLIP_SECRETS_MASTER_KEY (expected 32-byte base64, 64-char hex, or raw 32-char string)",
);
}
return fromEnv;
}
const keyPath = resolveMasterKeyFilePath(env);
if (existsSync(keyPath)) {
enforceKeyFilePermissionsBestEffort(keyPath);
const raw = readFileSync(keyPath, "utf8");
const decoded = decodeMasterKey(raw);
if (!decoded) {
throw badRequest(`Invalid secrets master key at ${keyPath}`);
}
return decoded;
}
if (!allowCreate) throw new Error(`Secrets master key file is missing: ${keyPath}`);
const dir = path.dirname(keyPath);
mkdirSync(dir, { recursive: true });
const generated = randomBytes(32);
const temporaryPath = `${keyPath}.${process.pid}.${randomBytes(12).toString("hex")}.tmp`;
writeFileSync(temporaryPath, generated.toString("base64"), { encoding: "utf8", mode: 0o600, flag: "wx" });
try {
// A hard link publishes a complete file without overwriting a competitor's
// key. Opening the destination with wx would expose an empty/partial file.
try {
linkSync(temporaryPath, keyPath);
} catch (err) {
if ((err as NodeJS.ErrnoException).code !== "EEXIST") throw err;
}
} finally {
unlinkSync(temporaryPath);
}
return loadOrCreateMasterKey(false, env);
}
/** Derive a purpose-specific instance secret without exposing or reusing the vault key. */
export function deriveLocalEncryptedInstanceSecret(purpose: string, env: LocalEncryptedKeyEnv = process.env, options: { allowCreate?: boolean } = {}): string {
return Buffer.from(hkdfSync("sha256", loadOrCreateMasterKey(options.allowCreate ?? true, env), Buffer.alloc(0),
`paperclip:instance-secret:v1\0${purpose}`, 32)).toString("base64url");
}
function enforceKeyFilePermissionsBestEffort(keyPath: string) {
try {
const mode = statSync(keyPath).mode & 0o777;
if ((mode & 0o077) !== 0) {
chmodSync(keyPath, 0o600);
}
} catch {
// best effort only; health checks surface persistent permission problems.
}
}
function sha256Hex(value: string): string {
return createHash("sha256").update(value).digest("hex");
}
function prepareManagedVersion(value: string): PreparedSecretVersion {
const masterKey = loadOrCreateMasterKey();
const valueSha256 = sha256Hex(value);
return {
material: encryptValue(masterKey, value),
valueSha256,
fingerprintSha256: valueSha256,
externalRef: null,
};
}
async function inspectLocalEncryptedHealth(): Promise<SecretProviderHealthCheck> {
const envKeyRaw = process.env.PAPERCLIP_SECRETS_MASTER_KEY;
if (envKeyRaw && envKeyRaw.trim().length > 0) {
if (!decodeMasterKey(envKeyRaw)) {
return {
provider: "local_encrypted",
status: "error",
message:
"PAPERCLIP_SECRETS_MASTER_KEY is invalid; expected 32-byte base64, 64-char hex, or raw 32-char string",
};
}
return {
provider: "local_encrypted",
status: "ok",
message: "Local encrypted provider is using PAPERCLIP_SECRETS_MASTER_KEY",
backupGuidance: [
"Back up the configured master key separately from the database.",
"A restore needs both the database metadata and the same master key.",
],
details: { keySource: "env" },
};
}
const keyPath = resolveMasterKeyFilePath();
if (!existsSync(keyPath)) {
return {
provider: "local_encrypted",
status: "warn",
message: `Secrets key file does not exist yet: ${keyPath}`,
warnings: ["The first managed secret write will create this key file with 0600 permissions."],
backupGuidance: [
"Back up the key file together with database backups.",
"The database alone cannot restore local encrypted secret values.",
],
details: { keySource: "file", keyFilePath: keyPath },
};
}
let mode: number | null = null;
try {
mode = statSync(keyPath).mode & 0o777;
} catch (err) {
return {
provider: "local_encrypted",
status: "error",
message: `Could not stat secrets key file: ${err instanceof Error ? err.message : String(err)}`,
details: { keySource: "file", keyFilePath: keyPath },
};
}
try {
const raw = readFileSync(keyPath, "utf8");
if (!decodeMasterKey(raw)) {
return {
provider: "local_encrypted",
status: "error",
message: `Invalid key material in ${keyPath}`,
details: { keySource: "file", keyFilePath: keyPath },
};
}
} catch (err) {
return {
provider: "local_encrypted",
status: "error",
message: `Could not read secrets key file: ${err instanceof Error ? err.message : String(err)}`,
details: { keySource: "file", keyFilePath: keyPath },
};
}
const warnings =
mode !== null && (mode & 0o077) !== 0
? [`Secrets key file permissions are ${mode.toString(8)}; run chmod 600 ${keyPath}`]
: [];
return {
provider: "local_encrypted",
status: warnings.length > 0 ? "warn" : "ok",
message: `Local encrypted provider configured with key file ${keyPath}`,
warnings,
backupGuidance: [
"Back up the key file together with database backups.",
"The database alone cannot restore local encrypted secret values.",
],
details: { keySource: "file", keyFilePath: keyPath },
};
}
function encryptValue(masterKey: Buffer, value: string): LocalEncryptedMaterial {
const iv = randomBytes(12);
const cipher = createCipheriv("aes-256-gcm", masterKey, iv);
const ciphertext = Buffer.concat([cipher.update(value, "utf8"), cipher.final()]);
const tag = cipher.getAuthTag();
return {
scheme: "local_encrypted_v1",
iv: iv.toString("base64"),
tag: tag.toString("base64"),
ciphertext: ciphertext.toString("base64"),
};
}
function decryptValue(masterKey: Buffer, material: LocalEncryptedMaterial): string {
const iv = Buffer.from(material.iv, "base64");
const tag = Buffer.from(material.tag, "base64");
const ciphertext = Buffer.from(material.ciphertext, "base64");
const decipher = createDecipheriv("aes-256-gcm", masterKey, iv);
decipher.setAuthTag(tag);
try {
const plain = Buffer.concat([decipher.update(ciphertext), decipher.final()]);
return plain.toString("utf8");
} catch (err) {
const keyFingerprint = createHash("sha256").update(masterKey).digest("hex").slice(0, 12);
throw new Error(
`Secret decryption failed (master key fingerprint: ${keyFingerprint}). ` +
`This usually means the master key does not match the key used to encrypt the secret. ` +
`Original error: ${err instanceof Error ? err.message : String(err)}`,
);
}
}
function asLocalEncryptedMaterial(value: StoredSecretVersionMaterial): LocalEncryptedMaterial {
if (
value &&
typeof value === "object" &&
value.scheme === "local_encrypted_v1" &&
typeof value.iv === "string" &&
typeof value.tag === "string" &&
typeof value.ciphertext === "string"
) {
return value as LocalEncryptedMaterial;
}
throw badRequest("Invalid local_encrypted secret material");
}
function resolveLocalEncryptedValue<T>(operation: () => T): T {
try {
return operation();
} catch (cause) {
// This boundary contains only local key I/O and decryption, never database work.
throw new SecretProviderClientError({
code: "provider_unavailable", provider: "local_encrypted", operation: "resolveVersion",
message: cause instanceof Error ? cause.message : "Local encrypted secret could not be resolved",
cause,
});
}
}
/** Reuse one resolved key within a single batch; reload it for the next operation. */
export function createLocalEncryptedCipher(mode: "read" | "write") {
const masterKey = mode === "write"
? loadOrCreateMasterKey()
: resolveLocalEncryptedValue(() => loadOrCreateMasterKey(false));
return {
encrypt(value: string): StoredSecretVersionMaterial {
return encryptValue(masterKey, value);
},
decrypt(material: StoredSecretVersionMaterial): string {
return resolveLocalEncryptedValue(() => decryptValue(masterKey, asLocalEncryptedMaterial(material)));
},
};
}
export const localEncryptedProvider: SecretProviderModule = {
id: "local_encrypted",
descriptor() {
return {
id: "local_encrypted",
label: "Local encrypted (default)",
requiresExternalRef: false,
supportsManagedValues: true,
supportsExternalReferences: false,
configured: true,
};
},
async validateConfig(input): Promise<SecretProviderValidationResult> {
const warnings: string[] = [];
if (input?.deploymentMode === "authenticated" && input.strictMode !== true) {
warnings.push("Strict secret mode should be enabled for authenticated deployments");
}
const health = await inspectLocalEncryptedHealth();
if (health.status === "error") {
throw badRequest(health.message);
}
warnings.push(...(health.warnings ?? []));
return { ok: true, warnings };
},
async createSecret(input) {
return prepareManagedVersion(input.value);
},
async createVersion(input) {
return prepareManagedVersion(input.value);
},
async linkExternalSecret() {
throw badRequest("local_encrypted does not support external reference secrets");
},
async resolveVersion(input) {
return createLocalEncryptedCipher("read").decrypt(input.material);
},
async deleteOrArchive() {
// Secret metadata deletion is handled in Paperclip DB; the local key is shared and must remain.
},
async healthCheck() {
return inspectLocalEncryptedHealth();
},
};