Nine Skills, One Complete Agent Toolkit: A SkillHub Equipment Map

> In 2026, AI skills have evolved from "plugins" into "equipment." You're not installing one feature on an agent — you're giving it a complete survival toolkit. Memory, action, learning, review, search, writing — only when these skills combine does an agent move from "able to chat" to "able to get things done." The nine skills on SkillHub.cn roughly sketch out a modern AI agent's complete equipment library.

1. What It Is: A Map of 9 Skills

1.1 Self-Improving Agent: The Agent That Takes Notes

The core is simple: teach the agent to take notes. It creates three files — LEARNINGS.md (what it learned), ERRORS.md (mistakes made), and FEATURE_REQUESTS.md (what users want).

Key design:

• Automatic triggers: failed commands, user corrections, knowledge gaps, feature requests are logged automatically
• ID format: TYPE-YYYYMMDD-XXX, a unique ID for every entry
• Promotion mechanism: when a pattern appears 3+ times across 2+ tasks, it's promoted from temporary log to permanent memory (written into AGENTS.md / CLAUDE.md / SOUL.md / TOOLS.md)
• Cross-platform: supports OpenClaw, Claude Code, Codex CLI, GitHub Copilot

In essence, this is a metacognitive skill — the agent doesn't just do things, it supervises how it does them.

1.2 Summarize: One-Click Summarization

Details pending (search results did not return full technical specs). By naming, likely a text-summarization skill supporting long-document compression, multi-format input (PDF, web pages, chat logs), key-information extraction, and customizable length/style.

1.3 Github: A Cross-Platform Skill Manager

Not a GitHub operations skill but a skill manager, from Ray0907/skillhub:

• Cross-platform sync: Claude Code, Codex CLI, Gemini CLI
• Multi-source pulling: Git repos + local directories
• Namespace isolation: @scope/name format to avoid conflicts
• Automatic sync: configurable background intervals
• Platform filtering: skills can declare which platforms they support

In essence: it's the "package manager" for skills — npm for JavaScript, pip for Python, skillhub for AI skills.

1.4 Agent Browser: The Agent's Eyes and Hands

From Vercel Labs, 20.1k stars, Multi-Platform Verified.

Features:

• Web navigation, form filling, clicking
• Screenshots, data extraction
• Web app testing
• iOS simulator support
• Snapshots of interactive elements

Install: npx skills add vercel-labs/agent-browser --skill agent-browser

This is the base infrastructure for agents to interact with the internet. Without it, agents only handle text; with it, they can operate web pages, fetch live data, and run automation.

1.5 Skill Vetter: The Security Reviewer

Details pending. Likely a security-review skill: scanning skill code for malicious patterns, checking API calls and data-leak risks, verifying source trustworthiness, possibly via static analysis and rule matching.

1.6 Ontology: Knowledge Graphs

Details pending. Likely a knowledge-management skill: domain knowledge graphs, entity-relation extraction, knowledge reasoning and querying, possibly RDF/OWL standards.

1.7 Proactive Agent: Acting Without Being Asked

From Hal 9001 (@halthelobster), v3.1.0.

Core philosophy: "Act without being asked, survive context loss."

Full Agent Stack combo:

• Proactive Agent (autonomous action)
• Bulletproof Memory
• PARA Second Brain
• Agent Orchestration

Key features:

• Autonomous vs Prompted Crons
• WAL Protocol (Write-Ahead Log)
• Working Buffer Protocol
• Compaction Recovery Protocol
• Unified Search Protocol
• Self-Improvement Guardrails (ADL/VFM)

Slogan: "Every day, ask: How can I surprise my human with something amazing?"

This is a high-level agent framework. It doesn't solve specific problems; it solves "how does an agent keep running continuously."

1.8 Multi Search Engine: Aggregated Search

Details pending. Likely queries multiple search engines simultaneously (Google, Bing, DuckDuckGo, academic search), with deduplication, ranking, possible vertical search, and automatic best-result selection.

1.9 Humanizer: De-AI-ifying Text

From Aboudjem (Adam Boudjemaa), MIT License, zero dependencies.

Core capabilities:

• 43 AI writing pattern detections: flags clichés like "In today's rapidly evolving landscape"
• 5 voice styles: casual, professional, technical, warm, blunt
• 3 modes: detect (score), rewrite, edit
• AI-tell score: 0–100; 87 = "pure AI," 12 = "sounds human"
• Burstiness restoration: AI text has uniform sentence lengths; human text varies — Humanizer restores that burstiness

Install: a single Markdown file, one curl command. Supports 8 editors (Claude Code, Cursor, VS Code + Copilot, Codex CLI, Gemini CLI, Windsurf, Continue.dev, OpenClaw).

It's a stylistic scalpel — keeps the content, changes the expression, turning AI-ese into human-ese.

2. What It's For: Assembling a Full Capability Map

2.1 Three-Layer Architecture: Perception → Cognition → Action

• Perception layer: Agent Browser (see and operate the web), Multi Search Engine (search everything), Summarize (compress information)
• Cognition layer: Ontology (structured knowledge), Self-Improving Agent (learn, remember, evolve), Humanizer (adjust expression)
• Action layer: Proactive Agent (act on schedule), Github (manage the skill ecosystem), Skill Vetter (security review)

2.2 A Day-in-the-Life Workflow

A freelancer's day with an agent:

8:00 AM (Proactive Agent triggers): Agent Browser opens email/calendar; Summarize compresses overnight mail; Multi Search Engine scans the day's news.

9:00 AM (work starts): Self-Improving Agent loads yesterday's learnings; Humanizer polishes last night's AI draft; Ontology builds the project knowledge graph.

3:00 PM (code work): Github syncs skills to Claude Code; Skill Vetter audits newly installed skills; Agent Browser tests deployed web pages.

8:00 PM (wrap-up): Self-Improving Agent logs errors and learnings; Proactive Agent schedules tomorrow's crons.

2.3 Complementary Ecosystem: A Puzzle, Not a Competition

• Vercel (Agent Browser) — what the agent can see
• Hal 9001 (Proactive Agent) — when the agent acts
• Aboudjem (Humanizer) — how the agent speaks
• Ray0907 (Github) — what tools the agent uses
• pskoett (Self-Improving Agent) — how the agent learns

Nobody built the full stack. But together, they form a complete agent operating system.

3. How to Use It: Boundaries and Unresolved Questions

3.1 Who Vets the Skill Vetter?

A recursive problem: if the Vetter itself has flaws, the whole trust chain breaks. Possible fixes — multi-Vetter cross-validation (multi-signature style), community audits, sandboxed execution. Currently unclear; Skill Vetter is the most opaque of the nine.

3.2 The Self-Improving Agent's Promotion Trap

Recurrence count ≥ 3 triggers automatic promotion — but promotion only checks frequency, not correctness. A repeatedly-made error could be recorded, linked, and finally promoted into "best practice" — a wrong dogma. Mitigations: human confirmation before promotion, distinguishing "corrections" from "errors" (hard to automate), periodic manual audits of promoted memories.

3.3 Proactivity vs. Intrusiveness

"How can I surprise my human?" sounds romantic, but a 3 AM push notification may be a scare, not a surprise. Boundaries: time windows, priority filtering, user feedback loops. v3.1.0 distinguishes Autonomous vs Prompted Crons, but the boundary still depends on user configuration.

3.4 Head Concentration and Long-Tail Neglect

Agent Browser has 20.1k stars; Humanizer, Self-Improving Agent, and Proactive Agent have solid communities. Summarize, Skill Vetter, Ontology, and Multi Search Engine have little documentation. Possible interpretation: browser automation and humanized writing are刚需-level demands; learning/autonomy is for advanced users; security review, knowledge graphs, and aggregated search may be infrastructure — important but not installed by everyone, or already served by mature alternatives.

3.5 Memory System Conflicts

Three skills involve memory: Self-Improving Agent (Markdown promotion), Proactive Agent (Bulletproof Memory + PARA), Ontology (knowledge graph). If all three are installed, where is memory stored? Are formats compatible? Could they overwrite each other? Currently: Markdown files vs. possibly SQLite/JSON vs. possibly RDF/graph databases. A unified memory interface does not yet exist.

4. Conclusion: The Agent of the Skills Era

2023: prompts. 2024: RAG for memory. 2025: MCP for tool connections. 2026: Skills equipping agents with full capabilities.

These nine skills are the embryo of an ecosystem, pointing in one direction: agents must evolve from "able to answer questions" to "able to complete tasks independently." The key isn't stronger models but better equipment — enabling agents to see, search, remember, learn, act, write, and review.

Platforms like SkillHub.cn are becoming the agent "App Store" — except the "apps" are for agents, not humans. Their users are AI, their install commands are curl, their runtime is Markdown.

A future agent's résumé might not say "I run GPT-5" but "I run Agent Browser, Proactive Agent, Self-Improving Agent, and Humanizer, support OpenClaw and Claude Code, and passed Skill Vetter certification."

In the skills era, agents compete on equipment too.

Project Links

1. Self-Improving Agent: https://github.com/peterskoett/self-improving-agent 2. Summarize: skillhub.cn/skills/summarize (URL to be confirmed) 3. Github (SkillHub Manager): https://github.com/Ray0907/skillhub 4. Agent Browser: https://github.com/vercel-labs/agent-browser 5. Skill Vetter: skillhub.cn/skills/skill-vetter (URL to be confirmed) 6. Ontology: skillhub.cn/skills/ontology (URL to be confirmed) 7. Proactive Agent: https://github.com/halthelobster/proactive-agent 8. Multi Search Engine: skillhub.cn/skills/multi-search-engine (URL to be confirmed) 9. Humanizer: https://github.com/Aboudjem/humanizer-skill