foundry-ops
foundry-ops covers everything between a merged commit and a running system: the pipeline, the
image, the cluster, the infrastructure code, the deployment target and the release train.
Three of its six agents declare isolation: worktree, because they write files while other agents
may also be writing files.
Install
Section titled “Install”/plugin install foundry-ops@foundryRequires foundry-core, which is installed automatically as a dependency.
When to install it
Section titled “When to install it”- The repository has no CI, or PR feedback is slow enough that people stop reading it.
- The workflow uses long-lived cloud secrets or unpinned third-party actions.
- Images are large, slow to build, or a scanner reports vulnerabilities and nobody knows which ones matter.
- Pods are OOMKilled or CPU-throttled, or probes cause restart storms and 502s during deploys.
- Infrastructure code has grown by copy-paste across environments and drifts silently.
When not to use it
Section titled “When not to use it”- Do not install it to “get Kubernetes”.
cloud-architectexists partly to argue you out of Kubernetes when a PaaS target is sufficient; installing this plugin does not commit you to a cluster. - It does not decide application architecture — that is
foundry-dev. - It does not set SLOs or design alerting.
sre-plannerandobservability-engineerinfoundry-qualitydo.
Agents
Section titled “Agents”| Agent | What it does | Model | Effort | Isolation |
|---|---|---|---|---|
pipeline-engineer |
Designs, hardens and speeds up GitHub Actions pipelines: caching, job graph, pinned actions, OIDC instead of long-lived secrets. | sonnet |
medium |
worktree |
container-engineer |
Writes and hardens container images: multi-stage builds, size and build time, scanner findings triaged by what is actually reachable. | sonnet |
medium |
worktree |
kubernetes-engineer |
Workload design and rollout strategy: sizing requests and limits, diagnosing OOMKills and CPU throttling, probes that do not cause restart storms. | opus |
high |
worktree |
iac-engineer |
Terraform and OpenTofu at scale: module boundaries, remote state and locking, environment separation without copy-paste, drift detection, plan review discipline. | opus |
high |
worktree |
cloud-architect |
Chooses the deployment target and its boundaries across AWS, Azure, GCP and PaaS (Vercel, Netlify, Fly.io, Render, Railway). | opus |
high |
— |
release-engineer |
Runs release trains: SemVer decisions, conventional-commit changelogs, tagging, promoting one artefact across environments by digest, feature flags to decouple deploy from release. | sonnet |
medium |
— |
Skills
Section titled “Skills”| Skill | When it fires |
|---|---|
scaffold-pipeline |
A repository needs a complete hardened GitHub Actions pipeline for the project type detected in it — Node/Angular, Maven or Gradle, or a containerised service. |
containerise |
A hardened multi-stage Dockerfile plus a compose file for local development, matched to the stack detected in the repository. |
deploy-strategy |
Choosing between rolling update, blue-green and canary, sizing its parameters from measurements, and writing the rollback procedure as a runbook. |
provision-environment |
Creating or extending Terraform/OpenTofu for a new environment, with remote state and locking, secrets handling and least-privilege identity for CI and workloads. |
release |
Running a release end to end — derive the version from the commit range and an API diff, generate the changelog, tag, publish and promote artefacts by digest, write release notes. |
All five accept --dry-run, --plan-only or an equivalent, so the generated output can be
inspected before anything is written or promoted.
Output contracts
Section titled “Output contracts”| Agent | Input | Output |
|---|---|---|
pipeline-engineer |
requirement.v1 |
plan.v1 |
container-engineer |
requirement.v1 |
review.v1 |
kubernetes-engineer |
plan.v1 |
adr.v1 |
iac-engineer |
plan.v1 |
risk.v1 |
cloud-architect |
requirement.v1 |
adr.v1 |
release-engineer |
plan.v1 |
handoff.v1 |
deploy-strategy also writes a rollback runbook to .foundry/runbooks/, which means the next
incident starts from a procedure rather than from memory.
Interaction with the guard hooks
Section titled “Interaction with the guard hooks”foundry-core protects .github/workflows/** and **/*.lock by default: a write there triggers
a PreToolUse escalation asking you to confirm. That is intentional — scaffold-pipeline changing
your CI should be a decision, not a side effect. See Hooks for how
to adjust protectedPaths.
Limits
Section titled “Limits”- Pipeline scaffolding targets GitHub Actions. Other CI systems are not generated.
- Infrastructure code targets Terraform and OpenTofu. CloudFormation, Pulumi and CDK are not.
- These agents write configuration; they do not hold your cloud credentials and do not apply
anything.
terraform apply,kubectl applyanddocker pushremain yours to run.