- Status: Active
- Last substantive update: 2026-11-07
- Owner: svelderrainruiz
A public, project-agnostic engine for provisioning self-hosted build/test infrastructure and running native SBOM reporting pipelines, maintained by VI Technologies.
Anyone forking this repository can use it for their own projects by
supplying their own .toml/.json project configuration — this repository
ships with zero pre-committed project configs, VI
Technologies-specific identifiers, or private infrastructure state, except
in a small number of scripts explicitly disclosed as narrow-purpose,
consumer-vendored reference templates for one specific known consumer
(see #68
for the full history and remediation evidence).
Private consumers (such as VITechnologies/SandboxSBOMReporting) supply
their own real project configuration and infrastructure state separately,
and consume this repository's reusable workflows and scripts as a
versioned, pinned dependency — never the other way around. No credentials
or private configuration ever flow into this repository.
Assessed against svelderrainruiz/repo-standards-review's release-gate
profile, version v0.2.23 (commit ba72187). See
docs/architecture/standards-compliance.md
for the full gate status and the living re-certification policy,
docs/architecture/adr/ADR-0001-adopt-gitflow-for-release-governance.md
for the branch-governance decision this assessment drove, and
CONTRIBUTING.md for the branch model and the policy
on proactively filing issues for spec drift and future-agent-success
risks.
- Reusable GitHub Actions workflows are called by downstream repositories
via
uses: VITechnologies/projectbuddy/.github/workflows/<name>.yml@<commit-sha> # <tag>(pinned to the exact release commit SHA, with the tag as a trailing comment for readability -- seetoolchain-smoke-test.yml's own header comment for the canonical example, the only workflow here that is actuallyworkflow_call-consumable this way), matching the pattern already established for LabVIEW plugin module releases in thelabview-benchmark-actor-pluginsorg. workflow_dispatch-only workflows (add-project.yml,provision-windows-labview-vm.yml,provision-ubuntu-vm.yml,vm-fleet-status.yml) cannot be pinned viauses:@sha-- a consumer vendors its own copy of these instead (seeAdd-Project.ps1's own$WorkspaceRootdoc comment for the rationale, resolved as intentional in#48), accepting that vendored copies drift from this engine's own until a consumer deliberately re-syncs them.- Directly-invoked scripts (not wrapped in a GitHub Actions workflow) are
consumed via a second
actions/checkoutstep pinned to the exact release commit SHA (not just the tag), inside any downstream workflow that needs them — the same trust model GitHub's ownuses: owner/repo@<sha>syntax already applies to every action pinned throughout this engine's callers. This repository does not use git submodules for this purpose. - Upgrade path: downstream repositories bump their pinned tag/SHA
deliberately, as a reviewed code change (a version-bump commit), never
automatically and never by tracking a floating branch or
latest.
See CHANGELOG.md for released versions.
The candidate .NET source configuration CLI derives JSON/TOML drafts from a pinned checkout's existing VIPM manifest and a selected packed-library build. Configuration validity requires deterministic checks and local Ollama review; native-build proof remains separate. It does not register projects, overwrite existing outputs, or enable cleanup.
Dry-run-first VirtualBox builders that create self-hosted build/test VMs
(Windows 11 + LabVIEW/VIPM, or Ubuntu 24.04 + LabVIEW/VIPM), register them
as GitHub Actions self-hosted runners, and track them in a fork-local
runner-inventory.json (copy vm/windows-labview/runner-inventory.example.json
to get started — it is the committed schema/template; your own real
instance is fork-local and never committed here).
The VM identity values that actually name a fork's own infrastructure
(VM_NAME, BASEFOLDER, GUEST_FULLNAME, GUEST_HOSTNAME) are mandatory
inputs with no built-in default — these scripts never assume any fork's
own naming. (A few non-identity parameters, like the ISO path, guest
username, and WinRM port, do have generic, non-fork-specific defaults for
convenience; see each script's own param()/argument-parsing block for
the authoritative list.) vm/windows-labview/ additionally includes an NSIS/VNC
WinRM-bootstrap mechanism (winrm-setup-installer.nsi,
build-nsis-bootstrap-iso.sh, vnc-bootstrap.py) that configures WinRM on
a freshly-installed Windows guest without depending on FirstLogonCommands
surviving an in-guest restart — see the .nsi file's own header comment
for the full rationale.
.github/workflows/provision-windows-labview-vm.yml and
provision-ubuntu-vm.yml are the corresponding provisioning entry points,
each taking every real identity value as an explicit workflow_dispatch
input (some carry a generic, clearly fictional placeholder default; none
default to any specific fork's real infrastructure).
vm-fleet-status.yml is a separate, read-only verification workflow with
zero workflow_dispatch inputs — it re-derives ground truth from a
fork's own committed vm/windows-labview/runner-inventory.json (never
committed in this public engine repo itself, so its 30-minute schedule
trigger here finds no inventory and cleanly no-ops; a consuming fork with
a real, committed inventory gets full per-VM verification). See
docs/vm-provisioning/procedure.md
for the full phase-by-phase procedure (both VM families), including the
deliberate human-only LabVIEW activation boundary (VIPM activation on the
Ubuntu-LabVIEW family is a deliberate, narrower automated exception) and
the active fleet-status verification design.
docs/vm-provisioning/ubuntu-lane-box.md
covers the LabVIEW-free Ubuntu lane box (vm/ubuntu-lane/,
scripts/Invoke-LaneBox.py), and
docs/vm-provisioning/ubuntu-lane.md
covers the dry-run lane plan and lane teardown by VM UUID
(scripts/Invoke-Lane.py), and
docs/vm-provisioning/ubuntu-vagrant-clone.md
covers the neutral clone verifier (scripts/vagrant_clone/) and its clone
profiles.
Run-Pipeline.ps1 runs the native Windows LabVIEW build-and-report
pipeline directly on a self-hosted runner (there is no native Linux
equivalent in this repo -- Run-Pipeline.sh is a separate, unrelated
LabVIEWCLI MassCompile/VI Analyzer tool for NI's own example VIs, not a
SBOM pipeline; see docs/sbom-pipeline/procedure.md's Native pipeline
section). Run-WindowsSbom.ps1/Run-LinuxSbom.ps1 run the equivalent
containerized (Docker) pipelines on either platform. All three read
project configuration from either the PROJECTS_OVERRIDE environment
variable (the documented CI mechanism) or an explicit
-ConfigName/-VipmTomlPath local file — neither has a built-in
default pointing at any specific project; supply your own.
Dependency-Track reporting (full mode) requires an explicit
DEPENDENCY_TRACK_URL — there is no built-in default pointing at any
particular organization's server; a fork without one configured fails
closed with a clear error rather than silently trying to reach someone
else's private instance.
docker/linux-sbom/ and docker/windows-sbom/ are the container images'
own Dockerfiles and in-container pipeline scripts. See
docs/sbom-pipeline/procedure.md for
the full stage-by-stage procedure (both native and containerized paths),
including the full-mode fail-fast reporting preflight, the LVLIBP
release-dependency attestation-verification chain, and the deliberate
native-vs-container difference in whether Grype's own findings gate the
run.
An ephemeral, no-VIPM, GitHub-hosted-runner container (separate from the
SBOM pipeline's own containers above) that runs a pinned commit of
ni/labview-icon-editor's real LUnit test suite via native LabVIEWCLI,
proving this engine can execute genuine, unattended LabVIEW automated
tests without depending on persistent-VM GUI activation. Both Windows
(Phase 1) and Linux (Phase 2) are implemented and live-dispatch-proven on
both the fork and upstream VITechnologies/projectbuddy -- per REQ-033
the dual-platform gate is now Active. See
docs/lunit-gate/procedure.md and
docs/architecture/adr/ADR-0003-ephemeral-container-lunit-release-gate-revised.md
for the full procedure, the SHA-256 package-pinning provenance, and the
named-waiver policy.
Register-WindowsRunner.ps1 and Register-WslRunnerKeepAlive.ps1 plan or
install a self-hosted Windows Actions runner as a service (optionally under
a dedicated local account) and keep a WSL-hosted runner's distribution
running, respectively. Add-Project.ps1 validates and dispatches a
project-registration workflow call (add-project.yml-equivalent) against
the schema in pipeline-config.schema.json.
A first-class, project-agnostic mechanism for flagging a blocker that
automated iteration cannot resolve on its own, instead of leaving that
determination buried in an ad-hoc issue comment. New-HumanEscalation.ps1
creates (or, re-run with the same -EscalationId, idempotently updates) a
structured GitHub issue in any caller-supplied repo: context, why the agent
could not resolve it, options considered, a recommendation if one exists,
and related issue links, all under a consistent needs-human-decision
label (configurable via -Label). Get-HumanEscalations.ps1 queries one
or more repos for open issues carrying that label and prints a single
consolidated dashboard, so a human reviewing multiple forks/repos in this
ecosystem does not have to check each one individually.
This is deliberately a plain script, not a GitHub Actions workflow: the decision to escalate is a judgment call made during an agent's own analysis, not a routine CI step, so it is invoked directly (second consumption model above) at the point that judgment is made.
Where an escalation only notifies a human, the approval gate lets a human
authorize one privileged action. An automated requester submits a request
with scripts/Invoke-ApprovalGate.py submit. The request becomes one run
of approval-request.yml, whose gated job waits in a protected GitHub
environment for a human required reviewer. After an approval, the job
re-verifies it, then runs the action class's executor with the credential
that only that environment holds. Every decision is recorded on a
tracking issue. Forks and other repositories adopt the gate by
configuration: an action-class registry, one environment per class, and a
readiness check that names anything missing. See
docs/approval-gate/procedure.md and
ADR-0007.
Phase 1 ships the gate with only a no-op known-answer class, so it is not
yet live-proven.
Many repos this engine builds for start with Issues disabled, which forces
their technical issues into whichever consumer repo happened to drive the
work instead of their own tracker. Enable-RepoIssueTracking.ps1 enables
Issues on a given owner/repo (idempotent, never disables anything) and
seeds the needs-human-decision label so the repo is immediately ready for
both normal issues and New-HumanEscalation.ps1 escalations.
add-project.yml calls this automatically, as a best-effort step, for
every GitHub-hosted repo a newly-registered project's config references
(repoUrl, sourceCheckouts[].repository,
build.lvlibpDependencies[].repository) -- this needs a repo-scoped
ACTIONS_REPO_ADMIN_TOKEN secret (administration:write on the target
repos; distinct from ACTIONS_VARIABLES_TOKEN, which is Variables-scoped
only). A missing or insufficiently-scoped token only logs a warning; it
never fails project registration itself.
Projects outside the PROJECTS_OVERRIDE registration mechanism (e.g. a
standalone source repo being brought into a separate cross-repo build-spec
effort) have no formal onboarding workflow yet -- run
Enable-RepoIssueTracking.ps1 directly against such a repo as the first
step of bringing it in, until/unless that kind of effort gets its own
onboarding workflow.