Oracle Skills: installable, source-backed guidance for Oracle Database, OCI and GraalVM

The routing problem Oracle Skills is trying to solve

Oracle guidance is spread across product documentation, sample repositories and blog posts. A developer asking an AI agent about an OKE cluster or a GraalVM native image gets an answer assembled from whatever the model already absorbed, with no citation and no version boundary. Oracle Skills takes a different position: it treats guidance as an installable artifact with a declared source, organized by Oracle domain.

The intended audience is narrow and worth stating plainly. It is for developers, administrators and AI agents working inside an agent runtime that supports skill packages. If you are looking for a general Oracle tutorial site, this is not that. The README frames the goal as giving developers and agents a single place to find source-backed Oracle guidance across Oracle Database, Oracle Cloud Infrastructure, GraalVM, Oracle Fusion and Oracle APEX. The source list at the end of the README points at Oracle documentation for OKE and IoT, an Oracle samples repository for IoT, and the GraalVM native image reference manual. Those links are the backing for the content.

How the domain layout and SKILL.md files work

The repository is organized as a set of top-level domain directories: db, oci, graal, apex and fusion. Each domain carries its own SKILL.md, and the README describes that file as the table of contents for the domain. For a populated domain, content is split into category directories under the domain root. The db domain, for example, has admin, agent, appdev, architecture, containers, design, devops, features, frameworks, migrations, monitoring, ords, performance, plsql, security, sql-dev, sqlcl and vecdb.

The oci domain is structured differently because its subject matter is broader. It contains enterprise-ai with subdirectories for models, agent-workflows, governance, data, cost and integrations; functions with oci-functions-deploy and oci-functions-troubleshoot; iot-platform with agents, references, scripts, templates and tests; and oke with cluster-design.md, troubleshooting.md, gva-node-pools.md and multus-multihome.md alongside its own skills, scripts, agents, shared, examples and tests directories.

The README prescribes the shape of a domain SKILL.md. It should normally include sections named How to Use This Domain, Directory Structure, Category Routing, Key Starting Points and Common Multi-Step Flows. That is a routing contract: an agent reads the table of contents, picks a category, then reads the specific file. Stub domains are told to keep SKILL.md minimal and point back to the README and the authoring guide.

There is also a version rule. Any skill that documents version-specific behavior must carry a section named Oracle Version Notes (19c vs 26ai), with 19c as the baseline compatibility target unless stated otherwise, and newer-release features called out with 19c-compatible alternatives where practical. That is the most concrete editorial constraint in the repository, and it is the part most likely to be violated as contributions arrive.

Installing a domain skill and using it for the first time

The README gives one install path through the skills CLI. You append the root-level domain directory to the repository name, so the command names the domain you want rather than the whole collection.

npx skills add oracle/skills/db
npx skills add oracle/skills/oci
npx skills add oracle/skills/graal

Run the first line if your task is Oracle Database work. The README shows the same pattern for oci and graal, and the trailing ellipsis in its example implies other domains follow the same form. What you should see after the command completes is the db domain skill available to your agent.

If you use Claude Code, the repository ships a second install path as a plugin marketplace, defined in .claude-plugin/marketplace.json. Each domain folder (apex, db, fusion, graal, oci) is published as its own plugin. Register the marketplace first, then install the domains you need.

/plugin marketplace add oracle/skills
/plugin install db@oracle-skills
/plugin install graal@oracle-skills

If you already cloned the repository, the README says to point the marketplace at the local path instead of the repository name.

/plugin marketplace add ./

Enabled plugins are tracked in .claude/settings.json under enabledPlugins, and the README says you can browse and toggle installed plugins at any time with /plugin. That settings file is the thing to inspect when you want to know what is actually active in a project.

The README's own starting sequence is three steps: pick the domain closest to your task, install that domain skill, and add other domain skills only when needed. That last instruction is the useful one. Installing every domain up front works against the routing design, because the point of the SKILL.md table of contents is to narrow the agent's attention to one area.

Where the collection is thin

Two of the five domains are placeholders. The README states that fusion is the root for future Oracle Fusion skills and apex is the root for future Oracle APEX skills, and the repository layout confirms it: each contains only a SKILL.md. Anyone arriving for APEX or Fusion guidance will find a stub and a pointer back to the README. That is an honest structure, but it is not coverage.

The db domain is described as active and lists the most categories, yet the README does not state how many skills each category holds or how current they are. The oci domain names specific skills (OCI Functions deployment and troubleshooting, OKE cluster design and troubleshooting, IoT Platform digital twin workflows, and Enterprise AI guidance covering OCI Generative AI, agents, RAG, governance, model endpoints, Autonomous Database, APEX and integrations), but that is a list of topics, not a count of finished files.

The version standard is the other soft spot. Requiring a section titled Oracle Version Notes (19c vs 26ai) only works if contributors follow it, and the README does not describe a validation step that enforces the naming. Treat the standard as a convention to check per skill rather than a guarantee.

Finally, the README documents installation and nothing else. There is no documented update path, no uninstall command, and no stated compatibility matrix for which agent versions support the skill format. If you need lifecycle management, you are writing it yourself.

Oracle Skills compared with a general-purpose agent skill collection

The obvious alternative is a broad, vendor-neutral skill or prompt library that covers many stacks. The difference is in the routing layer. A general collection typically exposes one large index and expects the model to find the relevant passage. Oracle Skills splits by Oracle domain first, then by category inside the domain, and requires each domain's SKILL.md to carry Category Routing and Key Starting Points sections. The agent narrows twice before it reads anything substantive.

The second difference is the source discipline. The README's source list is Oracle documentation and Oracle sample repositories plus the GraalVM reference manual, and the version standard forces a 19c baseline with explicit callouts for newer releases. A general-purpose collection has no reason to make that commitment for Oracle products.

The trade-off runs the other way too. A general collection is useful the moment you install it, across whatever you are working on. Oracle Skills is only useful for Oracle work, and only for the domains that are actually populated. If your work touches Oracle occasionally and everything else constantly, a domain-specific collection adds a routing layer you will rarely use. It also assumes your agent supports the skill format at all, which the README does not attempt to establish for any runtime beyond Claude Code.

Licence and maintenance cost

The repository is licensed under UPL-1.0, the Universal Permissive License. It is a permissive licence, and the LICENSE.txt file at the repository root is the authoritative text. This is not legal advice; read LICENSE.txt and your own organisation's policy before redistributing or modifying the skills.

Maintenance is the harder question. The last push to the default branch was on 2026-09-10, and the repository is not archived. There are no retrieved releases, so there is no versioned artifact to pin against. That combination matters for adoption: you are tracking a branch, not a release. If a skill changes in a way that affects your workflow, nothing in the README describes how you would detect that or roll back.

Upgrade cost therefore falls mostly on the consumer. The install commands pull the domain as it exists at the time you run them, and the README does not document a version flag or a lockfile. For a team, the practical implication is that the .claude/settings.json enabledPlugins entry is the record of what is installed, and keeping that file under review is the only documented mechanism for seeing what a project has enabled.

Editorial conclusion

Adopt Oracle Skills if you already work inside an agent that supports skill packages and you want Oracle Database or OCI guidance routed by domain instead of pasted into a prompt. Do not adopt it expecting a complete Oracle APEX or Fusion reference: the README describes both directories as roots for future skills, and the repository layout shows only a SKILL.md inside each. Before rollout, verify three things: that the domain you need is populated rather than a stub, that your agent reads the plugin marketplace at .claude-plugin/marketplace.json, and what your own process is for updating or removing an installed skill, because the README documents installation only.

Frequently asked questions

Official sources

1. Issues
2. License: UPL-1.0
3. oracle/skills on GitHub
4. README