---
name: robotico
description: Research and safely update humanoid-robotics intelligence on Robotico (companies, robots and products, industry stats, articles and signals, and the event timeline) through the Robotico MCP server. Use when the user asks to look up or compare humanoid robots or companies, enrich a robot or company profile, attach source evidence, report or de-duplicate a timeline event, or draft a Robotico article or market signal. Requires the Robotico MCP connected (add it by URL and approve access in your browser).
---

# Robotico

Robotico is the humanoid-robotics intelligence platform. This skill drives its MCP tools to read and safely write company, robot and product, industry, article, and timeline data.

## Prerequisite: connect the MCP first

This skill only works when the **Robotico MCP server is connected**.

- Set it up from the setup guide at https://robotico.market/agent: add the server by URL and approve access in your browser (headless clients use a device code). There is no token to copy.
- Reads and writes both require a connected, verified account. Writes are scoped to what the account's role allows.
- If Robotico tools are not available in the session, stop and ask the user to connect the MCP before continuing.

## Know your access before you write

Robotico enforces roles server-side. Match the task to the connected account's role:

- **Any connected account:** all `list_*`, `get_*`, `search_*`, evidence, industry, and timeline reads.
- **Author (contributor):** article and signal tools, plus timeline reporting (candidates, sources, entity links).
- **Company rep:** company profile and robot writes scoped to the rep's assigned company.
- **Admin and reviewer:** company creation, review-queue approve/reject/publish, and audit logs.

Never assume a write will succeed. If a tool returns an access error, report it plainly instead of retrying with a different tool. The full tool-by-tier reference is on the `/agent` page and stays in sync with the live server.

## Golden rules

1. **Resolve IDs first.** Use `search_*` or `list_*` to get a canonical ID before any detail read or write.
2. **Evidence with every fact change.** Include source URLs and, where the tool supports it, quoted excerpts and screenshots.
3. **Review-first for risky changes.** Identity and interpretation changes go through candidate queues, not direct writes.
4. **De-duplicate before creating.** Always check for an existing record before creating a new one.
5. **Research is not publishing.** Creating a candidate does not publish it. Publishing is a separate reviewer action.

## Playbook: robot and product research

1. `search_robots` or `list_robots` to resolve the robot ID.
2. `get_robot` for the normal profile: specs, aliases, classification, availability, and variants.
3. Only when you need provenance, call `get_robot` with `include_evidence: true`, or call `robot.evidence`. Treat `evidence.fields` (dot-path keys such as `specifications.height`) as the canonical provenance map.
4. When a value is correct but unsupported, attach a citation with `robot.link_source_to_field` (field path plus source URL and a quoted excerpt or screenshot). This does not change the value.

## Playbook: robot and product updates

1. Use direct writes only for current, low-risk facts the token may edit: `update_robot`, `update_robot_profile`, `update_robot_specifications`, `add_robot_alias`, `upload_robot_cover_image`, `set_robot_cover_image`. Include sources or notes.
2. Stage risky changes as candidates instead of writing: rename, status, classification, category, About or description rewrite, and variant merge or split. Use `propose_robot_rename` or `create_robot_update_candidate`.
3. `list_robot_update_candidates` inspects the queue. Approving or rejecting a candidate records review state only (admin and reviewer). Applying an approved change is still a separate explicit update.
4. After writes, `robot.audit_log`, `company.audit_log`, or `entity.audit_log` explain what changed.

## Playbook: timeline event reporting

1. Before creating anything, call `events.find_duplicates` with the title plus any summary, company and robot IDs, event date, and source URL.
2. If a duplicate is likely, do not create another canonical event. Add sources or entities to the existing one, or keep the new item as a review candidate.
3. Otherwise use `events.create` (reporter workflow) or `create_event_candidate`, each with source evidence.
4. Every source needs a canonical http or https URL with a path or query, plus a JPEG screenshot of that exact page (`source_screenshot_base64` or `references[].screenshot_base64`).
5. Enrich with `events.add_source`, `events.link_company`, and `events.link_robot`. Reviewers use `events.mark_reviewed` or `publish_event_candidate` to publish an approved candidate.

## Playbook: articles and market signals

1. `list_my_articles` finds your drafts. Use `create_article` or `update_article` for long-form (`type: article`) or signals (`type: signal` with a `signal_category` such as `funding`, `product`, or `partnership`).
2. New articles save as drafts. Direct publishing (`draft: false`) requires a verified contributor or admin account.
3. `upload_image` returns a public URL for article imagery. Through an agent, keep uploads small because the base64 payload transits the model context.

## Role-aware notes

- **Company rep:** you can write only your assigned company and its robots, and only with no pending claim. Public profile fields are shared, broader metadata is admin-only.
- **Author and contributor:** you own the article, signal, and event-reporting surface. You cannot approve or publish reviews.
- **Admin and reviewer:** you hold company creation, the approve, reject, and publish review queues, and full audit access. Prefer candidate review over direct edits for anything another role staged.

Exact tool inputs and the access tier for each tool are documented on the Robotico `/agent` page.