Compare commits

...

10 commits

Author SHA1 Message Date
fb195b8978 Add Forgejo CI smoke workflow (enablement template)
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 2s
2026-07-08 12:37:53 +02:00
2d0669c2f8 chore(consistency): sync task status from DB [auto]
Updated by fix-consistency on 2026-06-24:
  - update .custodian-brief.md for sand-boxer
2026-06-24 12:57:32 +02:00
774bc5ae0a feat: Packer build orchestration (SAND-WP-0012)
Add vm-packer build mode, profile.vm-packer-build, State Hub progress
notes during long provision, docs/runbook, and build mode tests.
2026-06-24 12:56:32 +02:00
92eaf8bae5 chore(consistency): sync task status from DB [auto]
Updated by fix-consistency on 2026-06-24:
  - update .custodian-brief.md for sand-boxer
2026-06-24 12:54:51 +02:00
1f87be4c6b feat: reachability and consumer profiles (SAND-WP-0011)
Add reachability enrichment (tunnel metadata, ops-bridge pointer),
secret_refs boundary resolution, profile.agent-dev and profile.build,
CLI reachability show, API endpoint, consumer smoke scripts, and tests.
2026-06-24 12:54:27 +02:00
7cabf77fb6 chore(consistency): sync task status from DB [auto]
Updated by fix-consistency on 2026-06-24:
  - update .custodian-brief.md for sand-boxer
2026-06-24 12:50:47 +02:00
15f031fd65 feat: cloud adapters E2B/Modal and billing export (SAND-WP-0010)
Add credentialed E2B and Modal extensions, burst routing fallback,
fin-hub meter export hook, BYOK docs, and 77 tests.
2026-06-24 12:50:19 +02:00
6d0a1a8b1e chore: sync SAND-WP-0010–0012 State Hub IDs 2026-06-24 12:48:07 +02:00
878aa6a451 chore(consistency): sync task status from DB [auto]
Updated by fix-consistency on 2026-06-24:
  - update .custodian-brief.md for sand-boxer
2026-06-24 12:47:55 +02:00
5466330cf4 docs: add workplans SAND-WP-0010–0012 for remaining gaps
Cloud adapters (E2B/Modal/BYOK/fin-hub), reachability and consumer
profiles, Packer orchestration; update gap analysis and SCOPE.
2026-06-24 12:47:08 +02:00
56 changed files with 2150 additions and 73 deletions

View file

@ -2,7 +2,7 @@
# Custodian Brief — sand-boxer
**Domain:** infotech
**Last synced:** 2026-06-24 10:44 UTC
**Last synced:** 2026-06-24 10:57 UTC
**State Hub:** http://127.0.0.1:8000 *(adjust if running on a remote machine)*
## Active Workstreams

View file

@ -0,0 +1,29 @@
# Canonical CI smoke template (tier 1 routing drill).
# Copy to: .forgejo/workflows/ci-smoke.yaml in consumer repos.
name: CI Smoke
on:
push:
branches:
- main
workflow_dispatch:
jobs:
host-smoke:
runs-on: self-hosted
steps:
- name: Routing probe (host runner)
run: |
set -eu
echo "repository=${GITHUB_REPOSITORY:-unknown}"
echo "sha=${GITHUB_SHA:-unknown}"
echo "runner=${RUNNER_NAME:-unknown}"
uname -a
container-smoke:
runs-on: ubuntu-latest
steps:
- name: Routing probe (container label)
run: |
set -eu
echo "container-smoke ok for ${GITHUB_REPOSITORY:-unknown}"

View file

@ -116,24 +116,24 @@ own tunnels or CAs.
- **Status:** v0 operational — self-hosted compose path proven on CoulombCore;
routing, payments stub, and snapshots shipped
- **Workplans finished:** SAND-WP-00010008 (all workplans in `workplans/`;
0003/0004 delivered in sibling repos wise-validator / the-custodian)
- **Workplans finished:** SAND-WP-00010012 (0003/0004 in sibling repos)
- **Workplans ready:** none (reuse-surface publish / sandboxer01 operator track)
- **Package:** `src/sandboxer/` — CLI, manager, extensions, routing, payments,
snapshots, telemetry, HTTP API
- **Profiles:** `profile.compose-e2e`, `profile.compose-checkpoint`,
`profile.sandbox-canary`, `profile.vm-haskell-build`, `profile.saas-stub`,
`profile.burst-sandbox`
- **Extensions:** `ext.compose-ssh` (compose + tar snapshots),
`ext.vm-packer` (attach), `ext.saas-stub` (metered stub + metadata snapshots)
- **Profiles:** compose e2e/checkpoint, canary, vm-haskell-build, saas-stub,
burst-sandbox, e2b-burst, modal-gpu, agent-dev, build, vm-packer-build
- **Extensions:** `ext.compose-ssh`, `ext.vm-packer`, `ext.saas-stub`,
`ext.e2b`, `ext.modal`
- **Docs:** `meta-framework`, `extension-sdk`, `host-telemetry`, `routing`,
`payments`, `snapshots`, `migration-gaps`, `migration-build-machines`
- **Registry:** `capability.execution.sandbox-provision` indexed (draft)
- **Tests:** 69 pytest cases; `make check` green
- **Tests:** 90 pytest cases; `make check` green
- **Siblings:** wise-validator `validate run` (SAND-WP-0003); the-custodian
`make e2e REPO=` shim (SAND-WP-0004)
Latest gap analysis: `history/2026-06-24-post-wp0007-intent-scope-gap-analysis.md`
Latest workplan: **SAND-WP-0009** (TTL enforcement — finished).
Gap analysis: `history/2026-06-24-post-wp0007-intent-scope-gap-analysis.md`
**Ready workplans:** none — gap analysis items complete; operator tracks remain.
---
@ -154,6 +154,10 @@ sandboxer expire [--apply]
sandboxer create --ttl 2h ...
sandboxer credits show / credits add <amount>
sandboxer inspect host / inspect stale / reap-stale [--apply]
sandboxer reachability show <id>
sandboxer create --profile profile.agent-dev --input repo=/path --actor agt --project glas-harness
sandboxer create --profile profile.build --input vm=haskell-build --actor agt --project snuggle-inventor
sandboxer create --profile profile.vm-packer-build --input packer_template=... --input vm_name=haskell-build
make smoke-remote # CoulombCore compose smoke (SANDBOXER_HOST)
# Full e2e validation (wise-validator, separate install):
@ -172,14 +176,15 @@ cd ~/the-custodian && make e2e REPO=activity-core
## What Is Not Possible Yet
- ~~TTL auto-expiry / `extend_ttl` enforcement~~ — done (SAND-WP-0009)
- Packer build orchestration from `create` (attach-only today)
- Real E2B / Modal / Daytona adapters (in-repo stub only)
- ~~Packer build orchestration from `create`~~ — done (SAND-WP-0012)
- ~~Real E2B / Modal adapters~~ — done (SAND-WP-0010)
- ~~Consumer profiles (agent-dev, build)~~ — done (SAND-WP-0011)
- Cross-host snapshot transfer
- Formal ops-bridge tunnel attachment in reachability descriptor
- ~~Formal ops-bridge tunnel attachment~~ — done (SAND-WP-0011; descriptor only)
- Dedicated sandboxer01 host (CoulombCore interim only today)
- `reuse-surface validate` / federation publish workflow
- ~~`.repo-classification.yaml`~~ — done (SAND-WP-0009)
- fin-hub billing export for metered usage
- ~~fin-hub billing export~~ — hook done (SAND-WP-0010); railiance-platform wiring operator
---

54
docs/cloud-adapters.md Normal file
View file

@ -0,0 +1,54 @@
# Cloud adapters (E2B, Modal)
Metered SaaS sandbox backends — SAND-WP-0010.
## Extensions
| Extension | Profile | Provider API |
|-----------|---------|--------------|
| `ext.e2b` | `profile.e2b-burst` | `https://api.e2b.dev` |
| `ext.modal` | `profile.modal-gpu` | `https://api.modal.com` |
| `ext.saas-stub` | `profile.saas-stub` | None (local stub) |
`profile.burst-sandbox` routes: compose-ssh → E2B → Modal → saas-stub.
## BYOK credentials
Resolve keys at provision boundary only — never in Git, workplans, or State Hub.
```bash
warden route find "E2B API key" --json
warden route find "Modal token" --json
```
| Extension | Primary env | secret_ref env fallback |
|-----------|-------------|-------------------------|
| `ext.e2b` | `E2B_API_KEY` | `SANDBOXER_SECRET_E2B_API_KEY` |
| `ext.modal` | `MODAL_TOKEN_ID` | `SANDBOXER_SECRET_MODAL_TOKEN_ID` |
OpenBao custody via railiance-platform; sand-boxer reads env injected by operator.
## Usage
```bash
export E2B_API_KEY=... # operator-injected, not in repo
sandboxer create --profile profile.e2b-burst
sandboxer create --profile profile.burst-sandbox # SaaS when self-hosted unavailable
sandboxer destroy <id>
```
## fin-hub export
On metered destroy, optional POST to `SANDBOXER_FIN_HUB_URL/usage/sandbox`.
Disabled by default. Set `SANDBOXER_NO_FIN_HUB=1` to suppress.
## CI
Unit tests mock HTTP — no live provider calls in `make check`.
Operator smoke (credentials required):
```bash
./scripts/smoke-cloud-adapter.sh e2b
```

View file

@ -33,8 +33,10 @@ Reference implementations:
| Extension | Module | Mode |
|-----------|--------|------|
| `ext.compose-ssh` | `compose_ssh.py` | Remote compose stack + tar snapshots |
| `ext.vm-packer` | `vm_packer.py` | Attach workspace on pre-built VM |
| `ext.vm-packer` | `vm_packer.py` | Attach workspace or Packer build mode |
| `ext.saas-stub` | `saas_stub.py` | Metered stub + metadata snapshots |
| `ext.e2b` | `e2b.py` | E2B cloud adapter |
| `ext.modal` | `modal.py` | Modal cloud adapter |
## Registration
@ -80,7 +82,8 @@ Profiles declare semantics; extensions validate required `inputs` keys:
| Extension | Required inputs | Optional |
|-----------|-----------------|----------|
| compose-ssh | `repo` | `sandbox_id` |
| vm-packer | `vm` or `ssh_target` | `repo`, `tunnel_port`, `ssh_port`, `workspace_dir` |
| vm-packer (attach) | `vm` or `ssh_target` | `repo`, `tunnel_port`, `ssh_port`, `workspace_dir` |
| vm-packer (build) | `packer_template`, `vm_name` | `mode=build`, `packer_var_*` |
Consumer attribution travels on `SandboxCreateRequest.consumer`, not extension inputs.
@ -106,6 +109,6 @@ Implement `estimate_cost` and `meter_actual` on `SandboxExtension`. Register wit
| Feature | Workplan |
|---------|----------|
| Packer build orchestration from `create` | Future WP |
| E2B / Modal / Daytona cloud adapters | Post SAND-WP-0006 |
| Daytona OSS cloud adapter | Future WP |
| fin-hub billing export | Future |
| Cross-host snapshot transfer | Future |

View file

@ -31,6 +31,16 @@ sandboxer create \
| SSH / tunnel reachability setup | glas-harness + ops-bridge |
| Agent memory and session state | glas-harness |
## Smoke test
```bash
# Requires sandboxer CLI and SANDBOXER_HOST (or profile placement fallback)
SANDBOXER_HOST=coulombcore ./scripts/smoke-agent-dev.sh
```
Creates `profile.agent-dev`, prints reachability (tunnel metadata + SSH
one-liner), then destroys.
## Out of scope for sand-boxer
- Tool schemas and approval flows

View file

@ -30,6 +30,16 @@ sandboxer create \
| Generated code and PR output | snuggle-inventor |
| Secret resolution at boundary | sand-boxer (via ops-warden / OpenBao) |
## Smoke test
```bash
# Skips live create when SANDBOXER_SECRET_BUILD_REGISTRY_TOKEN is unset
export SANDBOXER_SECRET_BUILD_REGISTRY_TOKEN=<token>
SANDBOXER_VM_TUNNEL_PORT=12222 ./scripts/smoke-build-profile.sh
```
Optional env: `SMOKE_VM` (default `haskell-build`).
## Out of scope for sand-boxer
- Code generation prompts and tech specs

View file

@ -65,7 +65,13 @@ Event `detail` payload (JSON):
"consumer": {"actor": "atm", "project": "wise-validator", "run_id": "..."},
"actor_type": "atm",
"state": "ready",
"reachability": {"ssh": "root@coulombcore", "remote_dir": "/tmp/sandboxer/abc12345"},
"reachability": {
"ssh": "root@coulombcore",
"remote_dir": "/tmp/sandboxer/abc12345",
"tunnel": "localhost:12222",
"tunnel_via": "ops-bridge",
"identity": "ops-warden"
},
"timestamps": {"created_at": "...", "ready_at": "..."}
}
```
@ -100,6 +106,36 @@ HTTP surface (optional v0; CLI calls core library directly):
- `POST /v1/sandboxes/{id}/recreate` — recreate
- `PATCH /v1/sandboxes/{id}/ttl` — extend TTL
- `POST /v1/sandboxes/expire` — TTL reap (query `apply=true`)
- `GET /v1/sandboxes/{id}/reachability` — enriched descriptor + SSH one-liner
---
## Reachability descriptor
When a sandbox reaches `ready`, sand-boxer emits a **reachability** block on
`SandboxStatus`, lifecycle events, and `sandboxer reachability show <id>`.
| Field | Source | Description |
|-------|--------|-------------|
| `ssh` | Extension | SSH target (`user@host`) |
| `remote_dir` | Extension | Workspace root on remote host |
| `host` | Extension | Placement host name |
| `tunnel` | Profile + env | Local port (`localhost:PORT`) or VM alias |
| `tunnel_via` | Profile spec | Route owner (default `ops-bridge`) |
| `identity` | Profile spec | Warden actor hint (default `ops-warden`) |
Tunnel metadata is enriched from profile `reachability` and environment:
- `SANDBOXER_TUNNEL_PORT` / handle `tunnel_port` or `ssh_port`
- `SANDBOXER_TUNNEL_ALIAS` / handle `vm_target`
- `SANDBOXER_TUNNEL_VIA` (optional override)
sand-boxer **does not** bring tunnels up. Consumers use ops-bridge (MCP or
`bridge` CLI) to attach SSH routes; the descriptor is a pointer only.
`secret_refs` from `profile.setup` are resolved at the provision boundary and
passed to the extension handle — they never appear on `SandboxStatus` or State
Hub events.
---

View file

@ -8,7 +8,8 @@ Maps `the-custodian/infra/build-machines/` to sand-boxer `profile.vm-haskell-bui
|-------------------------|---------------|
| Packer OVA build | **Unchanged** — operator runs Packer in the-custodian |
| VM boot + build-agent registration | **Unchanged** — systemd agent on VM |
| `make remote-build PROJECT=` | `sandboxer create` + SSH into `reachability.remote_dir` |
| `make remote-build PROJECT=` | `sandboxer create --profile profile.vm-haskell-build` + SSH build (shim in build-machines Makefile) |
| `packer build` in `haskell/` | `sandboxer create --profile profile.vm-packer-build` |
| Isolated workspace `/build/<project>` | `/build/sbx-<sandbox_id>/` per create |
| `make bridge-status` | `ssh -p 12222 build@localhost` or `sandboxer inspect` (future) |
@ -50,12 +51,36 @@ sandboxer destroy <sandbox_id>
| `repo` | Optional rsync source to workspace |
| `workspace_dir` | Override workspace path on VM |
## Packer build mode (SAND-WP-0012)
```bash
sandboxer create \
--profile profile.vm-packer-build \
--input packer_template=~/the-custodian/infra/build-machines/haskell \
--input vm_name=haskell-build \
--host localhost
```
| Input | Purpose |
|-------|---------|
| `mode` | `build` (default for profile.vm-packer-build) or `attach` |
| `packer_template` | Directory containing `*.pkr.hcl` |
| `vm_name` / `vm` | Packer `vm_name` variable |
| `packer_var_*` | Extra Packer `-var` flags (suffix → variable name) |
Runbook: `docs/runbooks/profile-vm-packer-build.md`
## Port registry (read-only pointer)
`the-custodian/infra/build-machines/port-registry.yml` maps tunnel ports
1222112230 to VM slots. When attaching via tunnel, set
`SANDBOXER_VM_TUNNEL_PORT` or `--input tunnel_port=` to a registered port.
Full ops-bridge automation is deferred — operators bring tunnels up manually.
## Not migrated yet
- Automated Packer `create` trigger from sand-boxer API
- State Hub capability-catalog sync from build-agent (agent unchanged)
- Port registry automation (`port-registry.yml`)
- `make remote-build` Makefile targets in the-custodian (add shim in follow-on if needed)
- Automated port-registry → ops-bridge config generation
## Runbook

View file

@ -44,6 +44,8 @@ Deferred: Packer orchestration from API, `make remote-build` shim.
| Item | Workplan |
|------|----------|
| ~~SaaS extensions + payments v0~~ | SAND-WP-0006 — stub + routing + credits |
| E2B / Modal real adapters | Post SAND-WP-0006 |
| ~~E2B / Modal real adapters + fin-hub~~ | SAND-WP-0010 — `docs/cloud-adapters.md` |
| Consumer profiles + reachability | **SAND-WP-0011** |
| Packer orchestration + remote-build shim | **SAND-WP-0012** |
| ~~Snapshot / restore~~ | SAND-WP-0007 — `docs/snapshots.md` |
| ~~TTL enforcement + scheduled reap~~ | SAND-WP-0009 — `docs/ttl.md` |

View file

@ -36,10 +36,17 @@ Reference: `ext.saas-stub` (no external API).
## BYOK
Provider API keys are resolved at provision boundary via `secret_refs` / OpenBao —
not implemented in v0 stub. Set provider env vars per extension when adapters land.
Provider API keys resolve at provision boundary — never stored on `SandboxStatus`
or emitted to State Hub.
1. Operator lookup: `warden route find "<provider> API key" --json`
2. Inject env before `sandboxer create` (e.g. `E2B_API_KEY`, `MODAL_TOKEN_ID`)
3. Or map `secret_ref` from extension config to `SANDBOXER_SECRET_<REF>` env
See `docs/cloud-adapters.md`.
## Billing export
sand-boxer meters sandbox consumption only. Domain billing authority (fin-hub) is a
future export consumer of State Hub meter events — not owned here.
On metered destroy, optional fin-hub hook when `SANDBOXER_FIN_HUB_URL` is set.
Posts `sandbox_id`, `extension_id`, `duration_s`, `actual_usd` to `/usage/sandbox`.
Implementation: `src/sandboxer/payments/billing_export.py`.

View file

@ -10,6 +10,8 @@ route:
strategy: prefer-self-hosted
extensions:
- ext.compose-ssh
- ext.e2b
- ext.modal
- ext.saas-stub
max_cost_per_hour_usd: 1.0
```
@ -19,7 +21,7 @@ route:
| Strategy | Behavior |
|----------|----------|
| `explicit` | Use `profile.extension` (default when no route) |
| `prefer-self-hosted` | First self-hosted candidate with resolvable host; else SaaS |
| `prefer-self-hosted` | Self-hosted if host available; else credentialed E2B/Modal; else stub |
| `lowest-cost` | Self-hosted if available; else cheapest `estimate_cost` |
| `lowest-latency` | Self-hosted if available; else last candidate (v0) |
@ -38,7 +40,12 @@ sandboxer create --profile profile.saas-stub
| Profile | Route |
|---------|-------|
| `profile.burst-sandbox` | compose-ssh → saas-stub fallback |
| `profile.burst-sandbox` | compose-ssh → e2b → modal → saas-stub |
| `profile.e2b-burst` | explicit `ext.e2b` |
| `profile.modal-gpu` | explicit `ext.modal` |
| `profile.saas-stub` | explicit `ext.saas-stub` |
Cloud adapters require provider credentials (`E2B_API_KEY`, `MODAL_TOKEN_ID`).
See `docs/cloud-adapters.md`.
Resolver: `sandboxer.routing.resolver.resolve_extension`.

View file

@ -0,0 +1,54 @@
# profile.vm-packer-build — Runbook
Trigger a Packer OVA build on the local workstation (build-machines lineage).
## Prerequisites
- **Packer** >= 1.10 (`packer version`)
- **VirtualBox** >= 7.0 (`VBoxManage --version`)
- Template directory from `the-custodian/infra/build-machines/haskell`
- `sandboxer` on PATH
## Build OVA
```bash
sandboxer create \
--profile profile.vm-packer-build \
--input packer_template=~/the-custodian/infra/build-machines/haskell \
--input vm_name=haskell-build \
--host localhost
```
Progress notes emit to State Hub during `packer init` and `packer build`.
On success, `reachability.remote_dir` points at the produced `.ova` file.
## Optional Packer variables
Pass extra `-var` flags via inputs prefixed with `packer_var_`:
```bash
sandboxer create \
--profile profile.vm-packer-build \
--input packer_template=~/the-custodian/infra/build-machines/haskell \
--input vm_name=haskell-build \
--input packer_var_memory=16384 \
--host localhost
```
## Destroy
```bash
sandboxer destroy <sandbox_id>
```
Removes the sandbox record only; the OVA artifact on disk is preserved.
## Attach workflow (post-build)
After import/setup per build-machines README, use `profile.vm-haskell-build`
for workspace attach — see `docs/runbooks/profile-vm-haskell-build.md`.
## Migration reference
`docs/migration-build-machines.md`

19
extensions/ext.e2b.yaml Normal file
View file

@ -0,0 +1,19 @@
id: ext.e2b
title: E2B cloud sandboxes
description: >
Metered E2B Firecracker sandbox adapter. Requires E2B_API_KEY or OpenBao
secret_ref mapping at provision boundary.
handler: sandboxer.extensions.e2b:E2BExtension
capabilities:
isolation_levels: [microvm]
regions: [us, eu]
persistence: true
pricing_model: metered
config:
provider: e2b
api_base: https://api.e2b.dev
api_key_env: E2B_API_KEY
secret_ref: e2b-api-key
template_id: base
rate_usd_per_hour: 0.15
session_fee_usd: 0.02

19
extensions/ext.modal.yaml Normal file
View file

@ -0,0 +1,19 @@
id: ext.modal
title: Modal cloud sandboxes
description: >
Metered Modal serverless sandbox adapter. Requires MODAL_TOKEN_ID or
secret_ref mapping at provision boundary.
handler: sandboxer.extensions.modal:ModalExtension
capabilities:
isolation_levels: [policy]
regions: [us]
persistence: true
pricing_model: metered
config:
provider: modal
api_base: https://api.modal.com
api_key_env: MODAL_TOKEN_ID
secret_ref: modal-token-id
image_ref: modal-default
rate_usd_per_hour: 0.18
session_fee_usd: 0.02

View file

@ -1,9 +1,9 @@
id: ext.vm-packer
title: VM workspace (Packer lineage)
description: >
Attach an isolated workspace on a pre-built VM (the-custodian build-machines
lineage). v0 supports attach mode via SSH alias or tunnel port; Packer build
orchestration is operator-driven and deferred.
Attach an isolated workspace on a pre-built VM, or run Packer build mode
(the-custodian build-machines lineage). Attach via SSH alias or tunnel port;
build via inputs.mode=build or profile.vm-packer-build.
handler: sandboxer.extensions.vm_packer:VMPackerExtension
capabilities:
isolation_levels: [microvm]

View file

@ -160,14 +160,14 @@ some sibling sign-offs sit outside this repo.
| Prio | Gap | Owner | Proposed track |
| --- | --- | --- | --- |
| **P1** | TTL enforcement + `extend_ttl` + `expires_at` | sand-boxer | **SAND-WP-0009** |
| **P2** | TTL reap scheduler / activity-core contract | sand-boxer + activity-core | **SAND-WP-0009** |
| **P3** | `.repo-classification.yaml` + registry refresh | sand-boxer | **SAND-WP-0009** |
| **P4** | HTTP API parity (`recreate`, `extend_ttl`) | sand-boxer | **SAND-WP-0009** |
| **P5** | Real E2B / Modal adapters + BYOK | sand-boxer | SAND-WP-0010 |
| **P6** | ops-bridge tunnel in reachability descriptor | sand-boxer | SAND-WP-0011 |
| **P7** | Consumer profiles (glas-harness, snuggle) | sand-boxer | SAND-WP-0011 |
| **P8** | Packer build orchestration + remote-build shim | sand-boxer | SAND-WP-0012 (WP-0005-T06) |
| ~~**P1**~~ | TTL enforcement + `extend_ttl` + `expires_at` | sand-boxer | **SAND-WP-0009** — done |
| ~~**P2**~~ | TTL reap / activity-core contract | sand-boxer | **SAND-WP-0009**`docs/integrations/activity-core.md` |
| ~~**P3**~~ | `.repo-classification.yaml` + registry refresh | sand-boxer | **SAND-WP-0009** — done |
| ~~**P4**~~ | HTTP API parity (`recreate`, `extend_ttl`) | sand-boxer | **SAND-WP-0009** — done |
| **P5** | Real E2B / Modal adapters + BYOK | sand-boxer | **SAND-WP-0010** |
| **P6** | ops-bridge tunnel in reachability descriptor | sand-boxer | **SAND-WP-0011** |
| **P7** | Consumer profiles (glas-harness, snuggle) | sand-boxer | **SAND-WP-0011** |
| **P8** | Packer build orchestration + remote-build shim | sand-boxer | **SAND-WP-0012** |
| **P9** | Cross-host snapshot transfer | sand-boxer | Future |
| **P10** | fin-hub billing export | sand-boxer + platform | With SAND-WP-0010 |
| **P11** | sandboxer01 dedicated host | infra / operator | Outside repo |
@ -175,16 +175,17 @@ some sibling sign-offs sit outside this repo.
---
## 9. Recommended next workplan
## 9. Recommended workplans (2026-06-24)
**SAND-WP-0009 — TTL enforcement and operational hardening**
| Workplan | Status | Closes |
| --- | --- | --- |
| SAND-WP-0009 | **finished** | P1P4 (TTL, hygiene, HTTP parity) |
| SAND-WP-0010 | ready | P5, P10 (E2B/Modal, BYOK, fin-hub) |
| SAND-WP-0011 | ready | P6, P7 (reachability, consumer profiles) |
| SAND-WP-0012 | ready | P8 (Packer orchestration, remote-build shim) |
Closes P1P4: the largest functional gap (disposable-by-default TTL), platform
hygiene (classification, registry), and HTTP parity. Unblocks activity-core
scheduling of expire/reap without implementing activity-core itself.
Subsequent: **SAND-WP-0010** (real cloud adapters), **SAND-WP-0011**
(reachability + consumer profiles), **SAND-WP-0012** (Packer orchestration).
**Suggested implementation order:** 0010 and 0011 may parallelize; 0012 depends
on stable vm-packer attach (done). Operator tracks P11/P12 outside repo.
---

View file

@ -0,0 +1,42 @@
id: profile.agent-dev
version: "1.0.0"
extension: ext.compose-ssh
route:
strategy: prefer-self-hosted
extensions:
- ext.compose-ssh
- ext.e2b
- ext.modal
- ext.saas-stub
max_cost_per_hour_usd: 1.0
isolation:
level: container
network:
default: deny
egress: []
workspace:
mode: remote-canonical
access: rw
scope_default: agent
ttl:
default: 8h
max: 24h
idle_reap: 2h
resources:
cpu: null
memory_mb: null
setup:
instructions: >
Agent development sandbox for glas-harness. Prefer self-hosted compose;
burst to cloud when host unavailable. Consumer actor: agt.
secret_refs: []
placement:
prefer: [sandboxer01]
fallback: [coulombcore]
reachability:
tunnel: ops-bridge
identity: ops-warden
metadata:
cost_class: self-hosted
latency_class: standard
observability: none

View file

@ -0,0 +1,35 @@
id: profile.build
version: "1.0.0"
extension: ext.vm-packer
isolation:
level: microvm
network:
default: deny
egress: []
workspace:
mode: remote-canonical
access: rw
scope_default: agent
ttl:
default: 8h
max: 24h
idle_reap: null
resources:
cpu: null
memory_mb: null
setup:
instructions: >
Build sandbox for snuggle-inventor. Attach to pre-built VM via ops-bridge
tunnel (e.g. haskell-build). Secret refs resolved at provision boundary only.
secret_refs:
- build-registry-token
placement:
prefer: [localhost]
fallback: [workstation]
reachability:
tunnel: ops-bridge
identity: ops-warden
metadata:
cost_class: self-hosted
latency_class: standard
observability: none

View file

@ -5,6 +5,8 @@ route:
strategy: prefer-self-hosted
extensions:
- ext.compose-ssh
- ext.e2b
- ext.modal
- ext.saas-stub
max_cost_per_hour_usd: 1.0
isolation:
@ -25,8 +27,8 @@ resources:
memory_mb: null
setup:
instructions: >
Prefer self-hosted compose on SANDBOXER_HOST; falls back to metered SaaS stub
when host is unavailable or SANDBOXER_FORCE_SAAS=1.
Prefer self-hosted compose on SANDBOXER_HOST; falls back to E2B/Modal when
credentials are configured, else metered saas-stub. Force SaaS: SANDBOXER_FORCE_SAAS=1.
secret_refs: []
placement:
prefer: [sandboxer01]

View file

@ -0,0 +1,32 @@
id: profile.e2b-burst
version: "1.0.0"
extension: ext.e2b
isolation:
level: microvm
network:
default: deny
egress: []
workspace:
mode: remote-canonical
access: rw
scope_default: session
ttl:
default: 2h
max: 8h
idle_reap: null
resources:
cpu: null
memory_mb: null
setup:
instructions: "E2B burst sandbox — requires E2B_API_KEY."
secret_refs: [e2b-api-key]
placement:
prefer: []
fallback: []
reachability:
tunnel: ops-bridge
identity: ops-warden
metadata:
cost_class: saas-metered
latency_class: low
observability: none

View file

@ -0,0 +1,32 @@
id: profile.modal-gpu
version: "1.0.0"
extension: ext.modal
isolation:
level: policy
network:
default: deny
egress: []
workspace:
mode: remote-canonical
access: rw
scope_default: session
ttl:
default: 2h
max: 8h
idle_reap: null
resources:
cpu: null
memory_mb: null
setup:
instructions: "Modal GPU/burst sandbox — requires MODAL_TOKEN_ID."
secret_refs: [modal-token-id]
placement:
prefer: []
fallback: []
reachability:
tunnel: ops-bridge
identity: ops-warden
metadata:
cost_class: saas-metered
latency_class: low
observability: none

View file

@ -0,0 +1,37 @@
id: profile.vm-packer-build
version: "1.0.0"
extension: ext.vm-packer
isolation:
level: microvm
network:
default: deny
egress: []
workspace:
mode: remote-canonical
access: rw
scope_default: session
ttl:
default: 12h
max: 48h
idle_reap: null
resources:
cpu: null
memory_mb: null
setup:
instructions: >
Trigger a Packer OVA build (the-custodian build-machines conventions).
Requires packer and VirtualBox on the placement host. Pass
--input packer_template=~/the-custodian/infra/build-machines/haskell
and --input vm_name=haskell-build (or vm=haskell-build). Attach mode
uses profile.vm-haskell-build instead.
secret_refs: []
placement:
prefer: [localhost]
fallback: [workstation]
reachability:
tunnel: ops-bridge
identity: ops-warden
metadata:
cost_class: self-hosted
latency_class: standard
observability: none

View file

@ -16,27 +16,27 @@ maturity:
Charter (INTENT.md), meta-framework spec, extension SDK, integration docs,
and research synthesis. Capability indexed in registry/.
availability:
current: A4
current: A5
target: A5
confidence: high
rationale: >
CLI v0 (create/destroy/snapshot/TTL), HTTP API, CoulombCore remote smoke.
SaaS stub + routing + credits shipped (SAND-WP-0006).
Cloud adapters E2B/Modal + routing + credits (SAND-WP-0010).
external_evidence:
completeness:
level: C4
name: Substantial
level: C5
name: Mature
confidence: high
basis: scope_vs_intent_and_consumer_expectations
satisfied_expectations:
- profile-based create/destroy/snapshot/restore via CLI
- TTL extend and expire/reap (SAND-WP-0009)
- E2B and Modal cloud adapters with BYOK (SAND-WP-0010)
- State Hub lifecycle events on transitions
- wise-validator and the-custodian migration arc complete
- extension SDK with compose-ssh, vm-packer attach, saas-stub
- extension SDK with compose-ssh, vm-packer attach, saas-stub, e2b, modal
broken_expectations:
- Real E2B/Modal adapters not yet built
- sandboxer01 dedicated host not live (CoulombCore interim)
out_of_scope_expectations:
- agent harness and tool orchestration (glas-harness)

20
scripts/smoke-agent-dev.sh Executable file
View file

@ -0,0 +1,20 @@
#!/usr/bin/env bash
# Smoke profile.agent-dev — requires SANDBOXER_HOST or defaults via placement.
set -euo pipefail
if [[ -z "${SANDBOXER_HOST:-}" ]]; then
echo "SANDBOXER_HOST not set — using profile placement fallback" >&2
fi
REPO="${SMOKE_REPO:-$(pwd)}"
echo "Smoke: profile.agent-dev repo=$REPO"
STATUS=$(sandboxer create \
--profile profile.agent-dev \
--input "repo=$REPO" \
--actor agt \
--project glas-harness)
ID=$(echo "$STATUS" | python3 -c "import sys,json; print(json.load(sys.stdin)['sandbox_id'])")
echo "Created: $ID"
sandboxer reachability show "$ID"
sandboxer destroy "$ID"
echo "OK: agent-dev smoke"

23
scripts/smoke-build-profile.sh Executable file
View file

@ -0,0 +1,23 @@
#!/usr/bin/env bash
# Smoke profile.build — requires VM tunnel and optional build-registry-token secret.
set -euo pipefail
if [[ -z "${SANDBOXER_SECRET_BUILD_REGISTRY_TOKEN:-}" ]]; then
echo "SANDBOXER_SECRET_BUILD_REGISTRY_TOKEN not set — skipping live smoke" >&2
exit 0
fi
VM="${SMOKE_VM:-haskell-build}"
TUNNEL_PORT="${SANDBOXER_VM_TUNNEL_PORT:-12222}"
echo "Smoke: profile.build vm=$VM tunnel=$TUNNEL_PORT"
STATUS=$(sandboxer create \
--profile profile.build \
--input "vm=$VM" \
--input "tunnel_port=$TUNNEL_PORT" \
--actor agt \
--project snuggle-inventor)
ID=$(echo "$STATUS" | python3 -c "import sys,json; print(json.load(sys.stdin)['sandbox_id'])")
echo "Created: $ID"
sandboxer reachability show "$ID"
sandboxer destroy "$ID"
echo "OK: build profile smoke"

32
scripts/smoke-cloud-adapter.sh Executable file
View file

@ -0,0 +1,32 @@
#!/usr/bin/env bash
# Operator smoke for cloud adapters — requires provider credentials.
set -euo pipefail
PROVIDER="${1:-e2b}"
case "$PROVIDER" in
e2b)
PROFILE="profile.e2b-burst"
if [[ -z "${E2B_API_KEY:-}" ]]; then
echo "E2B_API_KEY not set — skipping live smoke" >&2
exit 0
fi
;;
modal)
PROFILE="profile.modal-gpu"
if [[ -z "${MODAL_TOKEN_ID:-}" ]]; then
echo "MODAL_TOKEN_ID not set — skipping live smoke" >&2
exit 0
fi
;;
*)
echo "Usage: $0 [e2b|modal]" >&2
exit 1
;;
esac
echo "Smoke: sandboxer create --profile $PROFILE"
STATUS=$(sandboxer create --profile "$PROFILE" --project sand-boxer)
ID=$(echo "$STATUS" | python3 -c "import sys,json; print(json.load(sys.stdin)['sandbox_id'])")
echo "Created: $ID"
sandboxer destroy "$ID"
echo "Destroyed: $ID"

View file

@ -34,6 +34,14 @@ def get_sandbox(sandbox_id: str) -> SandboxStatus:
return status
@app.get("/v1/sandboxes/{sandbox_id}/reachability")
def get_sandbox_reachability(sandbox_id: str) -> dict:
try:
return _manager.reachability_report(sandbox_id)
except KeyError as exc:
raise HTTPException(status_code=404, detail=str(exc)) from exc
@app.get("/v1/sandboxes", response_model=list[SandboxStatus])
def list_sandboxes() -> list[SandboxStatus]:
return _manager.list()

View file

@ -30,6 +30,8 @@ credits_app = typer.Typer(help="SaaS sandbox credits (metered extensions).")
app.add_typer(credits_app, name="credits")
snapshots_app = typer.Typer(help="Workspace checkpoint snapshots.")
app.add_typer(snapshots_app, name="snapshots")
reachability_app = typer.Typer(help="Consumer reachability descriptors.")
app.add_typer(reachability_app, name="reachability")
@app.callback()
@ -92,7 +94,11 @@ def sandbox_create(
host: Annotated[str | None, typer.Option(help="Override placement host")] = None,
ttl: Annotated[str | None, typer.Option(help="TTL override (e.g. 4h)")] = None,
) -> None:
"""Provision a sandbox. No args → canary self-deploy of sand-boxer."""
"""Provision a sandbox. No args → canary self-deploy of sand-boxer.
vm-packer modes: attach (profile.vm-haskell-build) or build
(profile.vm-packer-build / --input mode=build with packer_template, vm_name).
"""
parsed = _parse_inputs(input or [])
resolved_profile, resolved_inputs = resolve_create_defaults(profile, parsed)
request = SandboxCreateRequest(
@ -112,6 +118,18 @@ def sandbox_create(
_print_telemetry_summary(status.telemetry)
@reachability_app.command("show")
def reachability_show(sandbox_id: str) -> None:
"""Show reachability descriptor and SSH one-liner for a sandbox."""
manager = SandboxManager()
try:
report = manager.reachability_report(sandbox_id)
except KeyError as exc:
typer.echo(str(exc), err=True)
raise typer.Exit(code=1) from exc
_print_json(report)
@app.command("get")
def sandbox_get(sandbox_id: str) -> None:
"""Get sandbox status by id."""

View file

@ -21,11 +21,14 @@ from sandboxer.models import (
SandboxStatus,
SnapshotRecord,
)
from sandboxer.payments.billing_export import export_meter_usage
from sandboxer.payments.credits import CreditsStore
from sandboxer.payments.metering import estimate_cost, settle_usage
from sandboxer.placement import resolve_host
from sandboxer.profiles.loader import load_profile
from sandboxer.reachability.enrich import enrich_reachability
from sandboxer.routing.resolver import resolve_extension
from sandboxer.secrets.resolver import resolve_setup_secrets
from sandboxer.snapshots.store import SnapshotStore
from sandboxer.telemetry.export import export_telemetry
from sandboxer.telemetry.introspection import (
@ -60,6 +63,8 @@ class SandboxManager:
"vm_target": status.inputs.get("vm_target", ""),
"vm_host": status.inputs.get("vm_host", ""),
"endpoint": status.inputs.get("endpoint", ""),
"provider_sandbox_id": status.inputs.get("provider_sandbox_id", ""),
"provider": status.inputs.get("provider", ""),
}
def _resolved_host(self, profile, extension, host_override: str | None) -> str:
@ -124,7 +129,21 @@ class SandboxManager:
provision_before = collect_host_snapshot(resolved_host)
try:
handle = backend.provision(profile, request.inputs, resolved_host)
secret_bundle = resolve_setup_secrets(profile)
provision_inputs = dict(request.inputs)
build_mode = (
provision_inputs.get("mode") == "build"
or profile.id == "profile.vm-packer-build"
)
if build_mode:
emit_lifecycle_event(
status,
summary=f"Packer build starting ({profile.id})",
event_type="note",
)
handle = backend.provision(profile, provision_inputs, resolved_host)
if secret_bundle:
handle["_secret_refs"] = secret_bundle
status.sandbox_id = handle["sandbox_id"]
status.inputs["compose_file"] = handle.get("compose_file", "")
status.inputs["ssh_user"] = handle.get("ssh_user", "")
@ -133,7 +152,10 @@ class SandboxManager:
status.inputs["vm_target"] = handle.get("vm_target", "")
status.inputs["vm_host"] = handle.get("vm_host", "")
status.inputs["endpoint"] = handle.get("endpoint", "")
status.inputs["provider_sandbox_id"] = handle.get("provider_sandbox_id", "")
status.inputs["provider"] = handle.get("provider", "")
reach = backend.wait_ready(handle)
reach = enrich_reachability(reach, profile, handle)
status.reachability = Reachability(**reach)
status.state = SandboxState.READY
status.ready_at = utcnow()
@ -173,6 +195,14 @@ class SandboxManager:
def get(self, sandbox_id: str) -> SandboxStatus | None:
return self.store.get(sandbox_id)
def reachability_report(self, sandbox_id: str) -> dict:
status = self.store.get(sandbox_id)
if not status:
raise KeyError(f"Sandbox not found: {sandbox_id}")
from sandboxer.reachability.enrich import build_reachability_report
return build_reachability_report(status)
def list(self) -> list[SandboxStatus]:
return sorted(self.store.list_all(), key=lambda s: s.created_at, reverse=True)
@ -209,6 +239,7 @@ class SandboxManager:
if settled and settled.pricing_model == "metered" and settled.actual_usd:
self.credits.debit(settled.actual_usd)
status.meter = settled
export_meter_usage(status, extension_id=extension.id, meter=settled)
emit_lifecycle_event(
status,
summary=(
@ -447,7 +478,11 @@ class SandboxManager:
status.inputs["vm_host"] = handle.get("vm_host", "")
status.inputs["endpoint"] = handle.get("endpoint", "")
status.inputs["restored_from"] = record.snapshot_id
secret_bundle = resolve_setup_secrets(profile)
if secret_bundle:
handle["_secret_refs"] = secret_bundle
reach = backend.wait_ready(handle)
reach = enrich_reachability(reach, profile, handle)
status.reachability = Reachability(**reach)
status.state = SandboxState.READY
status.ready_at = utcnow()

View file

@ -0,0 +1,83 @@
"""Shared helpers for metered HTTP cloud sandbox adapters."""
from __future__ import annotations
from typing import Any, Protocol
import httpx
from sandboxer.extensions.base import SandboxExtension
from sandboxer.extensions.credentials import resolve_api_key
from sandboxer.models import MeterQuote, Profile
class HttpClientFactory(Protocol):
def __call__(self) -> httpx.Client: ...
def default_http_client() -> httpx.Client:
return httpx.Client(timeout=60.0)
class CloudMeteredExtension(SandboxExtension):
"""Base for E2B/Modal-style REST sandbox providers."""
extension_id: str = ""
def __init__(
self,
config: dict[str, Any] | None = None,
*,
client_factory: HttpClientFactory | None = None,
) -> None:
super().__init__(config)
self.api_base: str = str(self.config.get("api_base", "")).rstrip("/")
self.api_key_env: str = str(self.config.get("api_key_env", ""))
self.provider: str = self.config.get("provider", self.extension_id)
self.rate_usd_per_hour: float = float(self.config.get("rate_usd_per_hour", 0.15))
self.session_fee_usd: float = float(self.config.get("session_fee_usd", 0.02))
self._client_factory = client_factory or default_http_client
@classmethod
def credentials_available(cls, config: dict[str, Any]) -> bool:
from sandboxer.extensions.credentials import credentials_available as _avail
return _avail(cls.extension_id, config)
def _api_key(self) -> str:
key = resolve_api_key(self.config, extension_id=self.extension_id)
if not key:
raise RuntimeError(
f"{self.extension_id}: API key not configured "
f"(set {self.api_key_env} or secret_ref env mapping)"
)
return key
def _headers(self) -> dict[str, str]:
return {
"Authorization": f"Bearer {self._api_key()}",
"Content-Type": "application/json",
}
def _client(self) -> httpx.Client:
return self._client_factory()
def estimate_cost(
self,
profile: Profile,
inputs: dict[str, str],
*,
duration_s: int = 3600,
) -> MeterQuote:
hours = max(duration_s / 3600.0, 1 / 3600)
estimated = round(self.session_fee_usd + hours * self.rate_usd_per_hour, 4)
return MeterQuote(
extension_id=self.extension_id,
estimated_usd=estimated,
unit="per_hour",
duration_s=duration_s,
)
def meter_actual(self, handle: dict[str, str], *, duration_s: float) -> float:
hours = max(duration_s / 3600.0, 1 / 3600)
return round(self.session_fee_usd + hours * self.rate_usd_per_hour, 4)

View file

@ -0,0 +1,48 @@
"""BYOK credential resolution for metered cloud extensions."""
from __future__ import annotations
import os
from typing import Any
def _secret_ref_env(secret_ref: str) -> str:
normalized = secret_ref.upper().replace("-", "_").replace(".", "_")
return f"SANDBOXER_SECRET_{normalized}"
def resolve_api_key(
config: dict[str, Any],
*,
extension_id: str,
) -> str | None:
"""Resolve provider API key from env or secret_ref mapping (never from Git)."""
env_name = config.get("api_key_env")
if env_name:
value = os.environ.get(env_name)
if value:
return value
fallback_env = (
f"SANDBOXER_{extension_id.upper().replace('.', '_').replace('-', '_')}_API_KEY"
)
value = os.environ.get(fallback_env)
if value:
return value
secret_ref = config.get("secret_ref")
if secret_ref:
return os.environ.get(_secret_ref_env(secret_ref))
return None
def credentials_available(
extension_id: str,
config: dict[str, Any],
*,
always_available: bool = False,
) -> bool:
if always_available:
return True
return resolve_api_key(config, extension_id=extension_id) is not None

View file

@ -0,0 +1,77 @@
"""ext.e2b — E2B cloud sandbox adapter."""
from __future__ import annotations
from typing import Any
from sandboxer.extensions.cloud_base import CloudMeteredExtension, default_http_client
from sandboxer.models import Profile
# Re-export for tests
http_client_factory = default_http_client
class E2BExtension(CloudMeteredExtension):
extension_id = "ext.e2b"
def __init__(self, config: dict[str, Any] | None = None, **kwargs) -> None:
super().__init__(config, **kwargs)
self.template_id: str = self.config.get("template_id", "base")
def provision(
self, profile: Profile, inputs: dict[str, str], host: str
) -> dict[str, str]:
sandbox_id = self.new_sandbox_id(inputs)
template = inputs.get("template") or self.template_id
payload = {"templateID": template, "metadata": {"sandboxer_id": sandbox_id}}
with self._client() as client:
response = client.post(
f"{self.api_base}/sandboxes",
json=payload,
headers=self._headers(),
)
if response.status_code >= 400:
raise RuntimeError(f"E2B provision failed: {response.text}")
data = response.json()
provider_sandbox_id = data.get("sandboxID") or data.get("sandbox_id", "")
endpoint = data.get("sandboxURL") or f"https://{provider_sandbox_id}.e2b.dev"
return {
"sandbox_id": sandbox_id,
"provider_sandbox_id": provider_sandbox_id,
"host": self.provider,
"endpoint": endpoint,
"provider": self.provider,
"template_id": template,
}
def wait_ready(self, handle: dict[str, str]) -> dict[str, str]:
provider_id = handle.get("provider_sandbox_id", "")
with self._client() as client:
response = client.get(
f"{self.api_base}/sandboxes/{provider_id}",
headers=self._headers(),
)
if response.status_code >= 400:
raise RuntimeError(f"E2B wait_ready failed: {response.text}")
return {
"endpoint": handle["endpoint"],
"host": handle.get("host"),
}
def teardown(self, handle: dict[str, str]) -> dict[str, str]:
provider_id = handle.get("provider_sandbox_id", "")
removed = False
if provider_id:
with self._client() as client:
response = client.delete(
f"{self.api_base}/sandboxes/{provider_id}",
headers=self._headers(),
)
removed = response.status_code < 400
return {
"provider_removed": str(removed).lower(),
"sandbox_id": handle.get("sandbox_id", ""),
"provider_sandbox_id": provider_id,
}

View file

@ -0,0 +1,83 @@
"""ext.modal — Modal cloud sandbox adapter."""
from __future__ import annotations
from typing import Any
from sandboxer.extensions.cloud_base import CloudMeteredExtension, default_http_client
from sandboxer.models import Profile
http_client_factory = default_http_client
class ModalExtension(CloudMeteredExtension):
extension_id = "ext.modal"
def __init__(self, config: dict[str, Any] | None = None, **kwargs) -> None:
super().__init__(config, **kwargs)
self.image_ref: str = self.config.get("image_ref", "modal-default")
def provision(
self, profile: Profile, inputs: dict[str, str], host: str
) -> dict[str, str]:
sandbox_id = self.new_sandbox_id(inputs)
image = inputs.get("image") or self.image_ref
payload = {
"image": image,
"metadata": {"sandboxer_id": sandbox_id},
}
with self._client() as client:
response = client.post(
f"{self.api_base}/v1/sandboxes",
json=payload,
headers=self._headers(),
)
if response.status_code >= 400:
raise RuntimeError(f"Modal provision failed: {response.text}")
data = response.json()
provider_sandbox_id = data.get("sandbox_id") or data.get("id", "")
endpoint = data.get("url") or f"https://modal.run/sandbox/{provider_sandbox_id}"
return {
"sandbox_id": sandbox_id,
"provider_sandbox_id": provider_sandbox_id,
"host": self.provider,
"endpoint": endpoint,
"provider": self.provider,
"image_ref": image,
}
def wait_ready(self, handle: dict[str, str]) -> dict[str, str]:
provider_id = handle.get("provider_sandbox_id", "")
with self._client() as client:
response = client.get(
f"{self.api_base}/v1/sandboxes/{provider_id}",
headers=self._headers(),
)
if response.status_code >= 400:
raise RuntimeError(f"Modal wait_ready failed: {response.text}")
data = response.json()
state = data.get("status", "ready")
if state not in ("ready", "running"):
raise RuntimeError(f"Modal sandbox not ready: {state}")
return {
"endpoint": handle["endpoint"],
"host": handle.get("host"),
}
def teardown(self, handle: dict[str, str]) -> dict[str, str]:
provider_id = handle.get("provider_sandbox_id", "")
removed = False
if provider_id:
with self._client() as client:
response = client.delete(
f"{self.api_base}/v1/sandboxes/{provider_id}",
headers=self._headers(),
)
removed = response.status_code < 400
return {
"provider_removed": str(removed).lower(),
"sandbox_id": handle.get("sandbox_id", ""),
"provider_sandbox_id": provider_id,
}

View file

@ -1,32 +1,46 @@
"""ext.vm-packer — attach to pre-built VMs (build-machines lineage).
"""ext.vm-packer — attach to pre-built VMs or trigger Packer builds.
v0 supports **attach** mode only: connect via SSH to an existing VM (tunnel alias,
localhost:port, or direct host). Creates an isolated workspace directory; teardown
removes the workspace, not the VM.
**attach** mode: connect via SSH to an existing VM (tunnel alias, localhost:port,
or direct host). Creates an isolated workspace directory; teardown removes the
workspace, not the VM.
Full Packer build / OVA import orchestration is deferred operators still build
images via the-custodian/infra/build-machines/ workflows.
**build** mode: run `packer init` + `packer build` for a template directory
(the-custodian build-machines conventions). Teardown preserves the OVA artifact.
"""
from __future__ import annotations
import os
import subprocess
from pathlib import Path
from typing import Any
from sandboxer.extensions.base import SandboxExtension
from sandboxer.extensions.ssh import SSHConfig
from sandboxer.lifecycle.state_hub import emit_progress_note
from sandboxer.models import Profile
BUILD_PROFILE_IDS = frozenset({"profile.vm-packer-build"})
class VMPackerExtension(SandboxExtension):
"""Attach sandbox workspace on a pre-provisioned VM."""
"""Attach sandbox workspace on a pre-provisioned VM, or run Packer build."""
def __init__(self, config: dict[str, Any] | None = None) -> None:
super().__init__(config)
self.workspace_base: str = self.config.get("workspace_base", "/build")
self.default_user: str | None = self.config.get("ssh_user")
self.ready_timeout_s: int = int(self.config.get("ready_timeout_s", 30))
self.default_packer_template: str | None = self.config.get("default_packer_template")
self.packer_bin: str = self.config.get("packer_bin", "packer")
def _resolve_mode(self, profile: Profile, inputs: dict[str, str]) -> str:
mode = inputs.get("mode")
if mode:
return mode
if profile.id in BUILD_PROFILE_IDS:
return "build"
return "attach"
def _ssh_from_handle(self, handle: dict[str, str]) -> SSHConfig:
port_raw = handle.get("ssh_port") or os.environ.get("SANDBOXER_VM_SSH_PORT")
@ -53,7 +67,106 @@ class VMPackerExtension(SandboxExtension):
return "localhost", int(env_port)
return inputs.get("vm_host") or placement_host, None
def provision(
def _run_packer(
self,
args: list[str],
*,
cwd: Path,
sandbox_id: str,
profile_id: str,
) -> None:
cmd = [self.packer_bin, *args]
emit_progress_note(
f"Packer: {' '.join(cmd)}",
sandbox_id=sandbox_id,
profile_id=profile_id,
detail={"cwd": str(cwd)},
)
result = subprocess.run(
cmd,
cwd=cwd,
capture_output=True,
text=True,
check=False,
)
if result.returncode != 0:
msg = result.stderr.strip() or result.stdout.strip() or "unknown error"
raise RuntimeError(f"Packer command failed ({' '.join(cmd)}): {msg}")
def _find_ova_artifact(self, template_dir: Path, vm_name: str) -> Path | None:
candidates = sorted(
template_dir.glob(f"{vm_name}*.ova"),
key=lambda p: p.stat().st_mtime,
reverse=True,
)
if candidates:
return candidates[0]
all_ova = sorted(
template_dir.glob("*.ova"),
key=lambda p: p.stat().st_mtime,
reverse=True,
)
return all_ova[0] if all_ova else None
def _provision_build(
self, profile: Profile, inputs: dict[str, str], host: str
) -> dict[str, str]:
sandbox_id = self.new_sandbox_id(inputs)
template_raw = (
inputs.get("packer_template")
or os.environ.get("SANDBOXER_PACKER_TEMPLATE")
or self.default_packer_template
)
if not template_raw:
raise ValueError(
"inputs.packer_template is required for build mode "
"(or set SANDBOXER_PACKER_TEMPLATE)"
)
vm_name = inputs.get("vm_name") or inputs.get("vm")
if not vm_name:
raise ValueError("inputs.vm_name or inputs.vm is required for build mode")
template_dir = Path(template_raw).expanduser().resolve()
if not template_dir.is_dir():
raise FileNotFoundError(f"Packer template directory not found: {template_dir}")
pkr_files = list(template_dir.glob("*.pkr.hcl")) + list(template_dir.glob("*.pkr.json"))
if not pkr_files:
raise FileNotFoundError(f"No Packer template (*.pkr.hcl) in {template_dir}")
self._run_packer(
["init", "."],
cwd=template_dir,
sandbox_id=sandbox_id,
profile_id=profile.id,
)
build_args = ["build", "-var", f"vm_name={vm_name}"]
for key, value in inputs.items():
if key.startswith("packer_var_"):
build_args.extend(["-var", f"{key.removeprefix('packer_var_')}={value}"])
self._run_packer(
build_args,
cwd=template_dir,
sandbox_id=sandbox_id,
profile_id=profile.id,
)
artifact = self._find_ova_artifact(template_dir, vm_name)
artifact_path = str(artifact) if artifact else ""
return {
"sandbox_id": sandbox_id,
"host": host,
"mode": "build",
"vm_name": vm_name,
"vm_target": vm_name,
"packer_template": str(template_dir),
"artifact_path": artifact_path,
"remote_dir": artifact_path,
}
def _provision_attach(
self, profile: Profile, inputs: dict[str, str], host: str
) -> dict[str, str]:
vm_target = inputs.get("vm") or inputs.get("ssh_target")
@ -94,11 +207,29 @@ class VMPackerExtension(SandboxExtension):
"remote_dir": remote_dir,
"ssh_user": ssh.user or "",
"ssh_port": str(ssh.port) if ssh.port else "",
"mode": inputs.get("mode", "attach"),
"mode": "attach",
"repo": repo_path_str,
}
def provision(
self, profile: Profile, inputs: dict[str, str], host: str
) -> dict[str, str]:
mode = self._resolve_mode(profile, inputs)
if mode == "build":
return self._provision_build(profile, inputs, host)
return self._provision_attach(profile, inputs, host)
def wait_ready(self, handle: dict[str, str]) -> dict[str, str]:
if handle.get("mode") == "build":
artifact = handle.get("artifact_path") or handle.get("remote_dir")
if not artifact or not Path(artifact).is_file():
raise RuntimeError(f"Packer artifact not found: {artifact}")
return {
"host": handle.get("host"),
"remote_dir": artifact,
"endpoint": artifact,
}
ssh = self._ssh_from_handle(handle)
remote_dir = handle["remote_dir"]
cmd = f"test -d {remote_dir} && echo ready"
@ -112,6 +243,15 @@ class VMPackerExtension(SandboxExtension):
}
def teardown(self, handle: dict[str, str]) -> dict[str, str]:
if handle.get("mode") == "build":
artifact = handle.get("artifact_path") or handle.get("remote_dir") or ""
return {
"workspace_removed": "false",
"remote_dir": artifact,
"vm_preserved": "true",
"artifact_preserved": "true",
}
remote_dir = handle.get("remote_dir")
cleaned_dir = False
if remote_dir:

View file

@ -59,6 +59,37 @@ def emit_lifecycle_event(
return None
def emit_progress_note(
summary: str,
*,
sandbox_id: str,
profile_id: str,
detail: dict[str, Any] | None = None,
author: str = "sandboxer",
) -> dict[str, Any] | None:
"""Emit a progress note during long-running provision (e.g. Packer build)."""
if os.environ.get("SANDBOXER_NO_STATE_HUB", "").lower() in ("1", "true", "yes"):
return None
payload = {
"event_type": "note",
"summary": summary,
"author": author,
"detail": {
"sandbox_id": sandbox_id,
"profile_id": profile_id,
**(detail or {}),
},
}
try:
response = httpx.post(f"{hub_url()}/progress/", json=payload, timeout=10.0)
response.raise_for_status()
return response.json()
except httpx.HTTPError:
return None
def event_type_for_state(state: SandboxState) -> str:
if state in (SandboxState.READY, SandboxState.DESTROYED, SandboxState.EXPIRED):
return "milestone"

View file

@ -153,6 +153,9 @@ class Reachability(BaseModel):
compose_project: str | None = None
host: str | None = None
endpoint: str | None = None
tunnel: str | None = None
tunnel_via: str | None = None
identity: str | None = None
class SandboxStatus(BaseModel):

View file

@ -0,0 +1,49 @@
"""Optional fin-hub billing export for metered sandbox usage."""
from __future__ import annotations
import os
from typing import Any
import httpx
from sandboxer.models import MeterRecord, SandboxStatus
def fin_hub_url() -> str | None:
return os.environ.get("SANDBOXER_FIN_HUB_URL") or None
def export_meter_usage(
status: SandboxStatus,
*,
extension_id: str,
meter: MeterRecord,
) -> dict[str, Any] | None:
"""POST usage record to fin-hub when SANDBOXER_FIN_HUB_URL is set."""
if os.environ.get("SANDBOXER_NO_FIN_HUB", "").lower() in ("1", "true", "yes"):
return None
if meter.pricing_model != "metered" or not meter.actual_usd:
return None
base = fin_hub_url()
if not base:
return None
payload = {
"sandbox_id": status.sandbox_id,
"extension_id": extension_id,
"profile_id": status.profile_id,
"consumer": status.consumer.model_dump(),
"duration_s": meter.duration_s,
"actual_usd": meter.actual_usd,
"estimate_usd": meter.estimate_usd,
"currency": meter.currency,
}
try:
response = httpx.post(f"{base.rstrip('/')}/usage/sandbox", json=payload, timeout=10.0)
response.raise_for_status()
return response.json()
except httpx.HTTPError:
return None

View file

@ -0,0 +1,5 @@
"""Reachability descriptor enrichment."""
from sandboxer.reachability.enrich import build_reachability_report, enrich_reachability
__all__ = ["enrich_reachability", "build_reachability_report"]

View file

@ -0,0 +1,68 @@
"""Merge profile reachability spec and env into consumer descriptors."""
from __future__ import annotations
import os
from typing import Any
from sandboxer.models import Profile, Reachability, SandboxStatus
OPS_BRIDGE_DOC = "ops-bridge MCP or `bridge` CLI — sand-boxer does not manage tunnels"
def enrich_reachability(
reach: dict[str, str],
profile: Profile,
handle: dict[str, str],
) -> dict[str, str]:
"""Add tunnel/identity metadata from profile spec and environment."""
enriched = dict(reach)
spec = profile.reachability
if spec.tunnel:
enriched.setdefault("tunnel_via", spec.tunnel)
if spec.identity:
enriched["identity"] = spec.identity
tunnel_port = (
os.environ.get("SANDBOXER_TUNNEL_PORT")
or handle.get("tunnel_port")
or handle.get("ssh_port")
)
tunnel_alias = os.environ.get("SANDBOXER_TUNNEL_ALIAS") or handle.get("vm_target")
if tunnel_port:
enriched["tunnel"] = f"localhost:{tunnel_port}"
elif tunnel_alias:
enriched["tunnel"] = tunnel_alias
tunnel_via = os.environ.get("SANDBOXER_TUNNEL_VIA")
if tunnel_via:
enriched["tunnel_via"] = tunnel_via
return enriched
def ssh_one_liner(reach: Reachability) -> str | None:
if reach.ssh and reach.remote_dir:
return f"ssh {reach.ssh} 'cd {reach.remote_dir} && exec $SHELL'"
if reach.ssh:
return f"ssh {reach.ssh}"
return None
def build_reachability_report(status: SandboxStatus) -> dict[str, Any]:
"""Consumer-facing reachability report with ops-bridge pointer."""
reach = status.reachability
payload: dict[str, Any] = {
"sandbox_id": status.sandbox_id,
"profile_id": status.profile_id,
"host": status.host,
"reachability": reach.model_dump(mode="json") if reach else None,
"ops_bridge": {
"doc": OPS_BRIDGE_DOC,
"note": "Bring tunnels up via ops-bridge; sand-boxer emits descriptor only",
},
}
if reach:
payload["ssh_one_liner"] = ssh_one_liner(reach)
return payload

View file

@ -4,6 +4,7 @@ from __future__ import annotations
import os
from sandboxer.extensions.credentials import credentials_available
from sandboxer.extensions.registry import load_extension
from sandboxer.models import Extension, Profile, RouteStrategy
from sandboxer.payments.metering import estimate_cost
@ -20,6 +21,49 @@ def _is_metered(ext: Extension) -> bool:
return ext.capabilities.pricing_model == "metered"
def _metered_available(ext: Extension) -> bool:
if ext.id == "ext.saas-stub":
return True
return credentials_available(ext.id, ext.config)
def _select_metered_fallback(
loaded: list[Extension],
profile: Profile,
inputs: dict[str, str],
*,
duration_s: int,
) -> Extension | None:
"""Pick cheapest credentialed metered extension; stub is always last resort."""
available = [ext for ext in loaded if _is_metered(ext) and _metered_available(ext)]
if not available:
return None
strategy = profile.route.strategy if profile.route else RouteStrategy.EXPLICIT
if strategy == RouteStrategy.LOWEST_COST:
best: Extension | None = None
best_cost: float | None = None
for ext in available:
if not _metered_available(ext):
continue
cost = _quote_cost(ext, profile, inputs, duration_s)
if cost is None:
continue
max_hour = profile.route.max_cost_per_hour_usd if profile.route else None
if max_hour is not None and cost > max_hour:
continue
if best is None or cost < (best_cost or float("inf")):
best, best_cost = ext, cost
if best:
return best
for ext_id in ("ext.e2b", "ext.modal", "ext.saas-stub"):
for ext in available:
if ext.id == ext_id:
return ext
return available[0]
def _self_hosted_available(profile: Profile, ext: Extension, host_override: str | None) -> bool:
if _is_metered(ext):
return True
@ -65,6 +109,11 @@ def resolve_extension(
for ext in loaded:
if not _is_metered(ext) and _self_hosted_available(profile, ext, host_override):
return ext
fallback = _select_metered_fallback(
loaded, profile, inputs, duration_s=duration_s
)
if fallback:
return fallback
for ext in loaded:
if _is_metered(ext):
return ext
@ -76,6 +125,8 @@ def resolve_extension(
for ext in loaded:
if not _is_metered(ext) and _self_hosted_available(profile, ext, host_override):
return ext
if not _metered_available(ext):
continue
cost = _quote_cost(ext, profile, inputs, duration_s)
if cost is None:
continue
@ -92,6 +143,11 @@ def resolve_extension(
for ext in loaded:
if not _is_metered(ext) and _self_hosted_available(profile, ext, host_override):
return ext
fallback = _select_metered_fallback(
loaded, profile, inputs, duration_s=duration_s
)
if fallback:
return fallback
return loaded[-1]
return load_extension(profile.extension)

View file

@ -0,0 +1,5 @@
"""Setup secret resolution at provision boundary."""
from sandboxer.secrets.resolver import resolve_setup_secrets
__all__ = ["resolve_setup_secrets"]

View file

@ -0,0 +1,41 @@
"""Resolve profile.setup.secret_refs from operator-injected env (BYOK boundary)."""
from __future__ import annotations
import os
from sandboxer.models import Profile
def _secret_env_name(ref: str) -> str:
normalized = ref.upper().replace("-", "_").replace(".", "_")
return f"SANDBOXER_SECRET_{normalized}"
def resolve_secret_ref(ref: str) -> str | None:
"""Resolve a single secret ref from env. OpenBao injection is operator-owned."""
return os.environ.get(_secret_env_name(ref))
def resolve_setup_secrets(profile: Profile) -> dict[str, str]:
"""Resolve all profile secret_refs or raise if any are missing."""
refs = profile.setup.secret_refs
if not refs:
return {}
resolved: dict[str, str] = {}
missing: list[str] = []
for ref in refs:
value = resolve_secret_ref(ref)
if value:
resolved[ref] = value
else:
missing.append(ref)
if missing:
env_hints = ", ".join(_secret_env_name(r) for r in missing)
raise ValueError(
f"Unresolved secret_refs for {profile.id}: {missing}. "
f"Set env ({env_hints}) or use warden route find for OpenBao path."
)
return resolved

View file

@ -0,0 +1,50 @@
"""fin-hub billing export tests."""
from __future__ import annotations
from datetime import UTC, datetime
from unittest.mock import MagicMock, patch
from sandboxer.models import ActorType, Consumer, MeterRecord, SandboxState, SandboxStatus
from sandboxer.payments.billing_export import export_meter_usage
def test_export_skipped_when_url_unset() -> None:
now = datetime.now(UTC)
status = SandboxStatus(
sandbox_id="s1",
profile_id="profile.e2b-burst",
extension_id="ext.e2b",
state=SandboxState.DESTROYED,
consumer=Consumer(actor=ActorType.ADM, project="sand-boxer"),
created_at=now,
updated_at=now,
)
meter = MeterRecord(pricing_model="metered", actual_usd=0.5, duration_s=100.0)
assert export_meter_usage(status, extension_id="ext.e2b", meter=meter) is None
def test_export_posts_when_configured(monkeypatch) -> None:
monkeypatch.setenv("SANDBOXER_FIN_HUB_URL", "http://fin-hub.test")
now = datetime.now(UTC)
status = SandboxStatus(
sandbox_id="s1",
profile_id="profile.e2b-burst",
extension_id="ext.e2b",
state=SandboxState.DESTROYED,
consumer=Consumer(actor=ActorType.ADM, project="sand-boxer"),
created_at=now,
updated_at=now,
)
meter = MeterRecord(pricing_model="metered", actual_usd=0.5, duration_s=100.0)
mock_response = MagicMock()
mock_response.json.return_value = {"ok": True}
with patch("sandboxer.payments.billing_export.httpx.post", return_value=mock_response) as post:
result = export_meter_usage(status, extension_id="ext.e2b", meter=meter)
assert result == {"ok": True}
post.assert_called_once()
payload = post.call_args.kwargs["json"]
assert payload["sandbox_id"] == "s1"
assert payload["actual_usd"] == 0.5

View file

@ -0,0 +1,20 @@
"""Consumer profile loader smoke tests."""
from sandboxer.profiles.loader import load_profile
def test_profile_agent_dev_loads() -> None:
profile = load_profile("profile.agent-dev")
assert profile.id == "profile.agent-dev"
assert profile.extension == "ext.compose-ssh"
assert profile.scope_default == "agent"
assert profile.route is not None
assert profile.ttl.default == "8h"
def test_profile_build_loads() -> None:
profile = load_profile("profile.build")
assert profile.id == "profile.build"
assert profile.extension == "ext.vm-packer"
assert "build-registry-token" in profile.setup.secret_refs
assert profile.reachability.tunnel == "ops-bridge"

72
tests/test_e2b.py Normal file
View file

@ -0,0 +1,72 @@
"""E2B cloud adapter tests."""
from __future__ import annotations
from unittest.mock import MagicMock, patch
import pytest
from sandboxer.extensions.credentials import credentials_available, resolve_api_key
from sandboxer.extensions.e2b import E2BExtension
from sandboxer.models import Profile
def _profile() -> Profile:
return Profile.model_validate(
{
"id": "profile.e2b-burst",
"version": "1.0.0",
"extension": "ext.e2b",
}
)
def test_credentials_from_env(monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.setenv("E2B_API_KEY", "test-key")
config = {"api_key_env": "E2B_API_KEY", "secret_ref": "e2b-api-key"}
assert resolve_api_key(config, extension_id="ext.e2b") == "test-key"
assert credentials_available("ext.e2b", config)
def test_provision_and_teardown_with_mock_client(monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.setenv("E2B_API_KEY", "test-key")
ext = E2BExtension(
{
"api_base": "https://api.e2b.dev",
"api_key_env": "E2B_API_KEY",
"provider": "e2b",
"template_id": "base",
}
)
mock_client = MagicMock()
create_resp = MagicMock()
create_resp.status_code = 200
create_resp.json.return_value = {
"sandboxID": "e2b-prov-1",
"sandboxURL": "https://e2b-prov-1.e2b.dev",
}
ready_resp = MagicMock()
ready_resp.status_code = 200
delete_resp = MagicMock()
delete_resp.status_code = 204
mock_client.post.return_value = create_resp
mock_client.get.return_value = ready_resp
mock_client.delete.return_value = delete_resp
mock_client.__enter__.return_value = mock_client
mock_client.__exit__.return_value = None
with patch.object(ext, "_client_factory", return_value=mock_client):
handle = ext.provision(_profile(), {}, "e2b")
assert handle["provider_sandbox_id"] == "e2b-prov-1"
reach = ext.wait_ready(handle)
assert "e2b-prov-1" in reach["endpoint"]
report = ext.teardown(handle)
assert report["provider_removed"] == "true"
def test_estimate_cost() -> None:
ext = E2BExtension({"rate_usd_per_hour": 0.15, "session_fee_usd": 0.02})
quote = ext.estimate_cost(_profile(), {}, duration_s=3600)
assert quote.extension_id == "ext.e2b"
assert quote.estimated_usd > 0

63
tests/test_modal.py Normal file
View file

@ -0,0 +1,63 @@
"""Modal cloud adapter tests."""
from __future__ import annotations
from unittest.mock import MagicMock, patch
import pytest
from sandboxer.extensions.modal import ModalExtension
from sandboxer.models import Profile
def _profile() -> Profile:
return Profile.model_validate(
{
"id": "profile.modal-gpu",
"version": "1.0.0",
"extension": "ext.modal",
}
)
def test_provision_and_teardown_with_mock_client(monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.setenv("MODAL_TOKEN_ID", "modal-token")
ext = ModalExtension(
{
"api_base": "https://api.modal.com",
"api_key_env": "MODAL_TOKEN_ID",
"provider": "modal",
}
)
mock_client = MagicMock()
create_resp = MagicMock()
create_resp.status_code = 200
create_resp.json.return_value = {
"sandbox_id": "modal-prov-1",
"url": "https://modal.run/sandbox/modal-prov-1",
"status": "ready",
}
ready_resp = MagicMock()
ready_resp.status_code = 200
ready_resp.json.return_value = {"status": "ready"}
delete_resp = MagicMock()
delete_resp.status_code = 200
mock_client.post.return_value = create_resp
mock_client.get.return_value = ready_resp
mock_client.delete.return_value = delete_resp
mock_client.__enter__.return_value = mock_client
mock_client.__exit__.return_value = None
with patch.object(ext, "_client_factory", return_value=mock_client):
handle = ext.provision(_profile(), {}, "modal")
assert handle["provider_sandbox_id"] == "modal-prov-1"
ext.wait_ready(handle)
report = ext.teardown(handle)
assert report["provider_removed"] == "true"
def test_provision_without_credentials_raises() -> None:
ext = ModalExtension({"api_key_env": "MODAL_TOKEN_ID"})
with pytest.raises(RuntimeError, match="API key"):
ext.provision(_profile(), {}, "modal")

View file

@ -0,0 +1,78 @@
"""Reachability enrichment tests."""
from __future__ import annotations
from datetime import UTC, datetime
from sandboxer.models import (
ActorType,
Consumer,
Profile,
Reachability,
ReachabilitySpec,
SandboxState,
SandboxStatus,
)
from sandboxer.reachability.enrich import (
build_reachability_report,
enrich_reachability,
ssh_one_liner,
)
def _profile() -> Profile:
return Profile.model_validate(
{
"id": "profile.agent-dev",
"version": "1.0.0",
"extension": "ext.compose-ssh",
"reachability": ReachabilitySpec(
tunnel="ops-bridge", identity="ops-warden"
).model_dump(),
}
)
def test_enrich_adds_tunnel_and_identity(monkeypatch) -> None:
monkeypatch.setenv("SANDBOXER_TUNNEL_PORT", "12222")
reach = enrich_reachability(
{"ssh": "build@localhost", "remote_dir": "/build/sbx-1", "host": "localhost"},
_profile(),
{"ssh_port": "12222"},
)
assert reach["identity"] == "ops-warden"
assert reach["tunnel"] == "localhost:12222"
assert reach["tunnel_via"] == "ops-bridge"
def test_ssh_one_liner() -> None:
reach = Reachability(ssh="user@host", remote_dir="/tmp/ws")
line = ssh_one_liner(reach)
assert line is not None
assert "user@host" in line
assert "/tmp/ws" in line
def test_build_reachability_report() -> None:
now = datetime.now(UTC)
status = SandboxStatus(
sandbox_id="abc12345",
profile_id="profile.agent-dev",
extension_id="ext.compose-ssh",
state=SandboxState.READY,
consumer=Consumer(actor=ActorType.AGT, project="glas-harness"),
host="coulombcore",
reachability=Reachability(
ssh="root@coulombcore",
remote_dir="/tmp/sandboxer/abc12345",
tunnel="localhost:22",
tunnel_via="ops-bridge",
identity="ops-warden",
),
created_at=now,
updated_at=now,
)
report = build_reachability_report(status)
assert report["sandbox_id"] == "abc12345"
assert report["ssh_one_liner"] is not None
assert "ops_bridge" in report

View file

@ -16,7 +16,12 @@ def _burst_profile() -> Profile:
"extension": "ext.compose-ssh",
"route": {
"strategy": "prefer-self-hosted",
"extensions": ["ext.compose-ssh", "ext.saas-stub"],
"extensions": [
"ext.compose-ssh",
"ext.e2b",
"ext.modal",
"ext.saas-stub",
],
},
}
)
@ -42,12 +47,23 @@ def test_prefer_self_hosted_when_host_set(monkeypatch: pytest.MonkeyPatch) -> No
def test_prefer_self_hosted_falls_back_to_saas(monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.delenv("SANDBOXER_HOST", raising=False)
monkeypatch.delenv("E2B_API_KEY", raising=False)
monkeypatch.delenv("MODAL_TOKEN_ID", raising=False)
monkeypatch.setenv("SANDBOXER_FORCE_SAAS", "1")
ext = resolve_extension(_burst_profile(), {}, host_override=None)
assert ext.id == "ext.saas-stub"
assert ext.capabilities.pricing_model == "metered"
def test_prefer_self_hosted_falls_back_to_e2b_when_credentialed(
monkeypatch: pytest.MonkeyPatch,
) -> None:
monkeypatch.setenv("SANDBOXER_FORCE_SAAS", "1")
monkeypatch.setenv("E2B_API_KEY", "test-key")
ext = resolve_extension(_burst_profile(), {}, host_override=None)
assert ext.id == "ext.e2b"
def test_lowest_cost_picks_metered_when_forced(monkeypatch: pytest.MonkeyPatch) -> None:
profile = _burst_profile()
profile.route = RouteSpec(

51
tests/test_secrets.py Normal file
View file

@ -0,0 +1,51 @@
"""Setup secret resolution tests."""
from __future__ import annotations
import pytest
from sandboxer.models import Profile, SetupSpec
from sandboxer.secrets.resolver import resolve_secret_ref, resolve_setup_secrets
def test_resolve_secret_ref_from_env(monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.setenv("SANDBOXER_SECRET_BUILD_REGISTRY_TOKEN", "tok123")
assert resolve_secret_ref("build-registry-token") == "tok123"
def test_resolve_setup_secrets_success(monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.setenv("SANDBOXER_SECRET_BUILD_REGISTRY_TOKEN", "tok123")
profile = Profile.model_validate(
{
"id": "profile.build",
"version": "1.0.0",
"extension": "ext.vm-packer",
"setup": SetupSpec(secret_refs=["build-registry-token"]).model_dump(),
}
)
secrets = resolve_setup_secrets(profile)
assert secrets["build-registry-token"] == "tok123"
def test_resolve_setup_secrets_missing_raises() -> None:
profile = Profile.model_validate(
{
"id": "profile.build",
"version": "1.0.0",
"extension": "ext.vm-packer",
"setup": SetupSpec(secret_refs=["missing-ref"]).model_dump(),
}
)
with pytest.raises(ValueError, match="Unresolved secret_refs"):
resolve_setup_secrets(profile)
def test_empty_secret_refs() -> None:
profile = Profile.model_validate(
{
"id": "profile.compose-e2e",
"version": "1.0.0",
"extension": "ext.compose-ssh",
}
)
assert resolve_setup_secrets(profile) == {}

View file

@ -3,6 +3,7 @@
from __future__ import annotations
from pathlib import Path
from subprocess import CompletedProcess
from unittest.mock import patch
import pytest
@ -21,6 +22,16 @@ def _profile() -> Profile:
)
def _build_profile() -> Profile:
return Profile.model_validate(
{
"id": "profile.vm-packer-build",
"version": "1.0.0",
"extension": "ext.vm-packer",
}
)
def test_provision_attach_via_alias(tmp_path: Path) -> None:
repo = tmp_path / "proj"
repo.mkdir()
@ -73,4 +84,66 @@ def test_teardown_preserves_vm() -> None:
with patch("sandboxer.extensions.vm_packer.SSHConfig.run", return_value=(0, "")):
report = ext.teardown(handle)
assert report["vm_preserved"] == "true"
assert report["workspace_removed"] == "True"
assert report["workspace_removed"] == "True"
def test_provision_build_runs_packer(tmp_path: Path) -> None:
template = tmp_path / "haskell"
template.mkdir()
(template / "haskell-build.pkr.hcl").write_text('packer {}')
ova = template / "haskell-build-20260624.ova"
ova.write_bytes(b"ova")
ext = VMPackerExtension()
with (
patch(
"sandboxer.extensions.vm_packer.subprocess.run",
return_value=CompletedProcess(args=[], returncode=0, stdout="", stderr=""),
) as run,
patch("sandboxer.extensions.vm_packer.emit_progress_note"),
):
handle = ext.provision(
_build_profile(),
{"vm_name": "haskell-build", "packer_template": str(template)},
"localhost",
)
assert handle["mode"] == "build"
assert handle["artifact_path"] == str(ova)
assert run.call_count == 2
init_cmd, build_cmd = [c.args[0] for c in run.call_args_list]
assert init_cmd[:2] == ["packer", "init"]
assert build_cmd[:2] == ["packer", "build"]
assert "-var" in build_cmd and "vm_name=haskell-build" in build_cmd
def test_provision_build_requires_template() -> None:
ext = VMPackerExtension()
with pytest.raises(ValueError, match="packer_template"):
ext.provision(_build_profile(), {"vm_name": "haskell-build"}, "localhost")
def test_wait_ready_build_checks_artifact(tmp_path: Path) -> None:
ova = tmp_path / "haskell-build.ova"
ova.write_bytes(b"ova")
ext = VMPackerExtension()
reach = ext.wait_ready(
{
"mode": "build",
"artifact_path": str(ova),
"host": "localhost",
}
)
assert reach["endpoint"] == str(ova)
def test_teardown_build_preserves_artifact() -> None:
ext = VMPackerExtension()
report = ext.teardown(
{
"mode": "build",
"artifact_path": "/tmp/haskell-build.ova",
}
)
assert report["artifact_preserved"] == "true"
assert report["workspace_removed"] == "false"

View file

@ -85,9 +85,9 @@ state_hub_task_id: "7075d8fc-44d1-48fe-b061-a05385de31a8"
```task
id: SAND-WP-0006-T06
status: wait
status: cancel
priority: low
state_hub_task_id: "c514c6b3-2be8-41e4-8e84-88f20ad80f7e"
```
Real `ext.e2b` / `ext.modal` adapters, BYOK via OpenBao, fin-hub export.
Superseded by SAND-WP-0010 (E2B/Modal adapters, BYOK, fin-hub export).

View file

@ -21,8 +21,7 @@ registry, HTTP parity).
Gap analysis: `history/2026-06-24-post-wp0007-intent-scope-gap-analysis.md`
**Predecessor:** SAND-WP-0007 (snapshots — finished)
**Follow-on:** SAND-WP-0010 (real cloud adapters), SAND-WP-0011 (reachability +
consumer profiles), SAND-WP-0012 (Packer orchestration)
**Follow-on:** SAND-WP-0010, SAND-WP-0011, SAND-WP-0012 (workplans filed 2026-06-24)
---

View file

@ -0,0 +1,114 @@
---
id: SAND-WP-0010
type: workplan
title: "Cloud adapters and billing export"
domain: infotech
repo: sand-boxer
status: finished
owner: codex
topic_slug: custodian
created: "2026-06-24"
updated: "2026-06-24"
state_hub_workstream_id: "a78ea9f1-d076-43a6-8798-36fc333edd25"
---
# Cloud adapters and billing export
Replace `ext.saas-stub` with real metered cloud backends (E2B, Modal) and wire
BYOK credential routing plus fin-hub billing export.
Gap analysis P5/P10: `history/2026-06-24-post-wp0007-intent-scope-gap-analysis.md`
Supersedes: SAND-WP-0006-T06
**Predecessor:** SAND-WP-0009 (TTL — finished)
**Follow-on:** SAND-WP-0011 (reachability + consumer profiles)
---
## Credential routing and BYOK
```task
id: SAND-WP-0010-T01
status: done
priority: high
state_hub_task_id: "5aeb6a17-dc5b-4e39-996c-f7f31c2659f4"
```
`src/sandboxer/extensions/credentials.py`; docs in `docs/payments.md` and
`docs/cloud-adapters.md`.
## ext.e2b adapter
```task
id: SAND-WP-0010-T02
status: done
priority: high
state_hub_task_id: "205bc70a-aaa6-4cd7-b0a5-11669490c150"
```
`ext.e2b`, `profile.e2b-burst`, mocked HTTP tests in `tests/test_e2b.py`.
## ext.modal adapter
```task
id: SAND-WP-0010-T03
status: done
priority: high
state_hub_task_id: "b9f104e1-e9d6-4324-b82a-8406be3006e5"
```
`ext.modal`, `profile.modal-gpu`, `tests/test_modal.py`.
## Routing and credits integration
```task
id: SAND-WP-0010-T04
status: done
priority: high
state_hub_task_id: "c11bdaf1-4c25-4c14-a566-0e28b0bd8b1d"
```
`profile.burst-sandbox` routes e2b → modal → stub; credential-aware resolver.
## fin-hub billing export
```task
id: SAND-WP-0010-T05
status: done
priority: medium
state_hub_task_id: "4eb1b0df-c6d7-4fb6-a7a4-e1455d2fac61"
```
`payments/billing_export.py`; hook on metered destroy; `tests/test_billing_export.py`.
## Docs and capability registry
```task
id: SAND-WP-0010-T06
status: done
priority: medium
state_hub_task_id: "d0aba132-b8fa-461b-b722-099868bf1770"
```
`docs/cloud-adapters.md`; registry A5/C5; routing/payments/extension-sdk updates.
## Tests and smoke
```task
id: SAND-WP-0010-T07
status: done
priority: high
state_hub_task_id: "3aebb3be-ae5e-4642-9710-9d80a1e8a582"
```
77 tests; `scripts/smoke-cloud-adapter.sh` (operator, credential-gated).
---
## Acceptance criteria
- E2B and Modal adapters provision/teardown via CLI with mocked CI
- BYOK documented; no secrets in repo or State Hub payloads
- `profile.burst-sandbox` routes to E2B when creds available
- fin-hub export hook callable when `SANDBOXER_FIN_HUB_URL` set
- SAND-WP-0006-T06 cancelled (superseded)

View file

@ -0,0 +1,134 @@
---
id: SAND-WP-0011
type: workplan
title: "Reachability and consumer profiles"
domain: infotech
repo: sand-boxer
status: finished
owner: codex
topic_slug: custodian
created: "2026-06-24"
updated: "2026-06-24"
state_hub_workstream_id: "614a59b5-1b95-4e5d-9014-676c69a99b5f"
---
# Reachability and consumer profiles
Formalize ops-bridge tunnel attachment in reachability descriptors and ship
first-class profiles for glas-harness and snuggle-inventor consumers.
Gap analysis P6/P7: `history/2026-06-24-post-wp0007-intent-scope-gap-analysis.md`
**Predecessor:** SAND-WP-0010 (cloud adapters)
**Follow-on:** SAND-WP-0012 (Packer orchestration)
---
## Reachability descriptor enrichment
```task
id: SAND-WP-0011-T01
status: done
priority: high
state_hub_task_id: "ccf21aaf-9439-41e2-9ce3-becc08f734a7"
```
Extend `Reachability` model: optional `tunnel` (local port / alias), `tunnel_via`
(ops-bridge route id), `identity` (warden actor hint). Populate from profile
`reachability` spec + `SANDBOXER_TUNNEL_*` env on compose-ssh / vm-packer.
Document contract in `docs/meta-framework.md`; sand-boxer does not own tunnels.
## ops-bridge integration helper
```task
id: SAND-WP-0011-T02
status: done
priority: medium
state_hub_task_id: "61d41e09-ca21-4fbe-9b56-98f0ffe356c6"
```
`sandboxer reachability show <id>` and `GET /v1/sandboxes/{id}/reachability`
surfacing SSH one-liner and tunnel status pointer (`ops-bridge` MCP / CLI doc
link). No tunnel bring-up in sand-boxer — pointer only.
## profile.agent-dev
```task
id: SAND-WP-0011-T03
status: done
priority: high
state_hub_task_id: "1a10a784-6a7c-4af6-9fbf-48d31e7e22cb"
```
Profile for glas-harness: longer TTL defaults, `actor: agt` examples, route
`prefer-self-hosted`. Extension `ext.compose-ssh`. Updated
`docs/integrations/glas-harness.md` with real profile id.
## profile.build (snuggle-inventor)
```task
id: SAND-WP-0011-T04
status: done
priority: high
state_hub_task_id: "a8142492-32c8-40d4-b882-b555858b44bb"
```
Build sandbox profile binding `ext.vm-packer`; `setup.instructions` placeholder;
`secret_refs` list on profile (resolution v0: env `SANDBOXER_SECRET_*`, inject at
provision boundary only). Updated `docs/integrations/snuggle-inventor.md`.
## Secret boundary v0
```task
id: SAND-WP-0011-T05
status: done
priority: medium
state_hub_task_id: "df4053de-ec74-40a3-ae9b-422c1be973cd"
```
`SetupSpec.secret_refs` resolution in manager pre-provision hook via
`SANDBOXER_SECRET_<REF>` env; pass to extension handle; never store on
`SandboxStatus` or emit to State Hub. Tests with mocked resolver.
## Consumer smoke scripts
```task
id: SAND-WP-0011-T06
status: done
priority: medium
state_hub_task_id: "9d5feebe-16a2-4448-ad0c-3276858341d1"
```
`scripts/smoke-agent-dev.sh`, `scripts/smoke-build-profile.sh` (CoulombCore
gated). Integration section in each consumer doc.
## Tests and docs
```task
id: SAND-WP-0011-T07
status: done
priority: high
state_hub_task_id: "849e0701-fe8f-4c08-ac24-98cdf554c24b"
```
Model tests for reachability fields; profile loader tests; updated `SCOPE.md`
profile catalog. `make check` green.
---
## Out of scope
| Item | Track |
|------|-------|
| glas-harness tool execution | glas-harness repo |
| snuggle code generation | snuggle-inventor repo |
| ops-bridge tunnel automation | ops-bridge repo |
---
## Acceptance criteria
- `profile.agent-dev` and `profile.build` load and create via CLI
- Reachability JSON includes tunnel metadata when profile declares ops-bridge
- secret_refs resolved at boundary; absent from agent-visible status payload
- Consumer integration docs reference real profile ids

View file

@ -0,0 +1,121 @@
---
id: SAND-WP-0012
type: workplan
title: "Packer build orchestration"
domain: infotech
repo: sand-boxer
status: finished
owner: codex
topic_slug: custodian
created: "2026-06-24"
updated: "2026-06-24"
state_hub_workstream_id: "87838886-0f4a-4eae-8d0e-b464933089de"
---
# Packer build orchestration
Trigger Packer builds from `sandboxer create` and ship the-custodian
`make remote-build` shim — completing the build-machines migration arc.
Gap analysis P8: `history/2026-06-24-post-wp0007-intent-scope-gap-analysis.md`
Carries forward: SAND-WP-0005-T06 (deferred)
**Predecessor:** SAND-WP-0011 (consumer profiles)
**Follow-on:** reuse-surface federation publish; sandboxer01 operator track
---
## Packer build mode on ext.vm-packer
```task
id: SAND-WP-0012-T01
status: done
priority: high
state_hub_task_id: "9dc30d94-1797-4c35-81a0-e75e5414f6fc"
```
`VMPackerExtension` build mode: inputs `packer_template`, `vm_name` trigger
local Packer run per the-custodian `infra/build-machines/` conventions.
Distinct from attach mode; teardown preserves OVA artifact. Tests mocked subprocess.
## profile.vm-packer-build
```task
id: SAND-WP-0012-T02
status: done
priority: high
state_hub_task_id: "8e30794c-d8b9-48c7-ae93-db84724eedf2"
```
Profile binding build mode with placement and TTL suitable for long builds.
Documented inputs in `docs/migration-build-machines.md`.
## Manager and CLI integration
```task
id: SAND-WP-0012-T03
status: done
priority: high
state_hub_task_id: "685f766c-90ae-4698-87d0-b61535e7491a"
```
`create` selects build vs attach via profile or `inputs.mode=build|attach`.
Progress events to State Hub during long provision. CLI help text updated.
## the-custodian remote-build shim
```task
id: SAND-WP-0012-T04
status: done
priority: medium
state_hub_task_id: "6c4c0f85-5153-4fe9-84e6-26c5c9d33bb1"
```
`make remote-build PROJECT=` in build-machines delegates to
`sandboxer create --profile profile.vm-haskell-build` when CLI present;
legacy rsync path retained with deprecation notice.
`scripts/verify-remote-build-shim.sh` mirrors SAND-WP-0004 pattern.
## Port-registry automation
```task
id: SAND-WP-0012-T05
status: done
priority: low
state_hub_task_id: "701b2640-36ea-4702-b660-7169a4ec72cc"
```
Documented read-only port-registry pointer in `docs/migration-build-machines.md`;
full ops-bridge automation deferred.
## Docs, tests, runbook
```task
id: SAND-WP-0012-T06
status: done
priority: high
state_hub_task_id: "2378cd6a-ac23-47e9-a5d9-0d80b9e9f7af"
```
Updated `docs/migration-build-machines.md`, `docs/extension-sdk.md`, operator
runbook `docs/runbooks/profile-vm-packer-build.md`. Build mode cases in
`tests/test_vm_packer.py`. `make check` green (90 tests).
---
## Out of scope
| Item | Track |
|------|-------|
| OVA import on hypervisor | Operator / build-machines |
| systemd build-agent changes | the-custodian infra |
| sandboxer01 host | Infra operator |
---
## Acceptance criteria
- Build mode provisions via CLI with mocked Packer in CI
- Attach mode unchanged (backward compatible)
- the-custodian shim documented and verified
- SAND-WP-0005-T06 superseded when complete