version: "1.0.0" name: tidybot-skill-dev description: Build, test, and iterate robot skills for Tidybot. Use when (1) asked to make a new robot skill, (2) prototyping or iterating on skill code, (3) running or recording skill trials, (4) tracking success/failure stats, (5) deciding whether to chain existing skills or build new, (6) debugging skill execution, (7) needing a backend API, model, or service that doesn't exist yet.

Tidybot Skill Development

Before You Start

1. Read the SDK guide first: GET http://<ROBOT_IP>:8080/docs/guide/html — do not guess or improvise. See robot-sdk-ref skill.
2. Check existing skills: Fetch https://raw.githubusercontent.com/tidybot-skills/wishlist/main/catalog.json. Use or extend what exists. Local skills use tb- prefix.
3. Chain if possible. Use tidybot-bundle <skill-name> to combine skills. Estimate success = product of individual rates. Chaining tested skills is always preferable to untested new code.
4. Build new only as last resort. Research first (internet, papers), then check services (see active-services skill).

Naming Convention

• Robot skills use the tb- prefix: tb-pick-up-object, tb-center-object
• In-development skills use the dev-tb- prefix: dev-tb-my-skill
• Management/reference skills use the tidybot- prefix (you won't create these)

All new robot skills must start with tb- (or dev-tb- while in development).

Skill Structure (OpenClaw format)

dev-tb-my-skill/          # dev-tb- prefix while developing
├── SKILL.md              # Frontmatter (name, description) + instructions
└── scripts/
    ├── main.py           # Skill code (submitted to robot)
    └── deps.txt          # Dependencies (one tb-skill-name per line)

Developing

1. Connect to the robot API using SDK guide patterns. Test sensor reads first.
2. Prototype iteratively — use SDK methods (not invented ones), test on hardware, use rewind on failure.
3. Save with dev-tb- prefix (e.g. dev-tb-my-skill/) and OpenClaw skill structure.

Testing & Trials

1. Run the skill multiple times in varied conditions.
2. Record success/failure per trial. Any code change resets stats — re-test from scratch.
3. Debug with print() + poll /code/status?stdout_offset=N&stderr_offset=N. Prefer recorded frames over live camera.
4. Max 10 trials/day per skill during heartbeat practice.
5. Use rewind freely — it's your safety net.

Publishing Readiness

A skill is ready when it has 70%+ success rate over 10+ unsupervised trials on the final code version. For publishing steps, see the tidybot-skill-publish skill.

Code Pattern Extraction

When the same logic appears in 2+ skills, extract into a shared skill. Add it to scripts/deps.txt in dependent skills. Abstraction emerges from use, not upfront design.

Reference

• Skills catalog: https://raw.githubusercontent.com/tidybot-skills/wishlist/main/catalog.json
• Skills wishlist: https://raw.githubusercontent.com/tidybot-skills/wishlist/main/wishlist.json