Build plugins
Prepared, unreleased authoring contract. This public page documents a reviewed future authoring workflow. Future authoring commands described here are not available in current releases; they do not claim that
plugin-kit-ai@2is available. Existing installer examples are identified separately. Public activation remains pending.
| Your job | Journey | Result |
|---|---|---|
| Install someone else's package | Use plugins | Managed client installation and activation instructions. |
| Write a new portable package | Continue below | Root plugin.json, Skills and/or MCP configuration. |
| Keep an existing YAML/runtime project working | Historical v1 | Version-pinned 1.2.4 context; no v2 migration. |
This journey describes the prepared future authoring contract at source revision 070663efb27f69ecae8609e6b839f86f843efbb0. It is not installation advice for a published v2 package. Use the accepted release supplied by the release owner before running these examples; an older executable may have different semantics.
Choose the smallest useful package
| Need | Template | Inputs beyond destination and identity |
|---|---|---|
| Instructions an agent can follow | Standalone Skill | --template skill; optional explicit Skill name. |
| Connect to an existing service | Remote MCP | --template mcp-remote --url with an explicit endpoint. |
| Provide a local MCP process | Stdio MCP | --template mcp-stdio --runtime node. |
| Instructions plus tools | Hybrid | --template hybrid --mcp-template plus the selected MCP inputs. |
A standalone Skill here means a standard package whose portable component is a Skill. It still has root plugin.json; it is not an external Skills installer. Add more instructions later with extra Skills.
Two equivalent authoring entrypoints
Choose one spelling and use it consistently:
| Shared authoring job | Installer executable | Authoring executable |
|---|---|---|
| Create a package | agentplugins author init | plugin-kit-ai init |
| Validate | agentplugins author validate | plugin-kit-ai validate |
| Inspect | agentplugins author inspect | plugin-kit-ai inspect |
| Static test | agentplugins author test | plugin-kit-ai test |
| Static client compatibility | agentplugins author compat | plugin-kit-ai compat |
| Project doctor | agentplugins author doctor | plugin-kit-ai doctor |
| Engine capabilities | agentplugins author capabilities | plugin-kit-ai capabilities |
| Add or validate Skills | agentplugins author skills | plugin-kit-ai skills |
The table names jobs, not complete invocations. The tutorials supply explicit paths and template inputs. Both entrypoints use the same prepared authoring engine; neither is a second YAML implementation.
Follow the authoring loop
- Choose a template and an absent destination under an existing parent.
- Review and edit the resulting package files.
- Run validate, inspect, and static test against the exact root.
- Evaluate explicit target compatibility and inspect project doctor evidence.
- Send the package and its limits to the installer planner.
Creating files is local mutation. Validation, inspection, static test, and compatibility do not install into clients or launch package code. A passing static report is useful evidence, but it does not establish runtime success.
What this journey does not expose
Runtime execution, dev loops, dependency bootstrap, import, normalization, project migration, client generation, export/bundle, and publication are deferred from the v2 journey. Their retained v1 source and documentation remain useful; they are not runnable v2 features. Hooks remain client-specific extensions.
Do not add legacy --platform, --strict, --typescript, --output, or --force flags to these examples. Select destinations positionally and read command boundaries before adapting automation.