Context-aware skills for safer, more focused DevOps and cloud AI agents.
OpsContext is an early-stage public Agent Plugin. It packages small, task-specific skills instead of loading one large instruction set into every request.
- Software developers working with delivery and infrastructure workflows
- DevOps, SRE, platform, and cloud engineers
- Teams connecting AI agents to private engineering systems through MCP
Guides branch naming, Conventional Commits, commit and push review gates, and removal of AI attribution from Git metadata.
Activation and behavior cases live in evals/git-workflow.json.
Resolves organization-specific source control, documentation, tickets, service catalogs, delivery systems, artifact registries, cloud accounts, Kubernetes clusters, observability, incidents, security, cost systems, and MCP capabilities without hardcoding private values into the public plugin.
Projects can copy examples/ops-context.yaml to .ops-context.yaml and replace the example values with non-secret organization context. The ops-context JSON Schema provides validation and editor completion. See the profile format for boundaries and field guidance.
Validate a profile without installing a project dependency:
uvx check-jsonschema \
--schemafile https://raw.githubusercontent.com/ehsanmsb/ops-context/main/schemas/ops-context.schema.json \
.ops-context.yamlActual MCP endpoints, authentication, internal policies, and sensitive references belong in the agent host configuration or a private companion plugin maintained by the organization.
Activation and behavior cases live in evals/organization-context.json.
Creates, reviews, debugs, and safely operates Terraform and OpenTofu configuration. It preserves repository conventions, reviews plan impact, protects state, and requires explicit approval before applies, destroys, imports, state mutations, force unlocks, or production operations.
Activation and behavior cases live in evals/terraform-opentofu.json.
Protects live systems from access lockout, outages, data loss, unsafe retries, and irreversible changes. It adds explicit safeguards for remote Linux administration, SSH and firewalls, networking, storage, cloud control planes, Kubernetes, databases, credentials, deployments, and incident operations.
Activation and behavior cases live in evals/operational-safety.json.
Finds the canonical operational runbook, checks its ownership, scope, recency, preconditions, verification, and recovery path, then follows it one approved and verified step at a time. Organization-specific runbooks remain in their owning repositories or knowledge systems.
Teams can start with the runbook template and register non-secret discovery locations in .ops-context.yaml.
Activation and behavior cases live in evals/runbook-executor.json.
- Load only the workflow relevant to the current request.
- Prefer repository conventions and existing tools.
- Keep changes small, reviewable, and validated.
- Require explicit approval before commits, pushes, or destructive operations.
- Never expose secrets or invent live infrastructure state.
Version 0.6.0 adds a versioned schema and validation workflow for organization profiles. Additional infrastructure workflows will be added incrementally after their activation and output behavior can be evaluated.