# Agent integration

The preferred image workflow is **agent vision → concise product evidence →
identify API → candidate verification → raw pricing**. Use the image understanding
already available in Muse or the calling agent. OCR is optional, not an API dependency or a required connector step.

## Reusable skill

[Read or download the TCG Identify skill](./skills/tcg-identify/SKILL.md).
It includes the image extraction rules, request example, ambiguity handling,
pricing boundaries, and quota behavior. It is a standard SKILL.md with name and
description frontmatter; it is not automatically installed by visiting this page.

For agents with skill discovery, copy the `tcg-identify` folder into the host's
supported skill location. For hosts using connector/tool instructions, adapt the
same workflow to their supported format and configure the API base URL. Do not
assume Muse accepts a particular skill upload format; confirm its submission
contract separately. The agent needs its own image-reading capability and an HTTP
tool or connector capable of calling the endpoints.

## The minimal extraction contract

Send one blob per item. Include the actual printed title, exact legible collector
number, game/set/language, and visible variant clues. `shape:"ocr"` also covers
text read by an image-capable model. Use `hints` only for reliable facts. Keep
rules-text character names, artist names, and seller overlays out of the title.
Ask for a clearer photo when decisive characters are unreadable; do not fabricate
them. Verify returned candidates against the original image before quoting value.

API schemas: `/openapi.json`. All docs: `/llms.txt`.

## Known unsupported releases

When `source` is `unsupported` and `reason` is `unsupported_release`, explain `detail` and follow `nextAction`. Do not retry with English substitutes or price another printing. Preserve the card language in the query and `language` field. See https://stratatcg.com/connector/docs/coverage.md .