@runblackcat/sdk 0.1.0-alpha.100

package/LICENSE

@@ -0,0 +1,35 @@
+Blackcat is dual-licensed by component.
+
+SSPL-1.0 (Server Side Public License, version 1) — see LICENSE-SSPL
+  The server: everything that runs a fleet.
+    crates/blackcat-gateway
+    crates/blackcat-placer
+    crates/blackcat-agent
+    crates/blackcat-guest
+    crates/blackcat-local
+    crates/blackcat-aws
+    crates/blackcat-s3
+    crates/blackcat-plugin-supervisor
+    crates/blackcat-plugin-egress
+    crates/blackcat-github-actions
+    crates/blackcat-template-refresh
+
+Apache-2.0 — see LICENSE-APACHE
+  The clients and shared libraries, and everything not listed above.
+    crates/blackcat-cli (source)
+    crates/blackcat-core
+    crates/blackcat-mcp
+    crates/blackcat-plugin
+    crates/blackcat-plugin-sdk
+    sdk/ (the TypeScript SDK, @runblackcat/sdk)
+    the conformance suite, scripts, docs and examples
+
+Each crate's Cargo.toml states its license (SPDX: SSPL-1.0 or Apache-2.0).
+
+Binaries: the `blackcat` CLI binary also embeds SSPL server crates (the local
+fleet, node CA and maintenance executor link blackcat-gateway, blackcat-placer,
+blackcat-guest, blackcat-local and blackcat-s3), so the distributed CLI binary
+as a whole is subject to SSPL-1.0. The TypeScript SDK is Apache-2.0 throughout.
+
+Releases up to and including v0.1.0-alpha.81 were published under Apache-2.0
+and remain available under it.

package/README.md

@@ -0,0 +1,406 @@
+# blackcat
+
+A self-hosted [Firecracker](https://firecracker-microvm.github.io/) microVM fleet:
+multi-tenant sandboxes with snapshots, resumable workloads, and scale-to-zero.
+
+**Agnostic core, paved AWS path.** The core knows about Linux, not about any cloud. AWS is
+the one deployment path that comes fully assembled; everything else is possible with work.
+
+## Status: hosted development, pre-release
+
+The gateway, placer, node agent, guest, forkable checkpoint templates, durable
+processes, previews and the TypeScript SDK are implemented, and a hot claim from
+a checkpoint template restores in about 1.1 s on the development fleet. Nothing
+is shipped to customers or released as open source.
+
+What is missing before untrusted tenants can share workers is written down in
+[docs/SECURITY.md](docs/SECURITY.md); the ranked backlog is
+[docs/GROOMED.md](docs/GROOMED.md). Readiness levels and the product contract
+are in [docs/PRODUCT.md](docs/PRODUCT.md).
+
+## What it will do
+
+- **Sandbox API** — acquire an isolated microVM with retry-safe request identity
+  and a pinned immutable template version. Creation never blocks: the
+  request is admitted durably and answered at once, so a caller holds a claim
+  rather than a connection.
+- **Queue and backpressure** — a request the fleet has no room for waits durably
+  instead of failing, and is placed by a background drain as capacity arrives.
+  Fleet capacity and the durable queue are the only backpressure; there are no
+  per-tenant quotas.
+- **Demand-driven capacity** — the placer sizes the fleet from the backlog it
+  cannot place, through a provider seam, and retires idle workers back to zero.
+- **Forkable templates** — build a Docker image into a validated state, memory,
+  and disk checkpoint, then restore isolated writable clones with its declared
+  services already running. Sealed content-addressed chunks deduplicate within a
+  tenant and are fetched lazily into a bounded node cache.
+- **Snapshot / resume** — UFFD lazy paging, so a resumed sandbox pages memory in on demand
+  instead of waiting for a full restore.
+- **Scale to zero** — idle sandboxes snapshot to disk and release their RAM.
+  The next request resumes them when capacity is available.
+- **HTTP previews** — templates declare named ports; each sandbox gets its own
+  HTTPS hostname, with share-link or public access and optional custom domains.
+  See [HTTP previews and custom domains](docs/preview-domains.md).
+- **Multi-tenant** — tenants share hosts, isolated at the VM boundary by the Firecracker
+  jailer, seccomp, and explicit VM-to-VM network denial.
+
+## Design in one page
+
+| | |
+|---|---|
+| **Gateway** | Stateless. The single authenticated entry point; resolves sandbox→node and proxies. Scales horizontally. |
+| **Placer** | Single writer. Bin-packs by memory and vCPU, proposes placements, owns the queue and the node registry. Sole writer of substrate desired capacity. |
+| **Node agent** | One per host. Firecracker lifecycle, jailer, networking, UFFD handler. **Final authority on admission** — the placer proposes, the node accepts or rejects. |
+| **Guest agent** | Inside each sandbox, reached over vsock. Exec, file I/O, template-declared boot services. |
+| **Plugin supervisor (in development)** | Runs operator-installed, digest-pinned job-source plugins outside sandbox hosts. Each installation has its own identity, secret scope and sandbox grant. A small Rust SDK calls the scoped private API. |
+
+Nodes **self-register** with a join token — there is no host inventory file anywhere.
+Tenants are **API-managed rows**, not configuration, so an operator can provision
+and use one without restarting the service.
+
+No ZooKeeper, etcd, or Consul at any scale. Placement is sticky and sandboxes never move
+while running, so the problem partitions without consensus.
+
+## How it fits together
+
+```mermaid
+flowchart TB
+  CLI["blackcat-cli"]
+  SDK["TypeScript SDK"]
+
+  subgraph ctrl["Control host"]
+    direction LR
+    GW["blackcat-gateway<br/>auth · capabilities · routing<br/>listens :443"]
+    PL["blackcat-placer<br/>single writer · best-fit<br/>queue · node registry"]
+    DB[("placer.sqlite")]
+    GW <-->|"Unix socket"| PL
+    PL --- DB
+  end
+
+  subgraph plugins["Plugin host — planned integration"]
+    SUP["Plugin supervisor<br/>digest-pinned images · per-install identity"]
+    subgraph PIN["Per-install isolated bridge — no direct route"]
+      GHPL["GitHub Actions connector<br/>poll queued jobs · request JIT runners"]
+      CPX["Connector proxy<br/>GitHub DNS rules · pinned control target"]
+      GHPL -.->|"HTTPS CONNECT only"| CPX
+    end
+    EXT["Dedicated outbound bridge<br/>proxy only"]
+    RPX["Runner proxy<br/>GitHub DNS rules · no control target"]
+    SUP --> GHPL
+    SUP --> CPX
+    CPX -.-> EXT
+  end
+
+  SEC[("Plugin secret store<br/>local encrypted or AWS Secrets Manager")]
+  GITHUB["GitHub Actions<br/>job queue · results"]
+
+  OS[("Opaque object store<br/>S3 or filesystem")]
+
+  subgraph nodehost["Node host — one per machine, N machines"]
+    AG["blackcat-agent<br/>listens :8081"]
+    UF["UFFD handler<br/>one process per restore"]
+    FC["Firecracker + jailer"]
+    GU["blackcat-guest<br/>PID 1 inside the VM"]
+    RUN["Ephemeral Actions runner<br/>nonroot guest process"]
+    AG -->|spawns| UF
+    AG --> FC
+    UF -.->|"page faults"| FC
+    FC -->|vsock| GU
+    GU -->|"nonroot boot service"| RUN
+  end
+
+  CLI -->|HTTPS| GW
+  SDK -->|HTTPS| GW
+  CLI -.->|"admin plugin API"| GW
+  PL -.->|"installation state"| SUP
+  SUP -.->|"scoped Blackcat API"| GW
+  SUP -.->|"per-install secret refs"| SEC
+  EXT -.->|"GitHub API via proxy"| GITHUB
+  EXT -.->|"pinned private API via proxy"| GW
+  RPX -.->|"GitHub runner traffic"| GITHUB
+  RUN -.->|"one job + result"| RPX
+  GW ---|"metadata + grants"| OS
+  AG <-->|"short-lived object grants"| OS
+  GW -->|"node TLS"| AG
+  PL -->|"node TLS · propose"| AG
+  AG -->|"HTTPS · register, heartbeat, objects"| GW
+```
+
+Dashed plugin links show the intended integration, not a deployed runner. The
+connector observes external jobs and asks Blackcat for a sandbox; the external
+source remains the job and result authority. Runner code executes only inside a
+sandbox, without the connector's private key or plugin service credential. See
+the [plugin system contract](docs/plugin-system.md) for the current boundary.
+
+**Creating a sandbox.** The gateway checks the caller's capability token and opens a
+durable acquisition receipt keyed by `(tenant, request_id)`, so a retry is the same
+lease rather than a second VM. The placer picks candidate nodes by best fit over free
+memory and vCPU — snapshot-cache affinity first — and **proposes** to each in turn. The
+node accepts or rejects; it is the final admission authority, not the placer. On
+acceptance the agent starts Firecracker under the jailer, and `blackcat-guest` boots as
+PID 1 and serves the vsock protocol. Until the sandbox reports `placed`, the API answers
+`202 pending` and it is not ready for commands.
+
+Three things the picture above is drawn carefully to show:
+
+- **The placer talks to nodes itself.** It is not hidden behind the gateway — it opens
+  its own node-TLS connections to propose placements.
+- **Nodes receive only object-scoped grants.** The control plane authorizes a
+  specific artifact operation and the node transfers sealed bytes directly.
+  Nodes never receive a general object-store or tenant credential.
+- **There are two TLS trust domains**, not one. Clients and agents reach the gateway
+  under the control-plane CA; the gateway and placer reach node agents under a separate
+  node CA issuing 24-hour leaves.
+
+Gateway and placer speak over a Unix socket, so they are co-located on one host by
+construction. Everything else is a network hop.
+
+## Capacity and observability
+
+Creation is admitted durably and answered immediately. A request the fleet cannot
+place becomes a **queued acquisition** — it keeps its sandbox identity, and a background drain places it as capacity appears. Callers poll
+the claim they already hold:
+
+```
+POST /v1/sandboxes {request_id,template,lease:{ttl_secs,renewable_for_secs}}
+                                            -> 202 {"request_id":…, "id":…, "state":"pending", "expires_at":…}
+GET  /v1/sandboxes/acquisitions/{id}       -> 202 pending … -> 200 {"state":"placed","node":…}
+POST /v1/sandboxes/{id}/renew {ttl_secs}   -> 200 {"expires_at":…, "renewable_until":…}
+```
+
+Add `?wait_secs=N` (held at most 55 s) to the create or the lookup to have the gateway
+hold a still-pending claim and answer as soon as it is placed or fails; the
+placed receipt carries `ports`, so there is nothing to poll and no second call.
+A claim still pending when the hold ends is answered `pending`; ask again. It
+does not apply to a `count` batch: wait on each member by `request_id`.
+
+To start many identical sandboxes, add `"count": N` (1 to 1000) to the create.
+`request_id` then names the batch and member `i` is `request_id.i`. The whole batch
+is recorded in one durable transaction and answered `202` at once, with no
+connection held through restores; a retry returns the same members.
+
+```
+POST /v1/sandboxes {request_id,template,count,lease:{…}}
+                                            -> 202 {"request_id":…, "count":N, "counts":{"pending":N}, "members":[…]}
+GET  /v1/sandboxes/batches/{request_id}    -> 202 while any member is pending, then 200
+GET  /v1/sandboxes/acquisitions/{id}.{i}   -> one member, as an ordinary claim
+```
+
+`request_id` is caller-supplied and is the idempotency boundary, so repeating a
+create is always safe: it returns the same claim rather than allocating a second
+sandbox.
+
+Backpressure is fleet capacity and the durable queue; there are no per-tenant
+quotas. A queued acquisition that waits past the queue deadline is torn down by
+the ordinary expiry reconciler. Request-rate limits are operator-set fleet state
+(`blackcat admin rate-limits get|set`): session token-mint and OIDC limits are on by
+default, every workload limit is unlimited until set.
+
+The placer sizes the fleet from that backlog through the `CapacityProvider`
+seam, and is the **only** writer of substrate desired capacity — it raises the
+count from unplaceable demand and lowers it only by retiring workers it has
+proven empty. An external autoscaling policy driving the same knob would fight
+it, so deployments should cap the fleet (a maximum size) rather than scale it.
+
+`GET /metrics` exposes Prometheus text to the **admin credential**. The series
+are fleet-wide and keyed by tenant, so unlike `/v1/fleet` — which serves an
+`INSPECT` holder its own rows only — there is no tenant-scoped view to hand out.
+`/livez` and `/readyz` stay unauthenticated because they disclose nothing.
+
+```
+blackcat_demand_unsatisfied{tenant="…"}          queued acquisitions
+blackcat_demand_oldest_age_seconds               longest current wait
+blackcat_capacity_target_nodes                   workers last asked for
+blackcat_nodes{health="healthy"}                 workers actually registered
+blackcat_placement_rejected_total{reason="…"}    no_capacity
+blackcat_placements_total{host="…"}              kept (hosts scale-in keeps), elastic
+```
+
+A fleet pinned at its ceiling shows up as `blackcat_capacity_target_nodes`
+persistently above `blackcat_nodes`.
+
+## Recommended production shape: bare metal
+
+Run workers on bare metal, one or a few large `.metal` hosts, and keep nested virtualization (a VM with `/dev/kvm`) as the quick-start
+default: it launches in under a minute and needs no special quota. On metal every guest page fault costs about 7.6 times less than
+nested, so the same fleet restores 2 to 3 times faster, fans out 30 times faster at 100 clones, costs a third as much per clone and
+does not lose VMs to nested VM-entry failures.
+
+| | bare metal (`c7i.metal-24xl`) | nested (`m8i.4xlarge`) |
+|---|---|---|
+| create p50 | 29 ms | 60 ms |
+| first exec | 33 ms | 67 ms |
+| snapshot restore | 13 ms | 43 ms |
+| fan-out of 100 clones | 0.6 s | 17.9 s |
+| cost per clone-hour | $0.0043 | $0.0132 |
+| VM-entry deaths | 0 | 2.9 per 1,000 hot restores, 325 per 1,000 at density |
+
+(Bake-off, 2 vCPU / 1 GiB app template, `contrib/bench/bakeoff/RESULTS.md`, "Huge pages". The hosts are different CPU
+generations and metal has 12 times the CPUs, so the fan-out column is not virtualization alone. Metal is cheaper per clone because it
+carries about 1,000 clones where the nested host was measured at 64.)
+
+North star on one `c7i.metal-48xl` (500 clones of a built-from-source web app at once, each running a short test then destroyed):
+99% done 112 s after the PR arrived, all 500 passed with no replacements and no VMM deaths, about $0.34 per PR from a stopped node,
+against 300 s, 264 replacements and about $1.00 on five `m8i.8xlarge`.
+
+Trade-offs, plainly:
+
+* **Boot.** A metal reboot has been quoted at about 7.5 minutes. Measured on a `c7i.metal-48xl`: a fresh launch was registered with SSM
+  in 74 s and a stopped node was serving again 47 s after `StartInstances`, but stopping it took 320 s (unbilled). Time from a
+  planner launch to an enrolled node on metal has not been measured (see
+  `contrib/bench/results/2026-10-04-wave-c-metal-northstar.md`). Nested hosts register in 40 to 45 s.
+* **Coarse sizes.** Metal is `.metal` sizes only: 96 or 192 vCPUs on the `c7i` family. A small or spiky fleet pays for a whole box
+  or runs nested.
+* **Thin Spot capacity.** Metal Spot pools are small; the planner falls back to on-demand when one refuses, and the price you plan
+  with is the cheaper Spot one.
+* **Large blast radius.** One host carries up to about a thousand clones. Losing it, or restarting its agent, hits all of them; keep
+  `fleet max-clones-per-host` as the cap you can accept and keep more than one host for anything that must survive a host loss.
+
+The capacity planner prefers metal when a reservation's `needed_by` allows its boot time and the price per clone is lower
+([docs/capacity-planning.md](docs/capacity-planning.md)); `blackcat_vm_entry_deaths` counts the VM-entry failures that make nested
+hosts unreliable at density. Setup: [docs/bare-mode.md](docs/bare-mode.md#recommended-production-shape-bare-metal).
+
+## Install (alpha)
+
+Identical on bare metal and on AWS:
+
+```bash
+blackcat init --bind 0.0.0.0:8080 --hostname control.example
+# init installs and starts blackcat-placer and blackcat-gateway
+
+# Copy /etc/blackcat/tls/ca.pem to each node. Join tokens are single-use, so
+# mint a distinct token with `blackcat join-token` for each node.
+blackcat join https://control.example:8080 \
+  --token <node-1-token> --ca-cert /path/to/ca.pem
+blackcat join https://control.example:8080 \
+  --token <node-2-token> --ca-cert /path/to/ca.pem
+
+# Fleet and workload operations use a tenant capability token, not the admin
+# credential. Create a tenant/API key and mint that token through the API.
+blackcat fleet status --control-plane https://control.example:8080 \
+  --ca-cert /path/to/ca.pem --token <capability-token>
+blackcat template create --control-plane https://control.example:8080 \
+  --ca-cert /path/to/ca.pem --token <capability-token> --name base \
+  --docker <image> --vcpus <n> --memory-mib <m>
+blackcat sandbox create --control-plane https://control.example:8080 \
+  --ca-cert /path/to/ca.pem --token <capability-token> --template <t> \
+  --request-id <stable-acquisition-id> --ttl-secs 3600 \
+  --renewable-for-secs 86400
+```
+
+Create request IDs identify logical acquisitions: reuse the ID and exact
+parameters after an interrupted request, never for a new VM. A `pending`
+receipt is not ready for commands. EXEC also requires an explicit positive
+`--timeout-ms` (or JSON `timeout_ms`); timeout does not undo effects that already
+happened. These breaking pre-trial contracts are being verified together; old
+embedded guests and callers must be rebuilt/updated, not silently accepted.
+
+AWS adds exactly two things: an AMI that registers itself using the EC2 instance identity
+document, and a `CapacityProvider` that wraps an ASG. If AWS ever needs a *different*
+path, the design has failed — there is a test for that.
+
+**Bare-metal prerequisites:** Linux with `/dev/kvm`, root for the node agent, a hardened
+host kernel, and an uncompressed guest `vmlinux` ≥ 5.18. Releases ship a known-good guest
+kernel, because otherwise a new user's first task is a kernel build.
+Control-plane `blackcat init` additionally requires systemd and Python 3 at
+`/usr/bin/python3`. Init checks the interpreter before changing host state and
+installs the shared bounded placer-listener gate for every gateway start.
+The gate establishes startup ordering, not catalog or workload readiness.
+Filesystem template nodes also need `e2fsprogs` (`mkfs.ext4` and `resize2fs`)
+and GNU `cp` with sparse/reflink support. A compact template is verified before
+its private sandbox copy is grown to the declared writable capacity; missing
+growth tooling is a launch error, not permission to boot a smaller disk.
+
+Generic plain-Linux installs use the default-feature `blackcat-agent-x86_64-musl`,
+`blackcat-gateway-x86_64-musl`, and `blackcat-placer-x86_64-musl` release artifacts.
+The AWS AMI uses the corresponding `-aws` variants.
+
+## Authentication modes
+
+Private/VPC installations keep the existing static authentication mode: an
+administrator creates tenant API keys, each with a fixed operation set, and
+those keys mint short-lived Blackcat capabilities within that set. Optional hosted access adds external
+OIDC login for people while retaining those same capabilities as the workload
+authorization boundary. OIDC is not required for a private cluster, and enabling
+it does not expose admin or node authority.
+
+The context/login/claim CLI and the exact trust, storage and public deployment
+boundaries are specified in
+[docs/public-access-auth.md](docs/public-access-auth.md). Public reachability is
+network policy, not an application `public` switch.
+For an identity provider that requires an exact pre-registered loopback
+redirect, use `blackcat login --callback-port PORT` and register
+`http://127.0.0.1:PORT/callback` on its public OIDC client. The default login
+continues to choose a free ephemeral port.
+Hosted CLI sessions use a revocable 12-hour access credential and a distinct
+30-day, non-extendable refresh credential. The CLI renews and rotates both
+automatically before commands; browser login is required again at the absolute
+deadline or after revocation. See
+[the session lifecycle and tradeoff](docs/hosted-session-renewal.md). The
+[disposable dev trial runbook](docs/history/hosted-auth-dev-e2e.md#manual-cli-login-to-claim-window)
+includes a macOS Docker harness for a manual login-to-claim test; it is not a
+production deployment recipe.
+
+For a native, globally usable `blackcat` command on Apple-silicon macOS or
+Linux x86-64/arm64, run `curl -fsSL https://runblackcat.dev/install.sh | sh`;
+see [CLI installation](docs/cli-install.md). The installer uses release
+checksums from the public
+[`blackcat-releases`](https://github.com/TreebeardHQ/blackcat-releases) mirror
+and needs neither Docker nor Rust on the client. To run a fleet on plain
+Debian or Ubuntu servers, see [Bare mode](docs/bare-mode.md). The site itself lives in
+[`site/`](site/) and publishes an allowlisted subset of `docs/`.
+`blackcat template list` shows launchable templates to members without their
+source; `blackcat sandbox claim alpine-e2e-matched` resolves the latest version
+and pins its ID for safe retries. Intel Mac binaries are not published.
+
+## Documentation
+
+- **[docs/PRODUCT.md](docs/PRODUCT.md)** — the product contract and readiness levels.
+- **[docs/ARCHITECTURE.md](docs/ARCHITECTURE.md)** — processes, crates, trust domains, a sandbox's life.
+- **[docs/SECURITY.md](docs/SECURITY.md)** — isolation boundaries and known gaps.
+- **[docs/fleet-upgrades.md](docs/fleet-upgrades.md)** — stage, preview, run, and resume compatible fleet upgrades.
+- **[docs/GROOMED.md](docs/GROOMED.md)** — the ranked backlog.
+- **[AGENTS.md](AGENTS.md)** — how to work in this repository.
+- **[docs/forkable-templates.md](docs/forkable-templates.md)**,
+  **[docs/preview-domains.md](docs/preview-domains.md)**,
+  **[docs/public-access-auth.md](docs/public-access-auth.md)** — feature contracts.
+- **docs/history/** — the original plan, the progress ledger, trial acceptance
+  records and measurements. Rationale lives there; it is not required reading.
+
+## Prior art worth knowing
+
+[e2b-dev/infra](https://github.com/e2b-dev/infra) (Apache-2.0) solves much of this and is
+worth borrowing from — its AWS support is beta and it requires operating Nomad + Consul,
+which is the gap blackcat targets. Firecracker's own UFFD reference handler, the
+[rust-vmm](https://github.com/rust-vmm) crates, and Cloud Hypervisor are the other sources.
+
+## Contributing
+
+While the service is pre-0.01, **commit straight to `main`** — no PR review cycle,
+no branch discipline. Use conventional commit titles (`feat:`, `fix:`, `docs:`,
+`refactor:`, `test:`, `chore:`), and verify in proportion to risk before pushing.
+
+```sh
+git add <paths>
+git commit -m "fix(gateway): ..."
+git push origin main
+```
+
+Open a PR only when you want hosted CI to check something you cannot check
+locally — the Rust toolchain is pinned in `rust-toolchain.toml`, so a container
+gets you the same compiler without installing one:
+
+```sh
+docker run --rm -v "$PWD":/w -w /w rust:1.85.1 \
+  sh -c "cargo fmt --check && cargo clippy --workspace --all-targets -- -D warnings && cargo test"
+```
+
+This will tighten up before a real release. See [AGENTS.md](AGENTS.md) for the
+security and provider-boundary rules that do *not* relax.
+
+## License
+
+The server (gateway, placer, agent and the other fleet crates) is [SSPL-1.0](LICENSE-SSPL); the CLI
+source, the TypeScript SDK and the shared client libraries are [Apache-2.0](LICENSE-APACHE). The distributed
+`blackcat` CLI binary embeds server crates and is therefore subject to SSPL-1.0. See [LICENSE](LICENSE) for
+the per-crate split. Releases up to v0.1.0-alpha.81 remain available under Apache-2.0.

package/package.json

@@ -0,0 +1,29 @@
+{
+  "name": "@runblackcat/sdk",
+  "version": "0.1.0-alpha.100",
+  "description": "TypeScript SDK for the Blackcat sandbox gateway",
+  "license": "Apache-2.0",
+  "repository": "https://github.com/TreebeardHQ/blackcat",
+  "type": "module",
+  "exports": {
+    ".": {
+      "types": "./sdk/dist/index.d.ts",
+      "import": "./sdk/dist/index.js",
+      "default": "./sdk/dist/index.js"
+    }
+  },
+  "files": ["sdk/dist", "README.md", "LICENSE"],
+  "scripts": {
+    "build": "bun build sdk/src/index.ts --outdir sdk/dist --target node --format esm && tsc -p sdk/tsconfig.build.json",
+    "prepack": "bun run build",
+    "typecheck": "tsc -p sdk/tsconfig.json --noEmit",
+    "test": "bun test sdk/src/index.test.ts"
+  },
+  "devDependencies": {
+    "@types/bun": "1.3.12",
+    "typescript": "6.0.3"
+  },
+  "engines": {
+    "bun": ">=1.1"
+  }
+}