Files
PaperClipAI/server/src/services/skill-package-store.ts
T
DottaandPaperclip 0e15c92073 feat(skills): make managed content independent of web-server disk (#18334)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work.
> - Agents use installed skills and exact version pins during execution.
> - Managed skills still depended on mutable directories on the web
server.
> - A replacement server could lose acknowledged edits or remove an
unassigned skill whose directory was absent.
> - This pull request makes immutable versions the authority for managed
content and installed repository reads.
> - The server can rebuild delivery files from durable content after
replacement.
> - This establishes skills readiness for ephemeral web servers. General
server draining and workspace recovery remain separate work.

## Linked Issues or Issue Description

**Subsystem affected**

Skills services, shared contracts, database receipts, Skill Studio, and
runtime delivery.

**Problem or motivation**

A fresh web server must discover, read, edit, and deliver protected
installed skills using the same database, object storage, secrets, and
supported application artifacts. It must not need its predecessor's
files or process memory.

**Proposed solution**

Use `managed` ownership with a null source locator. Read installed
content by immutable version. Prepare packages before a short
publication transaction. Persist execution selections before
materialization. Recover exact pinned runtime assets and reject deletion
while active or recovering pins exist.

**Alternatives considered**

Reconstructing a mutable installed directory preserves the old
source-of-truth ambiguity. Permanent dual writes create two competing
authorities. This change uses one managed model after a bounded legacy
conversion.

**Roadmap alignment**

This improves the existing Skills Manager, Studio, version history, and
Cloud execution capabilities. The maintainer supplied the implementation
contract. This does not add a separate skills product or deploy fleet
changes.

Related work: #18082 supplies immutable package storage. #17273
addresses edits outside the API for linked local skills; this PR keeps
that source category separate. #2648 is a closed earlier
ephemeral-filesystem proposal.

## What Changed

- Add managed ownership and focused content publication and execution
materialization modules. Keep the archive format and package storage
tables.
- Publish managed create, save, delete, fork, rename, restore, and
catalog content with version checks. Add durable create receipts.
Preserve file-save receipt responses.
- Return file version identities. Keep Studio drafts after a conflict
and provide an explicit reload action. A delayed reload cannot overwrite
newer typing or navigation.
- Persist exact runtime selections, including empty selections. Rebuild
pinned assets and verify their admitted digests on a replacement server.
- Preserve source access on forks and disconnected copies. Keep the
existing workspace policy category for managed skills.
- Add bounded conversion, eligibility, and independent verification
phases. Load persisted Cloud configuration in the standalone CLI without
seeding or changing it. Reject partial current versions against
installed inventories; recover only complete saved content or the exact
source. Document storage migration blockers and the compatible rollback
floor.
- Add separate-process PostgreSQL/S3 qualification and update domain,
retention, policy, and UI regressions.
- Pass the plugin worker manager into Studio execution, admit restricted
`skill_test` mode through native Runner contracts, and preserve admitted
scope across native and REST tools. Native tests cannot create ordinary
tasks, browse company data, or use external tools.
- Select the exact durable Studio revision even when unassigned or Beta
skills is off. Persist Studio output and audit atomically with native
task completion, preserving the result on replay.

## Verification

- Final revision `13e8af480b88ba6fea5737bac4bd81104d604792` passed full
CI: all 49 workflow jobs completed successfully ([run
41127396098](https://github.com/paperclipai/paperclip/actions/runs/41127396098)).
This includes typecheck, test suites, Runner checks, browser checks and
build. Greptile review 15 is 5/5 on that same revision, with zero
unresolved threads. The PR is ready for review.
- Local workspace typecheck and build passed after integrating master.
Focused checks passed: 33 execution/completion cases, 52 Runner HTTP
cases, 16 managed-content cases, 17 Studio cases, the new real-process
CLI regression, map checks and UI token gates. The CLI regression proves
inspection does not seed missing Cloud configuration. Server typecheck
passed after that fix.
- The independent-process PostgreSQL/S3 matrix passed through normal
HTTP skill routes. It covers A → B → C replacement, lost commit
responses, upload crashes, concurrent saves, binary contents and
executable delivery. Metadata listing and warm selected delivery retain
zero object downloads. Local timing measurements are recorded; they do
not establish hosted latency.
- Hosted A → B → C passed on the owned disposable test tenant with
external Postgres, S3 and a real remote Paperclip Computer. Predecessor
temporary homes were absent. C read and executed B's acknowledged
managed save. New runs used repository v1 while v2 was installed. Hashes
matched an independently checked exact upstream commit.
- Compatible image rollback to `6518b323` passed. Studio retained the
unassigned fork's C save. A fresh remote run executed the newer managed
content and historical repository pin. The actual shell event confirmed
expected hashes, stdout, exit 0 and mode 0555.
- Runtime candidate `9200e3fcb4` deployed successfully ([Cloud run
41124147876](https://github.com/paperclipai/paperclip-cloud/actions/runs/41124147876)).
The final Studio test delivered only the unassigned fork's exact v3,
executed its script and persisted a successful output snapshot.
Independent command-event inspection confirmed all three hashes, stdout,
exit 0 and mode 0555. All 14 installations were version-backed.
Independent verification checked all 25 versions: 23 remote objects, two
bundled database snapshots, zero blockers.
- The deployed image is `9200e3fcb4`. Later master integration, the
client-only delayed-reload draft guard and the CLI bootstrap have
separate automated proof. The CLI bootstrap was also exercised against
the tenant's stored configuration and storage through the same read-only
services. This PR does not claim that the deployed image is the final PR
revision.
- Live Studio checks exposed worker wiring, admission, access scope,
exact selection and completion-publication gaps. They are fixed. Failed
attempts remain in the evidence. The full local suite exceeded 80
minutes under host contention and was stopped. Later combined/isolated
local attempts also timed out; they are not counted as passes. Full CI
covers the integrated suites. Earlier CI found an obsolete restart-scope
fixture, which was corrected and passed all 52 HTTP cases.
- Reproduction and migration:
[`doc/skill-ephemeral-web-servers.md`](doc/skill-ephemeral-web-servers.md).
Source revisions, deployments, versions, run IDs, hashes and failure
history:
[`doc/plans/2026-10-30-ephemeral-web-skills.md`](doc/plans/2026-10-30-ephemeral-web-skills.md).

## Risks

- Legacy conversion requires stopped admission and retired
filesystem-authoritative writers. Pre-conversion binaries are not
supported rollback targets afterward.
- Cloud readiness requires remote object storage. Local provider
receipts and provider mismatches remain blockers until objects are
copied and verified.
- Linked local sources retain filesystem semantics and are outside the
cloud guarantee.
- Cold reads still fetch the existing archive format. Warm runtime
preparation retains the verified-cache path.
- Bundled reconciliation preserves an installed head. Existing release
snapshots remain selectable. An older server must not replace a newer
head with its shipped files.
- No production volumes, fleet defaults, or general deployment
coordination change in this PR.

## Model Used

OpenAI Codex, GPT-6 family, with reasoning, repository tools, code
execution, and browser use. The exact backend model ID and
context-window size are not exposed in this session. No subagents were
used.

## 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 16:18:27 -05:00

688 lines
44 KiB
TypeScript

import { DELIVERY_QUEUES, notifyDeliveryWork } from "./delivery-work-notifications.js";
import { recoverSkillPackage, skillPackageRecoveryReader } from "./skill-package-recovery.js";
import { touchSkillPackage, trimSkillPackageCache } from "./skill-package-cache.js";
import { AsyncLocalStorage } from "node:async_hooks";
import { randomUUID } from "node:crypto";
import { constants, createReadStream, createWriteStream, promises as fs } from "node:fs";
import path from "node:path";
import { pipeline } from "node:stream/promises";
import { and, eq, getTableColumns, inArray, sql } from "drizzle-orm";
import { companySkillRepositorySnapshots as snapshots, companySkillVersions, companySkillSources, withDedicatedDbConnection, type Db, type SkillPackageFileMapping, type SkillPackageStorageReceipt, type SkillPackageRestoreRecipe } from "@paperclipai/db";
import type { CompanySkillVersionFileInventoryEntry } from "@paperclipai/shared";
import { getStorageService, type StorageService } from "../storage/index.js";
import { resolvePaperclipInstanceRoot } from "../home-paths.js";
import { conflict, notFound } from "../errors.js";
import { assertSkillSnapshotPath, skillFileBytes } from "./skill-snapshot.js";
import { ensureSkillCacheDirectory, SkillCacheBusy, withSkillVersionCacheLock } from "./skill-version-cache.js";
import { boundedHashStream, extractSkillPackageArchive, MAX_SKILL_MANIFEST_BYTES, packageSha256,
skillPackageManifest, skillPackageManifestBytes, validateSkillPackageManifest, verifySkillPackageDirectory, relocateSkillPackageVerification, writeSkillPackageArchive,
type SkillPackageManifest } from "./skill-package-files.js";
/** External storage is the default; false is an explicit rolling-upgrade escape hatch. */
export function skillPackageStorageEnabled() {
return process.env.PAPERCLIP_SKILL_PACKAGE_STORAGE_ENABLED !== "false";
}
const descriptorColumns = { id: snapshots.id, companyId: snapshots.companyId, contentHash: snapshots.contentHash,
storage: snapshots.storage, upload: snapshots.upload, createdAt: snapshots.createdAt,
restorable: sql<boolean>`${snapshots.restore} is not null`, restoreRetryAt: snapshots.restoreRetryAt };
type Descriptor = Pick<typeof snapshots.$inferSelect, "id" | "companyId" | "contentHash" | "storage" | "upload" | "createdAt" | "restoreRetryAt"> & { restorable: boolean };
export const skillVersionMetadataColumns = { ...getTableColumns(companySkillVersions),
fileInventory: sql<CompanySkillVersionFileInventoryEntry[]>`'[]'::jsonb` };
export type StoredSkillVersion = typeof companySkillVersions.$inferSelect;
export function skillVersionFileMap(files: Array<Pick<CompanySkillVersionFileInventoryEntry, "path" | "kind" | "encoding">>): SkillPackageFileMapping[] {
return files.map(file => ({ path: file.path, sourcePath: file.path, kind: file.kind, encoding: file.encoding ?? "utf8" as const }));
}
export function skillVersionUsesWholePackage(version: Pick<StoredSkillVersion, "packageFileMap">, manifest: SkillPackageManifest) {
if (!version.packageFileMap) return true;
const paths = new Set(version.packageFileMap.map(file => file.path));
return paths.size === manifest.files.length && version.packageFileMap.length === manifest.files.length
&& version.packageFileMap.every(file => file.path === file.sourcePath)
&& manifest.files.every(file => paths.has(file.path));
}
type ReadScope = { db: Db; descriptors: Map<string, Promise<Descriptor>>; durable: Map<string, Promise<boolean>>; versions: Map<string, StoredSkillVersion> };
const readScope = new AsyncLocalStorage<ReadScope>();
export function withSkillPackageReadScope<T>(db: Db, action: () => Promise<T>): Promise<T> {
return readScope.getStore()?.db === db ? action() : readScope.run({ db, descriptors: new Map(), durable: new Map(), versions: new Map() }, action);
}
const flights = new WeakMap<Db, Map<string, Promise<unknown>>>();
function flight<T>(db: Db, key: string, action: () => Promise<T>): Promise<T> {
let pending = flights.get(db);
if (!pending) { pending = new Map(); flights.set(db, pending); }
const existing = pending.get(key);
if (existing) return existing as Promise<T>;
const work = action().finally(() => pending!.delete(key));
pending.set(key, work);
return work;
}
type SharedFlight = { work: Promise<unknown>; controller: AbortController; waiters: number };
const sharedFlights = new WeakMap<Db, Map<string, SharedFlight>>();
function sharedFlight<T>(db: Db, key: string, cancellation: AbortSignal | undefined, action: (signal: AbortSignal) => Promise<T>): Promise<T> {
cancellation?.throwIfAborted();
let pending = sharedFlights.get(db);
if (!pending) { pending = new Map(); sharedFlights.set(db, pending); }
let entry = pending.get(key);
if (!entry) {
const controller = new AbortController();
entry = { controller, waiters: 0, work: Promise.resolve() };
const created = entry;
entry.work = action(controller.signal).finally(() => {
if (pending!.get(key) === created) pending!.delete(key);
});
pending.set(key, entry);
}
const current = entry;
current.waiters++;
return new Promise<T>((resolve, reject) => {
let settled = false;
const release = () => {
if (settled) return false;
settled = true;
cancellation?.removeEventListener("abort", aborted);
current.waiters--;
return true;
};
const aborted = () => {
if (!release()) return;
if (current.waiters) { reject(cancellation!.reason); return; }
if (pending!.get(key) === current) pending!.delete(key);
current.controller.abort(cancellation!.reason);
// The last caller owns cancellation and waits for storage/queue cleanup.
void current.work.then(() => reject(cancellation!.reason), () => reject(cancellation!.reason));
};
cancellation?.addEventListener("abort", aborted, { once: true });
current.work.then(value => { if (release()) resolve(value as T); }, error => { if (release()) reject(error); });
});
}
type PackageIoQueue = { active: number; waiting: Array<() => void> };
const storageIo: PackageIoQueue = { active: 0, waiting: [] };
// A decoded cold response remains owned until its post-fill cleanup finishes.
const coldResponses: PackageIoQueue = { active: 0, waiting: [] };
// Bound local response buffers independently so stalled storage does not prevent
// verified cached reads, including reads inside an existing DB transaction.
const inventoryIo: PackageIoQueue = { active: 0, waiting: [] };
async function withPackageIo<T>(action: () => Promise<T>, signal: AbortSignal, queue = storageIo): Promise<T> {
signal.throwIfAborted();
if (queue.active >= 2) {
if (queue.waiting.length >= 32) throw conflict("Skill package storage is busy; retry shortly.");
await new Promise<void>((resolve, reject) => {
const ready = () => { signal.removeEventListener("abort", aborted); resolve(); };
const aborted = () => {
const index = queue.waiting.indexOf(ready);
if (index >= 0) queue.waiting.splice(index, 1);
reject(signal.reason);
};
queue.waiting.push(ready);
signal.addEventListener("abort", aborted, { once: true });
});
} else queue.active++;
try { signal.throwIfAborted(); return await action(); }
finally { const next = queue.waiting.shift(); if (next) next(); else queue.active--; }
}
const manifests = new Map<string, { manifest: SkillPackageManifest; bytes: number }>();
let manifestBytes = 0;
const MAX_MANIFEST_CACHE = 16 * 1024 * 1024;
function rememberManifest(key: string, manifest: SkillPackageManifest, bytes: number) {
if (bytes > MAX_MANIFEST_CACHE) return;
const previous = manifests.get(key);
if (previous) manifestBytes -= previous.bytes;
manifests.delete(key);
while (manifests.size && manifestBytes + bytes > MAX_MANIFEST_CACHE) {
const first = manifests.keys().next().value!;
manifestBytes -= manifests.get(first)!.bytes;
manifests.delete(first);
}
manifests.set(key, { manifest, bytes });
manifestBytes += bytes;
}
export function skillPackagePaths(companyId: string, snapshotId: string) {
if (![companyId, snapshotId].every(id => /^[a-f0-9-]{36}$/i.test(id))) throw new Error("Invalid skill package identity");
const parent = path.join(resolvePaperclipInstanceRoot(), "skills", companyId, "__packages__");
return { parent, directory: path.join(parent, snapshotId), manifest: path.join(parent, `${snapshotId}.json`) };
}
async function readBoundedObject(storage: StorageService, companyId: string, key: string, length: number, hash: string, signal: AbortSignal) {
if (!Number.isSafeInteger(length) || length < 0 || length > MAX_SKILL_MANIFEST_BYTES) throw new Error("Invalid skill manifest size");
const object = await storage.getObject(companyId, key, { signal });
const chunks: Buffer[] = [];
let bytes = 0;
try {
for await (const chunk of object.stream) {
const buffer = Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk);
bytes += buffer.length;
if (bytes > length) throw new Error("Skill manifest exceeds its recorded size");
chunks.push(buffer);
}
} finally { object.stream.destroy(); }
const result = Buffer.concat(chunks);
if (bytes !== length || packageSha256(result) !== hash) throw new Error("Skill manifest checksum mismatch");
return result;
}
/** Owns immutable package storage. Authorization belongs to the referencing
* skill/source, and must be checked even when package bytes are already cached.
* Transaction readers supply their originating DB as cacheOwner for dedicated
* cache locks; metadata queries remain on the caller's transaction. */
export function skillPackageStore(db: Db, suppliedStorage?: StorageService, cacheOwner: Db = db) {
const storage = () => suppliedStorage ?? getStorageService();
const scope = (companyId: string, id: string) => and(eq(snapshots.companyId, companyId), eq(snapshots.id, id));
async function descriptor(companyId: string, id: string): Promise<Descriptor> {
const load = async () => {
const [row] = await db.select(descriptorColumns).from(snapshots).where(scope(companyId, id));
if (!row || (!row.storage && !row.restorable && row.upload && !row.upload.legacy)) throw notFound("Installed skill package is unavailable");
return row;
};
const context = readScope.getStore();
if (context?.db !== db) return load();
const key = `${companyId}:${id}`;
if (!context.descriptors.has(key)) context.descriptors.set(key, load());
return context.descriptors.get(key)!;
}
async function version(companyId: string, skillId: string, id: string): Promise<StoredSkillVersion | null> {
const context = readScope.getStore();
const cached = context?.db === db ? context.versions.get(`${companyId}:${id}`) : null;
if (cached) return cached.companySkillId === skillId ? cached : null;
const [row] = await db.select(skillVersionMetadataColumns).from(companySkillVersions).where(and(
eq(companySkillVersions.companyId, companyId), eq(companySkillVersions.companySkillId, skillId), eq(companySkillVersions.id, id)));
return row ?? null;
}
async function prefetch(companyId: string, versionIds: string[]) {
const context = readScope.getStore();
if (context?.db !== db || versionIds.length < 2) return;
const rows = await db.select(skillVersionMetadataColumns).from(companySkillVersions).where(and(
eq(companySkillVersions.companyId, companyId), inArray(companySkillVersions.id, [...new Set(versionIds)])));
for (const row of rows) context.versions.set(`${companyId}:${row.id}`, row);
const ids = [...new Set(rows.flatMap(row => [row.repositorySnapshotId, row.packageSnapshotId]).filter((id): id is string => Boolean(id)))];
if (!ids.length) return;
const packages = await db.select(descriptorColumns).from(snapshots).where(and(eq(snapshots.companyId, companyId), inArray(snapshots.id, ids)));
for (const row of packages) if (row.storage || row.restorable || !row.upload || row.upload.legacy) {
context.descriptors.set(`${companyId}:${row.id}`, Promise.resolve(row));
}
}
async function restoreRecipe(row: Descriptor, database = db) {
const [saved] = await database.select({ restore: snapshots.restore }).from(snapshots).where(scope(row.companyId, row.id));
const recipe = saved?.restore;
if (!recipe || recipe.version !== 1) throw notFound("This skill package has no exact-commit restore recipe");
const url = new URL(recipe.repositoryUrl);
if (!/^[a-f0-9]{40}$/i.test(recipe.commitSha) || !recipe.sourceId || !recipe.repositoryId
|| url.protocol !== "https:" || url.hostname !== "github.com" || url.username || url.password
|| !/^\/[^/]+\/[^/]+$/.test(url.pathname) || url.search || url.hash) {
throw new Error("Invalid immutable skill package recipe");
}
if (recipe.skillPath) assertSkillSnapshotPath(recipe.skillPath);
const manifest = validateSkillPackageManifest(recipe.manifest);
if (manifest.contentHash !== row.contentHash || (row.storage && packageSha256(skillPackageManifestBytes(manifest)) !== row.storage.manifestSha256)) {
throw new Error("Skill package restore recipe does not match its receipt");
}
return recipe;
}
async function acquireFiles(row: Descriptor, signal: AbortSignal, database = db) {
signal.throwIfAborted();
const [current] = await database.select({ retryAt: snapshots.restoreRetryAt }).from(snapshots).where(scope(row.companyId, row.id));
if (current?.retryAt && current.retryAt.getTime() > Date.now()) throw conflict("The original skill package could not be acquired. Reconnect its repository and retry after the recovery backoff.");
try {
const recipe = await restoreRecipe(row, database);
const read = await skillPackageRecoveryReader(database, row.companyId, recipe);
const files = await recoverSkillPackage(recipe, read, signal);
await database.update(snapshots).set({ restoreRetryAt: null }).where(scope(row.companyId, row.id));
return files;
} catch (error) {
if (!signal.aborted) await database.update(snapshots).set({ restoreRetryAt: new Date(Date.now() + 5 * 60_000) }).where(scope(row.companyId, row.id));
throw error;
}
}
async function legacyFiles(row: Descriptor, database = db) {
const [legacy] = await database.select({ files: snapshots.fileInventory, storage: snapshots.storage, upload: snapshots.upload })
.from(snapshots).where(scope(row.companyId, row.id));
if (!legacy || (!legacy.files.length && !legacy.storage)) throw notFound("Installed skill package has no saved contents");
// A reader may have loaded its descriptor just before a migration clears
// the legacy payload. Upgrade this request's descriptor to the published
// receipt instead of reporting a missing historical version.
if (legacy.storage) { row.storage = legacy.storage; row.upload = legacy.upload; }
return legacy.files;
}
async function manifestFor(row: Descriptor, cancellation?: AbortSignal): Promise<SkillPackageManifest> {
cancellation?.throwIfAborted();
const paths = skillPackagePaths(row.companyId, row.id);
const key = `${paths.manifest}:${row.storage?.manifestSha256 ?? row.contentHash}`;
const cached = manifests.get(key);
if (cached) return cached.manifest;
const load = async (sharedSignal: AbortSignal) => {
const signal = AbortSignal.any([sharedSignal, AbortSignal.timeout(180_000)]);
await ensureSkillCacheDirectory(paths.parent, true);
// Legacy manifests have no independently stored checksum. Derive them once
// per process from the authoritative row until the operator migrates it.
let bytes: Buffer | null = null;
if (row.storage) {
try {
const file = await fs.open(paths.manifest, constants.O_RDONLY | constants.O_NOFOLLOW);
try {
const stat = await file.stat();
if (stat.isFile() && stat.size <= MAX_SKILL_MANIFEST_BYTES) bytes = await file.readFile();
} finally { await file.close(); }
} catch (error) {
if (!["ENOENT", "ELOOP", "EISDIR"].includes((error as NodeJS.ErrnoException).code ?? "")) throw error;
}
}
if (row.storage) {
if (!bytes || packageSha256(bytes) !== row.storage.manifestSha256) {
try {
if (row.storage.provider !== storage().provider) throw new Error("Skill package storage provider changed");
bytes = await withPackageIo(() => readBoundedObject(storage(), row.companyId, row.storage!.manifestKey,
row.storage!.manifestBytes, row.storage!.manifestSha256, signal), signal);
} catch (error) {
signal.throwIfAborted();
const saved = await withPackageIo(() => withDedicatedDbConnection(cacheOwner, async database => {
const [legacy] = await database.select({ files: snapshots.fileInventory }).from(snapshots).where(scope(row.companyId, row.id));
return legacy?.files.length ? skillPackageManifestBytes(skillPackageManifest(legacy.files)) : null;
}), signal);
if (saved && packageSha256(saved) === row.storage!.manifestSha256) bytes = saved;
else if (row.restorable) bytes = skillPackageManifestBytes((await restoreRecipe(row)).manifest);
else throw error;
}
}
} else if (row.restorable) {
bytes = skillPackageManifestBytes((await restoreRecipe(row)).manifest);
} else if (!bytes) {
bytes = await withPackageIo(() => withDedicatedDbConnection(cacheOwner, async database => {
const files = await legacyFiles(row, database);
return row.storage ? null : skillPackageManifestBytes(skillPackageManifest(files));
}), signal);
if (!bytes) return manifestFor(row, signal);
}
signal.throwIfAborted();
if (bytes.length > MAX_SKILL_MANIFEST_BYTES) throw new Error("Skill manifest exceeds its byte bound");
const manifest = validateSkillPackageManifest(JSON.parse(bytes.toString("utf8")));
if (row.storage && (manifest.contentHash !== row.contentHash || manifest.files.length !== row.storage.fileCount
|| manifest.totalBytes !== row.storage.totalBytes)) throw new Error("Skill package receipt does not match its manifest");
await ensureSkillCacheDirectory(paths.parent, true);
const temp = `${paths.manifest}.${randomUUID()}`;
try { await fs.writeFile(temp, bytes, { flag: "wx", mode: 0o600, signal }); signal.throwIfAborted(); await fs.rename(temp, paths.manifest); }
finally { await fs.rm(temp, { force: true }); }
rememberManifest(key, manifest, bytes.length);
return manifest;
};
return sharedFlight(db, `manifest:${key}`, cancellation, load);
}
async function restoreInto(row: Descriptor, manifest: SkillPackageManifest, staging: string,
archivePath: string, signal: AbortSignal, allowRecovery = true, database = db) {
let legacy: CompanySkillVersionFileInventoryEntry[] | null = null;
if (!row.storage && !row.restorable) legacy = await legacyFiles(row, database);
try {
if (row.storage) {
if (row.storage.provider !== storage().provider) throw new Error("Skill package storage provider changed");
const object = await storage().getObject(row.companyId, row.storage.archiveKey, { signal });
const digest = boundedHashStream(row.storage.archiveBytes);
await pipeline(object.stream, digest.stream, createWriteStream(archivePath, { flags: "wx", mode: 0o600 }), { signal });
const actual = digest.result();
if (actual.bytes !== row.storage.archiveBytes || actual.sha256 !== row.storage.archiveSha256) throw new Error("Skill package archive checksum mismatch");
await extractSkillPackageArchive(archivePath, staging, manifest, signal);
}
} catch (error) {
signal.throwIfAborted();
// A missing/corrupt archive must not hide a still-authoritative legacy
// body. Only then use the saved, authorized immutable Git recipe.
const [saved] = await database.select({ files: snapshots.fileInventory }).from(snapshots).where(scope(row.companyId, row.id));
if (saved?.files.length && skillPackageManifest(saved.files).contentHash === row.contentHash) legacy = saved.files;
else if (row.restorable && allowRecovery) legacy = await acquireFiles(row, signal, database);
else throw error;
await fs.rm(staging, { recursive: true, force: true });
await fs.mkdir(staging);
}
if (!row.storage && row.restorable && !legacy) {
if (!allowRecovery) throw notFound("Skill package bytes are not stored");
legacy = await acquireFiles(row, signal, database);
}
if (legacy) {
for (const file of legacy) {
signal.throwIfAborted();
const target = path.join(staging, assertSkillSnapshotPath(file.path));
await fs.mkdir(path.dirname(target), { recursive: true, mode: 0o700 });
await fs.writeFile(target, skillFileBytes(file), { flag: "wx", mode: file.executable ? 0o555 : 0o444 });
}
}
if (!await verifySkillPackageDirectory(staging, manifest.files)) throw new Error("Restored skill package does not match its manifest");
}
// Archive reads and verification own private scratch space outside the warm
// preparation lock. Administrative reads briefly try that lock to publish.
async function withIsolatedPackage<T>(companyId: string, id: string,
consume: (root: string, manifest: SkillPackageManifest) => Promise<T>, allowRecovery = true, cancellation?: AbortSignal, cacheRead = false) {
const signal = cancellation ? AbortSignal.any([cancellation, AbortSignal.timeout(180_000)]) : AbortSignal.timeout(180_000);
signal.throwIfAborted();
// Preserve the caller's transaction view before taking an I/O permit. All
// later SQL uses a private connection, so a queued transaction cannot hold
// the ordinary pool connection needed by an active cold response.
const row = await descriptor(companyId, id);
const manifest = await manifestFor(row, signal);
const read = () => withDedicatedDbConnection(cacheOwner, database =>
readIsolatedPackage(row, manifest, consume, allowRecovery, signal, cacheRead, database));
return cacheRead ? withPackageIo(read, signal, coldResponses) : read();
}
async function readIsolatedPackage<T>(row: Descriptor, manifest: SkillPackageManifest,
consume: (root: string, manifest: SkillPackageManifest) => Promise<T>, allowRecovery: boolean, signal: AbortSignal, cacheRead: boolean,
database: Db) {
const { companyId, id } = row;
let published = false;
const result = await withPackageIo(async () => {
const parent = skillPackagePaths(companyId, id).parent;
await ensureSkillCacheDirectory(parent, true);
const staging = await fs.mkdtemp(path.join(parent, ".verify-"));
const archive = `${staging}.tar.gz`;
try {
await restoreInto(row, manifest, staging, archive, signal, allowRecovery, database);
const result = await consume(staging, manifest);
if (cacheRead) {
// Network, extraction, hashing and response reads all happen before
// this short publication lock. Busy preparation keeps its priority.
const paths = skillPackagePaths(companyId, id);
const previous = await fs.mkdtemp(path.join(parent, ".verify-"));
try {
await withSkillVersionCacheLock(cacheOwner, companyId, async () => {
let moved = false;
try { await fs.rename(paths.directory, previous); moved = true; }
catch (error) { if ((error as NodeJS.ErrnoException).code !== "ENOENT") throw error; }
try { await fs.rename(staging, paths.directory); }
catch (error) { if (moved) await fs.rename(previous, paths.directory); throw error; }
relocateSkillPackageVerification(staging, paths.directory, manifest.files);
touchSkillPackage(paths.directory);
published = true;
}, true);
} catch (error) { if (!(error instanceof SkillCacheBusy)) throw error; }
finally { await fs.rm(previous, { recursive: true, force: true }); }
}
return result;
} finally { await fs.rm(staging, { recursive: true, force: true }); await fs.rm(archive, { force: true }); }
}, signal);
if (published) {
// Verify possible eviction candidates outside both the preparation lock
// and the storage slot used for this fill; take the lock only to remove.
const reader = skillPackageStore(database, storage(), cacheOwner);
await trimSkillPackageCache(database, companyId, skillPackagePaths(companyId, id).directory,
(id, signal) => reader.verifyDurable(companyId, id, signal).catch(() => false),
async remove => {
try { await withSkillVersionCacheLock(cacheOwner, companyId, remove, true); }
catch (error) { if (!(error instanceof SkillCacheBusy)) throw error; }
}, signal);
}
signal.throwIfAborted();
return result;
}
async function materialize(row: Descriptor, manifest: SkillPackageManifest) {
const paths = skillPackagePaths(row.companyId, row.id);
return flight(db, `files:${paths.directory}`, async () => {
await ensureSkillCacheDirectory(paths.parent, true);
if (await verifySkillPackageDirectory(paths.directory, manifest.files)) return { directory: paths.directory, created: false };
const signal = AbortSignal.timeout(180_000);
return withPackageIo(async () => {
const staging = await fs.mkdtemp(path.join(paths.parent, ".package-"));
const archivePath = `${staging}.tar.gz`;
try {
await restoreInto(row, manifest, staging, archivePath, signal);
// All consumers hold the company cache lock, including processes sharing a volume.
await fs.rm(paths.directory, { recursive: true, force: true });
await fs.rename(staging, paths.directory);
relocateSkillPackageVerification(staging, paths.directory, manifest.files);
return { directory: paths.directory, created: true };
} finally { await fs.rm(staging, { recursive: true, force: true }); await fs.rm(archivePath, { force: true }); }
}, signal);
});
}
async function withPackage<T>(companyId: string, id: string, consume: (root: string, manifest: SkillPackageManifest) => Promise<T>) {
const row = await descriptor(companyId, id);
const manifest = await manifestFor(row);
return withSkillVersionCacheLock(db, companyId, async () => {
const restored = await materialize(row, manifest);
touchSkillPackage(restored.directory);
if (restored.created) await trimSkillPackageCache(db, companyId, restored.directory,
(id, signal) => verifyDurable(companyId, id, signal).catch(() => false));
return consume(restored.directory, manifest);
});
}
async function files(companyId: string, id: string, isolated = false): Promise<CompanySkillVersionFileInventoryEntry[]> {
return (isolated ? withIsolatedPackage : withPackage)(companyId, id, async (root, manifest) => {
const result: CompanySkillVersionFileInventoryEntry[] = [];
for (const entry of manifest.files) result.push({ path: entry.path, kind: entry.kind, content: (await fs.readFile(path.join(root, entry.path))).toString(entry.encoding),
...(entry.encoding === "base64" ? { encoding: entry.encoding } : {}), ...(entry.executable ? { executable: true } : {}) });
return result;
});
}
async function prepare(companyId: string, files: CompanySkillVersionFileInventoryEntry[], migrateLegacy = false, cancellation?: AbortSignal, restore?: SkillPackageRestoreRecipe): Promise<string> {
cancellation?.throwIfAborted();
const manifest = skillPackageManifest(files);
if (restore && !skillPackageManifestBytes(restore.manifest).equals(skillPackageManifestBytes(manifest))) throw new Error("Restore recipe does not describe these package bytes");
const preparedId = await flight(db, `prepare:${companyId}:${manifest.contentHash}`, async () => {
const signal = cancellation ? AbortSignal.any([cancellation, AbortSignal.timeout(180_000)]) : AbortSignal.timeout(180_000);
const token = randomUUID();
const prefix = `${companyId}/skill-packages/${token}`;
const upload = { token, legacy: false, provider: storage().provider, archiveKey: `${prefix}/files.tar.gz`, manifestKey: `${prefix}/manifest.json`, expiresAt: new Date(Date.now() + 10 * 60_000).toISOString() };
const [created] = await db.transaction(async tx => {
await notifyDeliveryWork(tx, DELIVERY_QUEUES.skillPackages);
return tx.insert(snapshots).values({ companyId, contentHash: manifest.contentHash, upload, restore,
migrationAt: null, maintenanceAt: new Date(Date.now() + 24 * 60 * 60_000) }).onConflictDoNothing().returning(descriptorColumns);
});
let existing = created ?? (await db.select(descriptorColumns).from(snapshots).where(and(eq(snapshots.companyId, companyId), eq(snapshots.contentHash, manifest.contentHash))))[0]!;
if (!created) {
if (existing.storage || (!existing.upload && !migrateLegacy && !existing.restorable)) {
// Content deduplication must not attach a recipe with different kind or
// encoding metadata to an already published manifest.
if (restore && !existing.restorable && existing.storage?.manifestSha256 === packageSha256(skillPackageManifestBytes(manifest))) {
await db.update(snapshots).set({ restore }).where(and(scope(companyId, existing.id), sql`${snapshots.restore} is null`));
}
return existing.id;
}
// A stale claim is replaced atomically; the DB trigger retains its exact
// keys in the deletion outbox. Never reuse keys from another attempt.
upload.legacy = !existing.upload || Boolean(existing.upload.legacy);
const [claimed] = await db.update(snapshots).set({ upload }).where(and(scope(companyId, existing.id),
sql`${snapshots.storage} is null`,
sql`(${snapshots.upload} is null or (${snapshots.upload}->>'expiresAt')::timestamptz <= now())`)).returning(descriptorColumns);
if (!claimed) throw conflict("This skill package is being saved. Retry after its upload settles.");
existing = claimed;
}
return withPackageIo(async () => {
const root = skillPackagePaths(companyId, existing.id).parent;
await ensureSkillCacheDirectory(root, true);
const staging = await fs.mkdtemp(path.join(root, ".upload-"));
const archivePath = path.join(staging, "files.tar.gz");
const manifestBuffer = skillPackageManifestBytes(manifest);
if (manifestBuffer.length > MAX_SKILL_MANIFEST_BYTES) throw new Error("Skill manifest exceeds its byte bound");
const backend = storage();
try {
const archive = await writeSkillPackageArchive(archivePath, files, signal);
await backend.putFile({ companyId, namespace: "skill-packages", objectKey: upload.archiveKey, originalFilename: "files.tar.gz",
contentType: "application/gzip", body: createReadStream(archivePath), byteSize: archive.bytes, sha256: archive.sha256, signal });
await backend.putFile({ companyId, namespace: "skill-packages", objectKey: upload.manifestKey, originalFilename: "manifest.json",
contentType: "application/json", body: manifestBuffer, signal });
const receipt: SkillPackageStorageReceipt = { format: "tar-gzip-v1", provider: backend.provider, ...(!restore ? { durable: true } : {}),
archiveKey: upload.archiveKey, archiveSha256: archive.sha256, archiveBytes: archive.bytes,
manifestKey: upload.manifestKey, manifestSha256: packageSha256(manifestBuffer), manifestBytes: manifestBuffer.length,
fileCount: manifest.files.length, totalBytes: manifest.totalBytes };
// A complete read verifies storage independently before publishing any reference.
await readBoundedObject(backend, companyId, receipt.manifestKey, receipt.manifestBytes, receipt.manifestSha256, signal);
const verify = await backend.getObject(companyId, receipt.archiveKey, { signal });
const digest = boundedHashStream(receipt.archiveBytes);
const downloaded = path.join(staging, "verified.tar.gz");
await pipeline(verify.stream, digest.stream, createWriteStream(downloaded, { flags: "wx", mode: 0o600 }), { signal });
const actual = digest.result();
if (actual.bytes !== receipt.archiveBytes || actual.sha256 !== receipt.archiveSha256) throw new Error("Uploaded skill package checksum mismatch");
const restored = path.join(staging, "restored");
await fs.mkdir(restored);
await extractSkillPackageArchive(downloaded, restored, manifest, signal);
signal.throwIfAborted();
const [saved] = await db.transaction(async tx => {
await notifyDeliveryWork(tx, DELIVERY_QUEUES.skillPackages);
return tx.update(snapshots).set({ storage: receipt, upload: null, ...(restore ? { restore } : {}) })
.where(and(scope(companyId, existing.id), sql`${snapshots.upload}->>'token' = ${token}`)).returning({ id: snapshots.id });
});
if (!saved) throw conflict("Skill package upload ownership changed");
return saved.id;
} finally {
// Failed/ambiguous uploads retain their durable claim. Maintenance reconciles
// exact owned keys; never delete a successful object on an uncertain SQL result.
await fs.rm(staging, { recursive: true, force: true });
}
}, signal);
});
{
const [retained] = await db.update(snapshots).set({ maintenanceAt: new Date(Date.now() + 24 * 60 * 60_000),
...(!restore ? { storage: sql`jsonb_set(${snapshots.storage}, '{durable}', 'true'::jsonb)` } : {}) })
.where(and(scope(companyId, preparedId), sql`${snapshots.storage} is not null`)).returning({ id: snapshots.id });
if (!retained) return prepare(companyId, files, true, cancellation, restore);
}
return preparedId;
}
async function bindRestore(companyId: string, id: string, restore: SkillPackageRestoreRecipe) {
const manifest = validateSkillPackageManifest(restore.manifest);
// Called in the source publication transaction. A failed import must not
// leave a later successful import bound to an uncommitted source identity.
await db.update(snapshots).set({ restore }).where(and(scope(companyId, id),
eq(snapshots.contentHash, manifest.contentHash),
sql`${snapshots.storage}->>'manifestSha256' = ${packageSha256(skillPackageManifestBytes(manifest))}`,
sql`exists (select 1 from ${companySkillSources} s where s.company_id = ${companyId}
and s.id = ${restore.sourceId} and s.removed_at is null)`,
sql`not exists (select 1 from ${companySkillSources} s where s.company_id = ${companyId}
and s.id::text = ${snapshots.restore}->>'sourceId' and s.removed_at is null)`));
}
async function verifyDurable(companyId: string, id: string, cancellation?: AbortSignal,
validate?: (root: string, manifest: SkillPackageManifest) => Promise<boolean>): Promise<boolean> {
const action = async () => {
cancellation?.throwIfAborted();
const row = await descriptor(companyId, id);
if (!row.storage) {
await manifestFor(row, cancellation);
if (!row.storage) return validate ? withIsolatedPackage(companyId, id, validate, false, cancellation) : true;
}
if (row.storage.provider !== storage().provider) throw new Error("Skill package storage provider changed; migrate its objects before switching storage");
const signal = cancellation ? AbortSignal.any([cancellation, AbortSignal.timeout(180_000)]) : AbortSignal.timeout(180_000);
return withPackageIo(async () => {
const receipt = row.storage!;
const bytes = await readBoundedObject(storage(), companyId, receipt.manifestKey, receipt.manifestBytes, receipt.manifestSha256, signal);
const manifest = validateSkillPackageManifest(JSON.parse(bytes.toString("utf8")));
if (manifest.contentHash !== row.contentHash) return false;
const parent = skillPackagePaths(companyId, id).parent;
await ensureSkillCacheDirectory(parent, true);
const staging = await fs.mkdtemp(path.join(parent, ".verify-"));
try {
const object = await storage().getObject(companyId, receipt.archiveKey, { signal });
const digest = boundedHashStream(receipt.archiveBytes);
const archive = path.join(staging, "files.tar.gz");
await pipeline(object.stream, digest.stream, createWriteStream(archive, { flags: "wx", mode: 0o600 }), { signal });
const actual = digest.result();
if (actual.bytes !== receipt.archiveBytes || actual.sha256 !== receipt.archiveSha256) return false;
const restored = path.join(staging, "files");
await fs.mkdir(restored);
await extractSkillPackageArchive(archive, restored, manifest, signal);
return validate ? await validate(restored, manifest) : true;
} finally { await fs.rm(staging, { recursive: true, force: true }); }
}, signal);
};
const context = readScope.getStore();
const key = `${companyId}:${id}`;
if (context?.db !== db || validate || cancellation) return action();
if (!context.durable.has(key)) context.durable.set(key, action());
return context.durable.get(key)!;
}
async function readPackageFile(root: string, entry: SkillPackageManifest["files"][number], signal?: AbortSignal) {
signal?.throwIfAborted();
const filename = path.join(root, entry.path);
await ensureSkillCacheDirectory(path.dirname(filename));
const handle = await fs.open(filename, constants.O_RDONLY | constants.O_NOFOLLOW);
try {
const stat = await handle.stat();
if (!stat.isFile() || stat.size !== entry.size) throw new Error("Invalid cached skill file");
const bytes = Buffer.alloc(entry.size);
let offset = 0;
while (offset < bytes.length) {
signal?.throwIfAborted();
const { bytesRead } = await handle.read(bytes, offset, bytes.length - offset, offset);
if (!bytesRead) throw new Error("Invalid cached skill file");
offset += bytesRead;
}
if (packageSha256(bytes) !== entry.sha256) throw new Error("Invalid cached skill file");
return bytes;
} finally { await handle.close(); }
}
async function versionFiles(version: StoredSkillVersion, allowRecovery = true, cancellation?: AbortSignal): Promise<CompanySkillVersionFileInventoryEntry[]> {
const signal = cancellation ? AbortSignal.any([cancellation, AbortSignal.timeout(180_000)]) : AbortSignal.timeout(180_000);
signal.throwIfAborted();
if (!version.packageSnapshotId) {
const [row] = await db.select({ ...skillVersionMetadataColumns, files: companySkillVersions.fileInventory }).from(companySkillVersions).where(and(
eq(companySkillVersions.companyId, version.companyId), eq(companySkillVersions.id, version.id)));
if (row?.packageSnapshotId) return versionFiles(row, allowRecovery, signal);
return row?.files ?? [];
}
const id = version.packageSnapshotId;
const manifest = await manifestFor(await descriptor(version.companyId, id), signal);
const read = async (root: string) => {
const byPath = new Map(manifest.files.map(file => [file.path, file]));
const result: CompanySkillVersionFileInventoryEntry[] = [];
for (const mapping of version.packageFileMap ?? manifest.files.map(file => ({ ...file, sourcePath: file.path }))) {
assertSkillSnapshotPath(mapping.path);
assertSkillSnapshotPath(mapping.sourcePath);
const file = byPath.get(mapping.sourcePath);
if (!file) throw new Error("Skill version references a missing package file");
const encoding = mapping.encoding ?? file.encoding;
result.push({ path: mapping.path, kind: mapping.kind,
content: (await readPackageFile(root, file, signal)).toString(encoding),
...(encoding === "base64" ? { encoding } : {}), ...(file.executable ? { executable: true } : {}) });
}
return result;
};
// Verify bytes as they are read, without holding the warm preparation lock.
// A descriptor survives eviction; a missing/racing file falls back to storage.
const root = skillPackagePaths(version.companyId, id).directory;
const result = await withPackageIo(() => read(root).catch(() => null), signal, inventoryIo);
if (result) {
touchSkillPackage(root);
return result;
}
signal.throwIfAborted();
const acquire = async () => {
const cached = await withPackageIo(() => read(root).catch(() => null), signal, inventoryIo);
return cached ?? withIsolatedPackage(version.companyId, id, read, allowRecovery, signal, true);
};
return cancellation ? acquire() : flight(db, `version-read:${version.companyId}:${version.id}:${allowRecovery}`, acquire);
}
async function readVersionFile(version: StoredSkillVersion, relativePath: string, allowRecovery = true) {
assertSkillSnapshotPath(relativePath);
if (!version.packageSnapshotId) return (await versionFiles(version, allowRecovery)).find(file => file.path === relativePath);
const id = version.packageSnapshotId;
const manifest = await manifestFor(await descriptor(version.companyId, id));
const mapping = version.packageFileMap?.find(file => file.path === relativePath);
if (version.packageFileMap && !mapping) return undefined;
const entry = manifest.files.find(file => file.path === (mapping?.sourcePath ?? relativePath));
if (!entry) return undefined;
const signal = AbortSignal.timeout(180_000);
const read = async (root: string) => {
const bytes = await readPackageFile(root, entry, signal);
const encoding = mapping?.encoding ?? entry.encoding;
return { path: relativePath, kind: mapping?.kind ?? entry.kind, content: bytes.toString(encoding),
encoding, executable: entry.executable };
};
// Immutable cache files can be read without the preparation lock. An open
// descriptor survives eviction; missing/racing files fall back to storage.
const cached = await withPackageIo(() => read(skillPackagePaths(version.companyId, id).directory).catch(() => null), signal, inventoryIo);
if (cached) return cached;
signal.throwIfAborted();
return flight(db, `file-read:${version.companyId}:${version.id}:${relativePath}:${allowRecovery}`,
async () => {
const cached = await withPackageIo(() => read(skillPackagePaths(version.companyId, id).directory).catch(() => null), signal, inventoryIo);
return cached ?? withIsolatedPackage(version.companyId, id, read, allowRecovery, signal, true);
});
}
async function versionInventory(version: StoredSkillVersion): Promise<Array<{ path: string; kind: CompanySkillVersionFileInventoryEntry["kind"] }>> {
if (version.packageSnapshotId) {
const index = version.packageFileMap ?? (await manifestFor(await descriptor(version.companyId, version.packageSnapshotId))).files;
return index.map(({ path, kind }) => ({ path, kind }));
}
const [row] = await db.select({ ...skillVersionMetadataColumns, files: sql<Array<{ path: string; kind: CompanySkillVersionFileInventoryEntry["kind"] }>>`
coalesce((select jsonb_agg(jsonb_build_object('path', f->>'path', 'kind', f->>'kind'))
from jsonb_array_elements(${companySkillVersions.fileInventory}) f), '[]'::jsonb)` })
.from(companySkillVersions).where(and(eq(companySkillVersions.companyId, version.companyId), eq(companySkillVersions.id, version.id)));
if (row?.packageSnapshotId) return versionInventory(row);
return row?.files ?? [];
}
return { descriptor, prepare, bindRestore, files, version, prefetch, versionFiles, versionInventory, readVersionFile, withPackage, withIsolatedPackage, verifyDurable,
recoveryRecipe: async (companyId: string, id: string) => restoreRecipe(await descriptor(companyId, id)),
manifest: async (companyId: string, id: string, cancellation?: AbortSignal) => manifestFor(await descriptor(companyId, id), cancellation) };
}