SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsawesome-agent-skills

awesome-agent-skills

Agents & models

View repository

Updated 10-15 19:53

About this skill

awesome-agent-skills

A routing-first front door for the Awesome Agent Skills collection from Shubhamsaboo/awesome-llm-apps — 19 expert-persona skills plus a self-improving optimizer, consolidated into one jeo-skill with 6 reference pipelines. Each persona ships its own framework and output format; this skill classifies the request, picks the right pipeline and persona, then executes with that persona's discipline.

One skill, many experts. Instead of 20 loose skill folders competing for trigger matches, this front door routes deterministically: identify the work type → pick the pipeline → load the persona's framework → produce the persona's output packet.

Read the matching pipeline reference before executing:

Plugin Installation

# This routing skill via jeo-skills (verified path)
npx skills add https://github.com/akillness/jeo-skills --skill awesome-agent-skills

# Global install for one or more agents
npx skills add -g https://github.com/akillness/jeo-skills --skill awesome-agent-skills -a claude-code -a codex -y

# Upstream full collection (all 20 folders, upstream-documented command)
npx skills add shubhamsaboo/awesome-agent-skills

The upstream skills follow the open Agent Skills specification — plain SKILL.md files with YAML frontmatter, usable in Claude Desktop, Cursor/VSCode, custom agents, and frameworks like LangChain or AutoGen.

When to use this skill

  • The user asks for work that maps to an expert persona (review my code, debug this, research X, write docs, draft an email, plan a sprint, make a decision matrix, design a user flow, analyze this dataset)
  • You want a consistent framework + output format for a common knowledge-work task
  • The user wants to optimize/improve an existing agent skill (→ self-improving pipeline)

When not to use this skill

  • The task is a multi-agent orchestration run (build/verify loop) → use omc / omx / ohmg
  • The task is a full academic research-to-publication pipeline with citation gates → use academic-research
  • The task is spec-first execution with frozen acceptance criteria → use ooo / spec-stack
  • The task is token-efficient code discovery in a repo → use semble
  • The persona overlaps a deeper jeo-skill (see Route-out map) → prefer the dedicated skill

Required intake packet

Before routing, identify:

  1. Pipeline — coding · research · writing · planning · data · self-improving
  2. Persona — the specific expert within that pipeline (routing table below)
  3. Input — the code, claim, topic, document, dataset, or skill folder to work on
  4. Output target — the persona's default format, or a user-specified one (PR comment, README, abstract, decision matrix, chart spec, …)

Pipeline & Persona Routing Table

What the user says Pipeline Persona
"review this code", "PR review", "security audit", "find vulnerabilities" coding code-reviewer
"debug this", "why is this broken", "trace this stack", "not working" coding debugger
"write/optimize Python", "PEP 8", "type hints", "clean Python" coding python-expert
"build a web app", "React/Next.js", "Node/Express API", "set up the database" coding fullstack-developer
"research X", "synthesize sources", "multi-source summary with citations" research deep-research
"fact-check", "verify this claim", "is this true", "misinformation" research fact-checker
"literature review", "analyze this paper", "academic writing", "format citations" research academic-researcher
"write docs", "README", "API reference", "tutorial", "user guide" writing technical-writer
"blog post", "social media", "marketing copy", "headline" writing content-creator
"proofread", "edit", "improve clarity", "fix grammar", "tighten this" writing editor
"draft an email", "reply to this", "decline politely", "follow-up" writing email-drafter
"meeting notes", "minutes", "summarize this discussion", "action items" writing meeting-notes
"plan this project", "work breakdown", "roadmap", "milestones", "estimate timeline" planning project-planner
"plan a sprint", "story points", "sprint goal", "capacity" planning sprint-planner
"strategy", "business direction", "competitive analysis", "long-term plan" planning strategy-advisor
"help me decide", "pros/cons", "decision matrix", "weigh trade-offs", "SWOT/ICE" planning decision-helper
"wireframe", "user flow", "persona", "usability review", "UX copy", "design system" planning ux-designer
"analyze this data", "SQL query", "pandas", "statistics", "explore this dataset" data data-analyst
"what chart should I use", "visualize this", "dashboard", "chart selection" data visualization-expert
"improve my skill", "optimize this SKILL.md", "make my skill pass its evals" self-improving ADK optimizer

Instructions

Step 1: Pick the pipeline and persona

Classify the request against the routing table. State the chosen pipeline → persona explicitly before producing output (e.g. "coding → code-reviewer"). If two personas fit, prefer the one whose default output format matches what the user asked for.

Step 2: Load the persona's framework

Read the matching reference file and adopt that persona's process and constraints:

coding          → references/coding-pipeline.md
research        → references/research-pipeline.md
writing         → references/writing-pipeline.md
planning        → references/planning-pipeline.md
data            → references/data-pipeline.md
self-improving  → references/self-improving-pipeline.md

Each reference contains, per persona: when to apply, the step-by-step framework, key checklists/heuristics, and the exact output format.

Step 3: Execute with that persona's discipline

  • code-reviewer / python-expert / ux-designer: follow priority order (Security → Performance → Correctness → Maintainability for code; Discover → Define → Ideate → Prototype → Handoff for UX). Never invent vulnerabilities or rules; flag only what you can justify.
  • fact-checker / academic-researcher / deep-research: every claim needs evidence; rate claims on the verification scale; never fabricate citations or sources — say "unverified" when you cannot confirm.
  • debugger: form hypotheses, test them, find the root cause before proposing a fix.
  • writing / planning / data personas: produce the persona's structured output template, not a freeform answer.

Step 4: Return the persona's output packet

Use the output format defined for that persona in its reference file (PR-style review, verdict + sources, structured doc, decision matrix, sprint plan, chart recommendation, …). Close with a one-line Next step pointing to a follow-on persona when relevant (e.g. code-reviewer → debugger; deep-research → fact-checker; project-planner → sprint-planner).

Integrity principles

  • No fabrication: fact-checker, academic-researcher, and deep-research never invent sources or citations. Unverifiable → mark "unverified".
  • Justify findings: code-reviewer and ux-designer flag only issues they can explain with a concrete reason and fix.
  • Persona fidelity: stay in the chosen persona's framework and output format; do not blend two personas in one answer unless the user asks for a handoff.
  • Human-in-the-loop: for high-stakes outputs (strategy, decisions, security findings), present options and trade-offs rather than asserting a single answer as final.

Route-out map

If the user needs… Route to
Full research-to-publication pipeline with citation gates academic-research
Multi-agent build/verify orchestration omc / omx / ohmg
Spec-first execution with frozen acceptance criteria ooo / spec-stack
Token-efficient code search across a repo semble
Behavior-preserving refactor of existing code code-refactoring
Editable diagrams / charts as artifacts drawio / mermaid / slides-grab
Long-form marketing campaign automation marketing-automation
Persistent knowledge capture / wiki llm-wiki / okf / obsidian