Turn a small deployment specification into consistent Docker, Kubernetes and CI files. Review exactly what changes before handing the files to your deployment system. Restore a previous artifact bundle when a change must be reversed.
This release replaces the historical model driven shell execution interface with a deterministic local planner. It never runs deployment commands, reads cloud credentials or calls a model. Existing Python users must follow the migration guide.
| Capability | Behavior |
|---|---|
| Artifact generation | Dockerfile, Deployment, Service, NetworkPolicy and GitHub CI template |
| Policy validation | Rejects mutable image tags, unknown fields, privileged ports, excessive resources and any edited bundle |
| Preview | Reports changed files while keeping resource name and namespace fixed |
| Rollback planning | Returns the exact prior validated bundle only when the current digest matches |
| CLI and MCP | Seven actions share the same implementation and policy |
| MetaHarness | Repository maintainer profiles, sessions, field memory adapter and host integration |
| Autogenous | Evidence gates without automatic production promotion |
| CI delivery | Tests, dependency audit and downloadable artifact/benchmark evidence |
Requires Node 22 or 24. No API key is required.
npm ci
npm test
node src/cli.mjs status
node src/cli.mjs plan < examples/spec.json > plan.json
node src/cli.mjs validate < plan.json
npm run benchmarkThe example digests are deliberately fictional fixtures. Replace image with an existing, reviewed application image digest and baseImage with a verified compatible base digest. The Dockerfile expects an executable app/server; it does not compile your application. Registry availability, image signatures, vulnerability scanning and an actual cluster rollout are deployment qualification steps, not claims made by this planner.
Commands plan, validate, preview and rollback accept JSON on stdin, bounded to 64 KiB. Output is JSON. Extract the fixed artifacts map to a review directory using your trusted deployment tooling. See operations for request examples and runtime assumptions.
{"mcpServers":{"agentic-devops":{"command":"node","args":["/absolute/path/agentic-devops/src/mcp.mjs"]}}}Tools: devops_status, devops_plan, devops_validate, devops_preview, devops_rollback, devops_test, devops_benchmark. Each takes { "input": { ... } }. Policy resource: ruv://agentic-devops/policy.
Tests launched through the CLI or MCP require the local operator to set AGENTIC_DEVOPS_ALLOW_TESTS=1. The subprocess has a 30 second deadline, 64 KiB output limit, one concurrent execution and a stripped environment. MCP callers cannot choose commands, filesystem paths or environment variables. Stdio runs with the launching user's identity and is intended for trusted local hosts.
Kubernetes output uses a nonroot UID, dropped capabilities, read only root filesystem, no service account token, resource limits and deny all egress. Same namespace pods may reach the service. NetworkPolicy requires a supporting CNI. A local digest proves byte consistency, not authorship or deployment authorization.
Read the security review, architecture decision, benchmark methodology and migration guide. Historical UI screenshots and source are retained for provenance, not offered as a supported deployment path.
Use RuFlo for orchestration, MetaHarness for repeatable evaluations, Autogenous for evidence gates, Guardrail for application policies, Dynamo MCP for project scaffolding and Federated MCP for public federation reads. x.ruv.io messages are data, never deployment permission.
MIT license. Legacy vendored components retain their own license notices.
See the generated MetaHarness guide and Autogenous gate guide. Generated sessions and field memory require deployment owned identity and storage configuration; their presence does not mean a remote memory service is deployed.