List of skills developed by aboneto for the community.
npx skills add aboneto/skills --skill <skill_name>Example:
npx skills add aboneto/skills --skill consult-claudeThe idea behind this skill is to use the claude CLI as an "expert colleague" you can consult when facing something complex. It's a thinking tool, not a delegation tool: your main LLM remains responsible for the solution, but you gain external perspective. Very useful when you're using Opencode and want to escalate to a more capable model for a difficult problem without changing the entire session or costing more tokens than necessary. It's a thinking tool, not delegation: you remain responsible for the solution, but you gain external perspective.
Use when:
- You've made 3 or more attempts without resolving a bug and suspect you're stuck in an incorrect mental model.
- You need to make a design decision where there are multiple defensible paths and the choice isn't obvious.
- You're about to implement a non-trivial algorithm (complex parsing, concurrency, non-standard data structures) and want to validate the approach before writing a lot of code.
- The user has asked you for something where a mistake has a high cost (data migration, security-related code, large refactor).
- You have a solution that "works" but you're not sure if it's the right one.
When NOT to use (important to avoid burning through the user's subscription):
- Simple tasks: file reads, trivial edits, questions with a clear answer.
- When you already have high confidence in your approach.
- For repetitive tasks where you've already validated the pattern earlier in the same session.
Tests: tests/consult-claude-workspace
A compact reference for working productively with Jekyll. Covers Liquid templates, kramdown Markdown, _config.yml, collections, themes, plugins, GitHub Pages/Actions deployment, and custom plugins/filters.
Use when:
- Setting up a new Jekyll site or migrating an existing one.
- Working with Liquid templates, kramdown Markdown, or YAML front matter.
- Configuring
_config.yml, collections, data files, or permalinks. - Installing or troubleshooting plugins and themes.
- Deploying to GitHub Pages, GitHub Actions, or Netlify.
- Debugging build errors, future-dated posts, or Liquid syntax errors.
- Adding responsive images, comments, or custom plugins/filters.
Tests: evals/jekyll-workspace
Benchmark results (4 iterations):
| Iteration | Model | With Skill | Without Skill | Delta |
|---|---|---|---|---|
| 1 | Claude Opus 4.7 | 100% | 80.4% | +19.6% |
| 2 | DeepSeek V4 Flash Free | 100% | 80.4% | +19.6% |
| 3 | Gemini 2.5 Flash Lite | 77.8% | 46.5% | +31.3% |
| 4 | MiMo v2.5 Pro | 100% | 69.5% | +30.5% |
The skill consistently improves output quality across all tested models and evaluations (4 evals covering GitHub Pages deployment, Liquid/kramdown features, docs site scaffolding, and responsive images).
Complete agent skill for the AG-UI Ruby SDK (ag-ui-protocol gem). Use when working with Ruby/Rails applications that implement the Agent-User Interaction Protocol — streaming text responses, tool calls, lifecycle events, state management, ActionController::Live, with_stream pattern, EventEncoder, RunAgentInput, or any event type from the SDK.
Use when:
- Working with Ruby applications using
ag-ui-protocol. - Implementing Rails applications with AG-UI streaming endpoints.
- Setting up event streaming, lifecycle management, or serialization for agent-user communication in Ruby.
Tests: evals/ag-ui-ruby-sdk-workspace
Benchmark results (6 iterations):
| Iteration | Model | With Skill | Without Skill | Delta |
|---|---|---|---|---|
| 1 | Gemini 3.5 Flash | 100% | 27.78% | +72.22% |
| 2 | DeepSeek V4 Flash Free | 92% | 11% | +81% |
| 3 | MiniMax M2.7 | 96.4% | 32.1% | +64.3% |
| 4 | MiniMax M2.7 | 100% | 17% | +83% |
| 5 | Kimi K2.6 | 100% | 58% | +42% |
| 6 | DeepSeek V4 Flash Free | 100% | 15% | +85% |
The skill guarantees correct type usage, event sequences, and reference document consultation, whereas runs without the skill fail to use the correct AgUiProtocol namespace and guess SDK methods/attributes incorrectly.
- Iteration 4 fix: Iteration 3 had incorrect
ActionController::Liveinclude and undefinedstreamvariable issue. Iteration 4 skill now correctly instructs to includeActionController::Liveand useresponse.streamproperly, achieving 100% pass rate. - Iteration 6: Expanded skill with reference docs (
0.2.0gem version) and patterns for reasoning events, interrupt/resume types (HITL), agent capabilities declaration, and multimodal content types. Added 5 new evals (reasoning-events-stream, reasoning-message-type, interrupt-resume-types, agent-capabilities-declaration, multimodal-content-types) for a total of 9 evals. Subagents ran in strict isolation:without_skillprompts contained zero SDK class names or module paths, andwith_skillsubagents read the skill files themselves to discover the correct API.with_skillscored 100% across all 9 evals (45/45 assertions).