The preflight is the documented gate before real SSO traffic, but it did
not check two things the dsh-demo install got wrong:
- The gateway Role must grant `patch` on
extensions.agents.x-k8s.io/sandboxclaims. The SSO landing path rotates
the claim's token-hash annotation whenever the browser has no valid
gateway session (first visit, expired or foreign cookie, gateway
restart), so a missing verb makes every such sign-in answer 503.
Verify with `kubectl auth can-i --as`; when a read-only kubeconfig
cannot impersonate, fall back to reading the rules of the Role (or
ClusterRole) the RoleBinding points at.
- The variant's NetworkPolicies must be applied. Without them any Pod in
the cluster can reach dsh-gateway:8080 and spoof X-Forwarded-User, so
the exposure is demo-only. Fail the preflight while either policy from
75-networkpolicies.yaml is missing.
Also fix a false positive in the cookie-secret check: oauth2-proxy accepts
16, 24 or 32 raw bytes, or URL-safe unpadded base64 that decodes to one of
those lengths (pkg/encryption.SecretBytes). The documented
`secrets.token_urlsafe(32)` value is 43 base64 characters that decode to 32
bytes, so it is valid and must not be reported as 43 raw bytes.
The docs that enumerate the preflight checks are updated to match, and
75-networkpolicies.yaml notes that the preflight enforces it.
Verified with shellcheck, yamllint, make check and the gateway test suite
(150 passed); the new checks were exercised read-only against the live
dsh-demo cluster.
Entire-Checkpoint: 01M6FNY2WTC4GMAJ8BAHABN88M
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
* fix(gateway): grant patch on sandboxclaims for token rotation
The SSO landing path calls K8sClient.patch_token(), which patches the
`dsh-<user>` SandboxClaim's token-hash annotation whenever the browser has
no valid gateway session: first visit, an expired or foreign cookie, or a
gateway restart that drops the in-memory session store. The gateway Role
only granted get/list/create/delete on sandboxclaims, so the patch returned
403 and, because ApiException is not handled, every such GET / answered a
bare 500 "Internal Server Error".
Grant the missing `patch` verb on
extensions.agents.x-k8s.io/sandboxclaims, and update the least-privilege
comment plus the two docs that enumerate the Role verbs.
Verified with `make check` (versions, example layout, manifests) and
`yamllint`; the 403 and the resulting 500 were reproduced from a running
install and cleared by applying the equivalent Role patch (RBAC is evaluated
per request, so no gateway restart is needed).
* fix(gateway): return 503 instead of a bare 500 on Kubernetes API errors
Every gateway route talks to the Kubernetes API, and ApiException used to be
unhandled: Starlette's ServerErrorMiddleware turned it into a bare
500 "Internal Server Error" plus an ERROR traceback. That masked a missing
RBAC verb as a client-visible crash on GET / (the SSO landing path), and it
would equally hide an API outage or throttling.
Register an ApiException handler that logs one credential-free line
(status/reason) and answers 503 + Retry-After with a fixed body, so the
Kubernetes Status payload (which can name cluster objects) is never proxied.
Guard the first get_claim() in the WebSocket route the same way: WebSocket
scopes bypass Starlette's HTTP exception handlers, so it now closes with 1011
instead of escaping as a traceback.
Tests assert the 503 + Retry-After response, cover the SSO landing path that
produced the reported 500 (forbidden claim patch), and assert the WebSocket
close code. The failure-semantics tables document the new behaviour.
---------
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
- Replace user:token cookies with opaque server-side sessions, revocable on logout, delete, recreate and expiry; bind every credential to the path user and add cross-user regression tests.
- Rate-limit /users and /login, add optional signup invite codes and form same-origin checks, and unify login failures to 401.
- Require explicit ALLOW_INSECURE_COOKIES for plain HTTP, use __Host-dsh_gw_session when secure, and fix secure cookie deletion.
- Strip client-supplied forwarding/SSO headers, add an optional upstream allowlist, per-user in-flight and WebSocket message caps, credential-free handshake errors, and sweep-time state pruning.
- Digest-pin base images, hash-pin gateway dependencies, verify DSH_COMMIT in the build, and run the gateway image as non-root with a pinned Python base.
- Update manifests, production checklist, architecture/security docs and add an SSO preflight script.
BREAKING CHANGE: existing dsh_gw_session=<user>:<token> cookies are invalid; users must sign in again. Unknown users on POST /login now return 401 instead of 404, and the gateway refuses COOKIE_SECURE=false unless ALLOW_INSECURE_COOKIES=1.
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
Update AGENT_SANDBOX_VERSION from v1.0.8 to v1.0.9, the pin the examples
and the CI manifest dry-run consume. Refresh the compatibility matrix in
docs/versioning.md and clarify the failedPodPolicy sentence in the
hardening doc, which described v1.0.8 as the pin.
Reviewed the v1.0.9 release for breaking changes: the new
volumeClaimTemplates admission rules accept the named "dsh-home" templates
in both examples, and the router NetworkPolicy / Go label-constant changes
do not apply to this repository.
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
Remove obsolete layout/migration prose and legacy gateway-alias references, reconcile the required-check policy, fix version-label/RBAC/workspace/channel drift, and de-duplicate the example and gateway docs.
Rename the minimal manifest to 10-sandboxtemplate.yaml and mount the dsh-home PVC's workspace subPath at /workspace so the documented persistence is real. Enforce numbered top-level manifests in check-examples.sh.
Verified with make check, yamllint, shellcheck, a Markdown link/anchor scan, and git diff --check.
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
The directory picker has no start-path setting: it lists $HOME whenever it
opens. With the image default HOME=/home/node the folder explorer started on
a throwaway directory, so point HOME at the PVC-backed /workspace in both
sandbox templates. `~` and $HOME-relative tool caches/config follow; DSH_HOME
stays /home/node/.dsh, so sessions, config and credentials are unaffected.
Document the trade-off in the image contract and both examples, add
troubleshooting for sandboxes whose pod env was stamped before the change,
and assert HOME/~ in the e2e walkthrough.
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
* fix(example): keep agents usable when DSH's inner sandbox has no backend
DSH confines bash commands itself (bubblewrap, else Landlock) and fails closed
with SANDBOX_UNAVAILABLE when no runner is usable. In this hardened pod -
non-root, capabilities dropped, seccomp RuntimeDefault - bubblewrap cannot
start (it needs CAP_SYS_ADMIN) and the pod exposes no Landlock runner, so the
shipped workspace-write default made every command fail:
sandbox mode "workspace-write" is requested but no sandbox backend is usable
The DSH base bundle at the pinned ref documents DSH_PERMISSION_MODE as the
explicit deployment override, so set it to danger-full-access in both sandbox
templates. The per-user pod is the isolation boundary in this example (its own
pod, PVC and Service; non-root; all capabilities dropped), so a working agent
is worth more than an inner file policy that cannot be enforced. Note that
danger-full-access also turns approval prompts off.
Trade-off documented rather than hidden: an agent command can now read its own
DSH_HOME (including .credentials.yaml) and write anywhere in the container.
Hosts where a runner works should keep workspace-write or read-only.
- main-image docs: DSH_PERMISSION_MODE row plus a note on the runners and the
fail-closed behavior
- example README: limits bullet and a troubleshooting row with the in-pod
backend check and how to restore a confined mode
- architecture: tenant-isolation bullet extended
- skill runbook/troubleshooting: operation guidance; never grant CAP_SYS_ADMIN
or seccompProfile: Unconfined to "fix" this without an explicit decision
Verified: make check (14 manifests), make lint, YAML assertions on both
templates, and the override expression at the pinned dsh-v0.2.1-rc.5 base
bundle (mode: !!js process.env.DSH_PERMISSION_MODE ?? 'workspace-write').
* fix(image): install bubblewrap for DSH's command sandbox
DSH's Linux sandbox runner chain is bubblewrap, then Landlock, and it fails
closed with SANDBOX_UNAVAILABLE when neither works. The image shipped neither,
so a sandboxed agent had no way to execute confined commands.
Add bubblewrap to the runtime stage. Debian ships it unsetuid (0.12.0), so it
still needs unprivileged user namespaces plus the namespace/mount syscalls: in a
hardened pod (all capabilities dropped, RuntimeDefault seccomp) it usually
cannot create one, and DSH then falls back to Landlock or, failing that, to the
DSH_PERMISSION_MODE=danger-full-access override the templates already set.
Where the host allows an unprivileged user namespace, the first-choice runner
now exists, so the confined modes can be restored.
Docs corrected to match: bubblewrap is no longer "absent"; the blocker is the
unsetuid binary's namespace requirement, not a missing CAP_SYS_ADMIN grant.
Verified: make check, make lint; the package exists in the pinned base
(node:24-trixie-slim, bubblewrap 0.12.0-1~deb13u1); a throwaway container with
the pod's flags (--user 1000 --cap-drop ALL --security-opt no-new-privileges)
still cannot create a namespace, which is why the override stays. No local image
build was run.
* ci(image): build and smoke-test the main image on PRs that touch it
The Dockerfile and the pins in versions.env could only be validated by a merge
to main or by running `make image` by hand; neither gives a PR-time signal.
Add ci-image.yaml: on pull requests that change Dockerfile, .dockerignore or
versions.env (and on manual dispatch) it builds linux/amd64 natively with the
pins sourced from versions.env - no registry login, no push - loads the image
locally and smoke-tests it:
- the packaged CLI runs and reports the pinned version
- the container is the non-root node user (1000) and owns /workspace + DSH_HOME
- the bubblewrap command-sandbox runner is present
- the OCI version label matches versions.env
- a container serves /favicon.svg (the same unauthenticated probe the shipped
manifests use), with logs filtered so the launch token is never printed
Path-filtered like the other on-demand checks, so it is not a required status
check; the published multi-arch build stays in build-main.yaml and release.yaml.
Docs updated in the same change: AGENTS.md repository map and testing table,
docs/gitflow.md checks section, docs/main-image.md build section.
Verified: actionlint 1.7.12 and yamllint 1.38.0 (CI versions) clean over the
whole repository, plus make check and make lint.
---------
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
/workspace - the DSH image's WORKDIR, where agents keep files - was plain
container storage, so every suspend/resume or pod replacement silently threw
agent work away while DSH state survived on the PVC.
Mount the same per-sandbox `dsh-home` PVC a second time at /workspace with
`subPath: workspace`. One PVC, no extra provisioning, and the subPath keeps
DSH state (sessions, config and .credentials.yaml at the volume root) out of
the agent-visible tree; mounting the volume root at /workspace would have
exposed the credential file.
The kubelet creates the missing subPath directory itself and chmods it to the
volume's fsGroup permissions, and the volume root's setgid bit makes the new
directory inherit gid 1000, so UID 1000 can write with no init container and
no root. The rare interrupted-kubelet case that leaves 0755 is documented with
a recovery procedure (kubernetes#142464) rather than papered over.
- e2e-kind: write a /workspace marker alongside the DSH_HOME marker, assert
the mount is the dsh-home subPath, and verify both survive suspend/resume.
- Docs: README (overview, steps 4-6, troubleshooting row), architecture
(PVC diagram, persistence, component table), main-image working-directory
row, platform runbook PVC row.
Verified: make check (14 manifests), make lint, plus YAML assertions on the
two mounts and the unchanged volumeClaimTemplates.
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
* fix(gateway): answer client-aborted requests with 499, not a traceback
A browser reload, back navigation or proxy timeout can abort a request while
its body is still arriving. Starlette then raises ClientDisconnect from
request.stream(); nothing caught it, so ServerErrorMiddleware logged a full
ERROR traceback per aborted request and buried real faults. The request never
reaches DSH (the body is buffered before the upstream send) and the peer can
no longer be answered, so this is not an error.
- Register a ClientDisconnect exception handler that answers 499
(CLIENT_CLOSED_REQUEST). One handler covers the proxy and the body-reading
control routes (signup, login). A request already streaming keeps its
upstream status because the response has started.
- Make _release await-free, and reach it even when the upstream close is
cancelled. CancelledError is a BaseException, so suppress(Exception) around
upstream.aclose() skipped the release and stranded the in-flight slot,
leaving that user's sandbox looking busy forever (the idle sweeper then
never suspends it).
- Add raw-ASGI regression tests: mid-body abort on the proxy and on signup,
mid-stream abort, a cancelled release, and a cancelled upstream close.
- Document the 499 semantics in the gateway README failure table and the
architecture failure modes.
Verified: make check, make test (114 passed), make lint. Against the pinned
uvicorn 0.54 the abort reproduces the original traceback before the fix and
logs no traceback after it.
* chore(deps): bump gateway starlette to 1.10.0 and ruff to 0.17.2
Absorbs Dependabot PR #27 against current main: that branch predates the
dsh-v0.2.1-rc.5 changelog/release commits, so merging it as-is would have
reverted them.
starlette 1.10.0 reworks URL-encoded form parsing (fields are built in the
parser callbacks, and the last field is kept when the stream ends without an
empty chunk) and clears tracebacks of handled exceptions. The gateway reads
JSON and form bodies, so the existing suite exercises signup/login forms and
the 499 client-abort path against the new release. ruff 0.17.2 only adds
preview and fix-safety changes; the configured rule set reports nothing new.
Verified: make check, make test (114 passed), make lint, make gateway (image
builds on python:3.12-slim and imports starlette 1.10.0), plus a direct ASGI
abort probe under 1.10.0 (499, in-flight slot released, nothing forwarded).
---------
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
Bump DSH_VERSION to dsh-v0.2.1-rc.5 and clear RELEASE_REVISION, making the release identity dsh-v0.2.1-rc.5. Update every derived manifest, image tag, and doc, and remove prose that described the previous identity's legacy-tag history.
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
agent-sandbox v1.0.8 adds spec.failedPodPolicy to Sandbox and
SandboxTemplate. The default, Retain, leaves a Failed Pod in place and
surfaces Finished=True; Recreate deletes it and starts a replacement with
the same Sandbox identity and PVCs.
Ship it as an opt-in toggle next to the other commented fields, with a
new section 5 in the optional-hardening doc: what it changes, how to
enable it for new, warm-pool and already-claimed Sandboxes (the policy is
stamped at creation), a failure drill, caveats and rollback. The base
walkthrough stays on Retain because a deterministic failure would
otherwise recreate forever, and because the templates use
restartPolicy OnFailure, so the policy only matters for terminal cases
such as node eviction.
Cross-references updated in the example README, the optional-manifests
index, the architecture failure-mode table and the platform skill's
troubleshooting table.
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
Update AGENT_SANDBOX_VERSION to v1.0.8 and the docs that quote the
control-plane pin (example prerequisites, architecture, versioning table,
minimal-single-tenant template comment, platform skill).
The v1.0.8 release adds only one CRD field, spec.failedPodPolicy
(Sandbox and SandboxTemplate, enum Retain|Recreate, default Retain), so
the existing manifests keep their behavior: Retain is the default and
neither the API version (v1beta1) nor any field we set changed. All five
custom resources still validate against the v1.0.8 CRD schemas.
Behavior changes in the controller itself (bounded MCP output, router
kustomization, creation-latency metric timing) do not affect the
platform because it uses neither the sandbox-router nor the MCP server.
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
Add opt-in hardening to the agents-as-a-service example:
- commented runtimeClassName/nodeSelector fields for sandboxes and Deployments
- optional sandbox NetworkPolicy plus exposure-variant policies
- optional dsh-provider-keys-extra Secret for extra provider keys
- docs/optional-hardening.md with verification, caveats and rollback
The base walkthrough stays portable; no gateway code, routes or auth changed.
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
* feat(repo)!: restructure around the dsh-harness image and one example per implementation
- the root Dockerfile is the primary artifact; versions.env is the single source of truth
- merge gateway-api/dsh-demo into examples/dsh-agents-as-a-service (manifests/, exposure variants)
- move the single-tenant template into examples/minimal-single-tenant
- publish the gateway as a nested package under the repository image
- mirror the DSH ref for release identities (first release dsh-v0.2.1-rc.2-r2)
- CI: lint/python/manifests/gateway-image, reusable multi-arch build, tag-driven release, weekly e2e
- docs: README, docs/{main-image,examples,gitflow,versioning}, example READMEs, AGENTS.md, governance
- SSO is opt-in again; gateway pull policy and cookie defaults match the documented demo
BREAKING CHANGE: example manifest paths, the gateway image name, and the removal
of deploy/ and gateway-api/ affect existing deployments; see the migration table
in README.md.
* docs(changelog): generate initial changelog from history
* fix(scripts): make set-version update Dockerfile and sample revision pins
* build(hadolint): document intentional lint exceptions for the main image
---------
Co-authored-by: SREDevOpsAgents <agents@sredevops.org>
