Repository navigation
Releases
v0.4.1
What's Changed
- fix!: stop pointing at
veris-daytona teardown, a command #42 deleted by @alikhan126 in #43 - fix: send the Veris API key on twin control_url calls by @vcshih in #53
- chore: 0.4.1 by @vcshih in #54
Full Changelog: v0.4.0...v0.4.1
v0.4.0
v0.3.1
v0.3.0
What's Changed
- ci: drop the provenance-landed assertion by @alikhan126 in #18
- fix: tell the agent TLS trust is preconfigured, so it stops prefixing commands by @alikhan126 in #20
- feat:
veris-daytona run— the SDK as one command, with a receipt-gated exit code by @ritiztambi in #19 - docs: point at @veris-ai/veris-sim-opencode, the renamed veris-sim plugin by @alikhan126 in #22
- fix: trust Node via NODE_OPTIONS=--use-openssl-ca, since Daytona overwrites the CA variables by @alikhan126 in #21
- chore: drop CHANGELOG.md, the GitHub release notes are the changelog by @alikhan126 in #23
- fix: run exports the Veris trust variables on every command, since Daytona overwrites them by @ritiztambi in #24
- fix: fit Daytona's 20-domain allowlist, allow the twin's own host, and read the whole receipt by @ritiztambi in #25
- fix: delete the sandbox a failed create left behind, and say why it failed by @ritiztambi in #27
- feat:
provisionandteardown— hand a wired sandbox over, and take it back by @ritiztambi in #26 - chore: land provision, teardown and the failed-create fixes on main by @ritiztambi in #28
- feat:
pushandexec— get code into a provisioned box, and run things in it by @ritiztambi in #29 - fix: read the CLI's login profile, and say why a teardown was refused by @alikhan126 in #30
- fix(opencode): correct session and receipt guidance by @alikhan126 in #31
- feat: add run-scoped receipts and session controls for skills by @alikhan126 in #36
- fix: pin Daytona to the gateway address, since a domain allowlist turns on TLS inspection by @vcshih in #38
New Contributors
- @ritiztambi made their first contribution in #19
- @vcshih made their first contribution in #38
Full Changelog: v0.2.1...v0.3.0
v0.2.1
-
https://www.npmjs.com/package/@veris-ai/daytona-opencode/v/0.2.1
-
NODE_EXTRA_CA_CERTSnow points at the merged bundle, not the lone Veris
cert. That variable extends Node's baked-in Mozilla roots, never the
system store — and the leaf Node actually validates in a Daytona sandbox is
signed by Daytona's proxy CA, which lives only in the system store. So with
the single-cert file, every Node HTTPS call to a vendor host failed with
UNABLE_TO_VERIFY_LEAF_SIGNATURE, and agents in the sandbox learned to
compensate by prependingNODE_EXTRA_CA_CERTS=/tmp/veris-ca-bundle.crtto
every command by hand. Every other trust variable already pointed at the
bundle (system store — Daytona's CA with it — plus ours); Node's was the one
exception, and now it isn't. Plainnodeworks in a fresh sandbox, no flag,
no prefix.
What's Changed
- fix: point NODE_EXTRA_CA_CERTS at the merged bundle by @alikhan126 in #17
Full Changelog: v0.2.0...v0.2.1
v0.2.0
Promotes 0.2.0-rc.1 unchanged, plus two errors that now name their cause.
- https://www.npmjs.com/package/@veris-ai/daytona/v/0.2.0
- https://www.npmjs.com/package/@veris-ai/daytona-opencode/v/0.2.0
Everything in 0.2.0-rc.1, promoted unchanged, plus two errors that named
a failure without naming its cause. Both were found the same way — by a user
losing an evening to a working setup that reported nothing useful.
VerisErrornow folds the wrapped error's message into its own.cause
was always attached, but almost nothing that shows an error to a human walks
the cause chain: OpenCode printserr.messageand stops. So an invalid
Daytona API key surfaced asDaytona sandbox create failed, full stop, when
the SDK hadDaytonaAuthenticationError: Invalid credentialsin hand the
whole time. It now readsDaytona sandbox create failed: Invalid credentials.
The cause stays reachable; a runaway message is truncated rather than pasted
whole.- A 404 creating a twin now explains itself. Environments belong to one
control plane, so the usual cause of that 404 is an id from the other one —
a prodVERIS_ENVIRONMENT_IDagainst a devVERIS_API_BASE, or the reverse.
The old message wascreate sandbox in environment kl833…: 404, which sends
people to check their API key and their network first; both are fine, and
neither is the problem. It now names the base it asked and says the two
variables must agree.
What's Changed
- fix: stop the provenance check racing the registry, and unblock repair runs by @alikhan126 in #10
- feat: give the agent the twin's lifecycle, via Veris's MCP rather than new tools by @alikhan126 in #12
- fix(release): widen the ingestion wait, and keep the forgot-to-bump signal by @alikhan126 in #11
- fix(ci): exempt Dependabot from the PR-title lint by @alikhan126 in #14
- chore(deps): Bump actions/checkout from 5.1.0 to 7.0.1 by @dependabot[bot] in #4
- chore(deps-dev): Bump vitest from 3.2.7 to 4.1.11 by @dependabot[bot] in #7
- chore(deps-dev): Bump @types/node from 22.20.1 to 26.4.0 by @dependabot[bot] in #9
- chore(deps): Bump actions/setup-node from 5.0.0 to 7.0.0 by @dependabot[bot] in #5
- chore(deps): Bump @opencode-ai/plugin from 1.1.59 to 1.18.25 by @dependabot[bot] in #8
- docs: say how to add opencode-veris-sim, and what it actually gives you by @alikhan126 in #13
- fix: make two errors name their cause instead of just their failure by @alikhan126 in #15
- ci: publish on release, and cut 0.2.0 by @alikhan126 in #16
New Contributors
- @dependabot[bot] made their first contribution in #4
Full Changelog: v0.1.1...v0.2.0
v0.2.0-rc.1
Release candidate for 0.2.0. Published under the rc dist-tag; latest remains 0.1.1.
- https://www.npmjs.com/package/@veris-ai/daytona/v/0.2.0-rc.1
- https://www.npmjs.com/package/@veris-ai/daytona-opencode/v/0.2.0-rc.1
Both carry SLSA provenance attestations. The workflow's provenance check timed out at 100s and skipped this step; the attestations landed shortly after.
The agent could see what the twin received. It can now also ask what the twin
is, and manage its lifecycle — without either package growing a tool surface
that a second engine integration would have to grow again.
- The Veris MCP server is registered automatically (
@veris-ai/daytona-opencode).
Twin lifecycle — read the environment, promote a sandbox to the environment's
baseline, reset it — comes from Veris's own MCP rather than from tools
reimplemented here, so it stays one codebase across every engine. It uses the
VERIS_API_KEYalready required, and registers nothing when that is unset:
a server wired with an empty credential fails every call and reads as a Veris
outage rather than a missing setting. Everything is??=, so a user's own
opencode.jsonwins and an existingopencode-veris-siminstall does not
double-register. create_sandboxanddelete_sandboxare denied by default, with
promote_sandboxandreset_sandboxset to ask. The plugin creates and owns
the session's twin, which makes the first two always wrong: a second twin takes
neither the traffic nor the receipts with it, so an agent would seed it and
report success while nothing changed — the exact failure the receipt exists to
catch, one layer up.- New tool
verisTwinreports the twin's id and the services it answers for,
and with a service argument returns that service's manual. The id is the piece
no MCP server can supply: the session→sandbox→twin mapping lives in the
plugin's session manager, which is also why the receipt is a tool rather than
an MCP call. - New
sbx.veris.manual(service)on@veris-ai/daytona, alongside a new
twin-stateerror phase. Likereceipt(), it reads the twin's control plane
from the host —control_urlstays off the sandbox's allowlist, because code
that could reach/veris/resetcould erase its own receipt. - The system prompt now states that the plugin owns the twin's lifecycle and
that nothing here runs underveris-proxy. This matters when
opencode-veris-simis also installed: its commands describe a
laptop-plus-proxy setup, and saying the premise is false is cheaper than
suppressing them.
v0.1.1
- https://www.npmjs.com/package/@veris-ai/daytona/v/0.1.1
- https://www.npmjs.com/package/@veris-ai/daytona-opencode/v/0.1.1
No functional change to either package. This is the release-pipeline release:
0.1.0 was published by hand and carries no provenance attestation, and cutting
0.1.1 through the workflow is the only way to prove the trusted-publishing
path — a dry run skips the publish entirely.
- The publish guard is per-package. It used to refuse the whole run if either
package was already on npm, which made a partial release unrecoverable: if the
SDK published and the plugin then failed, the retry aborted on the SDK it had
just successfully published. It now skips what is published, publishes the
rest, and fails only when there is nothing to do. The git tag and the GitHub
release are idempotent for the same reason. - Prereleases can be published. npm 11 refuses a prerelease without an explicit
--tag, and the workflow passed none, so everyx.y.z-rc.nwould have failed
at the publish step after the build, tests and pack had all passed. The
dist-tag is now derived from the version (0.2.0-rc.1→rc) and can be
overridden with the newdist_taginput. - The workflow asserts its own provenance: a publish that lands without a
dist.attestationsnow fails the run instead of shipping and looking fine. - Tarballs are checked for a stray
node_modulesand against a 1 MB ceiling. npm run releaseis gone. The workflow is the only publish path.version:setrefuses to write a1.xversion while the plugin's dependency
on the SDK is a caret range, which only pins a patch range below 1.0.0.- Hardening: job-scoped workflow permissions, a
releaseconcurrency group,
actions pinned by commit SHA, Dependabot, and Conventional Commits enforced on
the PR title.
v0.1.0
First release. Both packages version together.
- https://www.npmjs.com/package/@veris-ai/daytona/v/0.1.0
- https://www.npmjs.com/package/@veris-ai/daytona-opencode/v/0.1.0
Published manually to bootstrap the scope — npm cannot attach a Trusted Publisher to a package that does not exist yet. Releases from 0.1.1 onward run from the release workflow.
@veris-ai/daytona
- A drop-in for
@daytona/sdk: re-exports the whole SDK and overrides only
Daytona, so a single changed import puts a Veris twin behind every sandbox. create()provisions the twin, mints an egress credential, sets the
domainAllowListandoutboundProxyUrl, installs CA trust, and proves the
tunnel with a canary probe before it resolves.get()andlist()rehydrate the same surface from sandbox labels, so a
resumed session keeps its receipts.delete()is wrapped to remove the twin, so nothing leaks.sbx.veris:receipt(),assertTouched(),services(),
getDataPlaneEnv(),getTrustEnv(),deliverTo().
@veris-ai/daytona-opencode
- Fork of
@daytona/opencode0.192.0, four changed files: the SDK import, the
tool registry, the system prompt, and a check that the Veris coordinates are
set. - Adds the
verisReceipttool.