This repository hosts public agent skills maintained by the ToyWorks Team.
Agent Skills are folders of instructions, scripts, and resources that agents can discover and use to do things more accurately and efficiently. They work across any AI agent that supports the open Agent Skills standard.
| Skill | Description |
|---|---|
| t3-hardware-scoring | The MantaBase T3 Hardware Audit System classifies a hardware product as Tool, Toy, or Trash. It brand-blinds the source material, scores it with three independent auditors working from fixed rubrics and verbatim evidence, revalidates the high-risk findings with an Eagle Eye pass, and applies a safety veto in the final synthesis. |
Use npx skills to install skills directly:
# Install all skills
npx skills add toyworks/agent-skills
# Install specific skills
npx skills add toyworks/agent-skills --skill t3-hardware-scoring
# List available skills
npx skills add toyworks/agent-skills --listInstall via Claude Code's plugin system. The marketplace is defined in
.claude-plugin/marketplace.json and is named
toyworks-agent-skills:
/plugin marketplace add ToyWorks/agent-skills/plugin install t3-hardware-scoring@toyworks-agent-skillsClaude Code plugins are also supported in Factory's Droid.
Agent skills can also be installed by using the below commands from Playbooks or Context7:
# Playbooks
npx playbooks add skill toyworks/agent-skills
# Context7
npx ctx7 skills install /toyworks/agent-skillsCreate skills/<skill-name>/SKILL.md with YAML frontmatter (name,
description), add any supporting references/ and scripts/, then add a row
to the table above.
The description is what an agent matches against when deciding whether to load
the skill, so write it as a trigger, not a summary: name the user phrasings and
artifacts that should invoke it.
Skills that ship scripts should ship tests for them. t3-hardware-scoring is
the reference:
skills/t3-hardware-scoring/tests/run_tests.shIts tests/fixtures/ holds a complete worked example — a fictional product's
brand-blinded source text, the three auditor reports, and the classification
they synthesize to. It serves as both a regression test and the reference for
what correct output looks like.
CI runs each skill's test suite plus a link check on every push and pull request. See .github/workflows/test.yml.
We welcome contributions to improve existing skills or add new skills. Please
submit a pull request, and make sure run_tests.sh passes.
MIT