gaik-toolkit agent plugin

A portable Agent Plugins v1 package containing the skills an AI coding agent needs to build document-understanding pipelines with the gaik Python toolkit.

The skills carry decisions and failure modes that are expensive to rediscover — which parser preserves a table's structure, why an extraction request is refused, and what a confidence score is actually worth — rather than restating the API reference, which the agent can read from the package itself.

Skills

Skill	Use it when
`parsing-documents`	Converting PDFs, scans, or Word files to text or markdown; tables come out wrong; parsing costs more than expected
`extracting-structured-data`	Pulling fields or line items into a schema; a request fails with a 400 or truncates; adding page/quote evidence; measuring accuracy

Each skill is self-contained: SKILL.md holds the workflow, and references/ holds detail loaded only when the task needs it.

Installing

This package follows the portable format, so any client that implements Agent Plugins v1 can load it from a directory. The two common cases:

Claude Code — add the repository as a plugin marketplace, then install:

/plugin marketplace add GAIK-project/gaik-toolkit

/plugin install gaik-toolkit@gaik-toolkit

Any other Agent Plugins v1 client — point it at this directory, or copy the directory into wherever that client keeps plugins. Everything the format defines lives at a fixed location, so nothing needs configuring.

No plugin support at all — copy skills/parsing-documents/ and skills/extracting-structured-data/ into the agent's skills directory. They are ordinary Agent Skills and do not depend on the plugin wrapper.

Layout

agent-plugin/
├── plugin.json          # manifest — the only required file
└── skills/
    ├── parsing-documents/
    │   ├── SKILL.md
    │   └── references/
    └── extracting-structured-data/
        ├── SKILL.md
        └── references/

The format is deliberately small. Two things are worth knowing if you extend this package:

Component locations are fixed and cannot be redirected. plugin.json names and describes the plugin; it cannot relocate skills/, and it cannot declare a component inline. A client finds skills by looking in skills/ for immediate child directories containing a SKILL.md — it does not search deeper, so a skill nested two levels down is invisible.

v1 standardises exactly two component types: skills, and MCP servers declared in an mcp.json at the plugin root. Agents, slash commands, hooks, and prompts are outside the portable contract. A client that wants them uses a reverse-domain extension namespace — either a key under extensions in plugin.json, or a top-level directory named exactly for that namespace, such as com.example.client/. Putting a client's own field at the manifest's top level instead does not conform: the schema is closed, and a conforming client reports and ignores the unknown field.

Both properties are what make the package portable — a client that implements none of the extensions still loads every skill here correctly.

Adding a skill

Create skills/<skill-name>/SKILL.md with name (matching the directory) and description in the frontmatter.
Write the description in the third person, and say when to use the skill, not only what it does — that text is the whole basis on which an agent decides to open it.
Put anything long in references/ and tell SKILL.md when to read it, so it costs nothing until it is needed.

No registration step: the skill is discovered because of where it sits.

Verifying the manifest

python -c "import json; json.load(open('agent-plugin/plugin.json'))"

The manifest requires only $schema and name. name must be 1–64 characters of lowercase letters, digits, hyphens, and periods, starting and ending alphanumeric, with no doubled hyphens or periods. Any schema violation other than an unknown top-level field is fatal — the client rejects the whole plugin and loads none of its skills.

License

MIT, same as the toolkit.