---
title: "Insights | GEO, AI and security notes | TauX"
description: "TauX's technical notes on GEO, MCP, AI agents, LLM security and post-quantum cryptography. Each names its sources and where its conclusions stop applying."
url: "https://taux.io/en-US/insights"
locale: "en-US"
alternates:
  ja-JP: "https://taux.io/ja-JP/insights"
  ko-KR: "https://taux.io/ko-KR/insights"
  zh-Hans-CN: "https://taux.io/zh-Hans-CN/insights"
  zh-Hant-TW: "https://taux.io/zh-Hant-TW/insights"
---

# Insights

Technical notes we write for ourselves and for clients. Each one names its sources and says where its conclusions stop applying.

## Articles

*   [GEO Guide](https://taux.io/en-US/geo-guide)

    What Generative Engine Optimization is, how it differs from SEO, and the nine methods a peer-reviewed study found actually move citation rates.

*   [Reading the Threat Landscape](https://taux.io/en-US/threat-landscape)

    Eight headline figures from the CrowdStrike 2026 Global Threat Report, how each was measured, and what that means for how far it generalises.

*   [Post-Quantum Cryptography Migration](https://taux.io/en-US/pqc-migration)

    Governance, CBOM inventory, risk assessment, crypto-agility and supply chain — mapped to the FSC's seven strategies, with the US, EU and Taiwan timelines and where each comes from.

*   [Workspace AI: reading the ROI](https://taux.io/en-US/google-workspace-with-ai)

    The method, sample size and vintage behind four widely quoted Workspace AI figures, set against independent research, and why self-reported time savings need objective verification.

*   [LLMs.txt Explained](https://taux.io/en-US/what-is-llms-txt)

    What llms.txt is, what it expresses that robots.txt cannot, and an honest account of its status: a proposal in use, not a ratified standard.

*   [What is MCP](https://taux.io/en-US/what-is-mcp)

    Host, client and server; why prompts, resources and tools differ by who decides rather than read versus write; and what the 2026-07-28 revision removed.

*   [Prompt Injection Explained](https://taux.io/en-US/what-is-prompt-injection)

    How prompt injection works, how it differs from jailbreaking, and the layered defences that actually reduce the risk.

*   [OWASP LLM Top 10](https://taux.io/en-US/owasp-llm-top-10)

    Each of the ten risks with its definition, real attack surface and defences — plus the 2025-to-2026 reordering, the four entries that widened, and how the ranking method changed.

*   [Prompting Guide](https://taux.io/en-US/agent-prompting-guide)

    What to specify, what to leave open, and how to tell whether it worked. Covers role and task framing, step instructions, output formats, and designing agents.

*   [A practical guide to building Claude Skills](https://taux.io/en-US/claude-skills-guide)

    Turning an SOP or a piece of expert knowledge into a reusable automated workflow: folder structure, the YAML rules that decide whether it loads, testing, and release.

*   [ADK Skill Patterns](https://taux.io/en-US/adk-skill-patterns)

    Tool Wrapper, Generator, Reviewer, Inversion and Pipeline — the five patterns from Google Cloud Tech, and what problem each one is actually solving.

*   [An agent development workflow](https://taux.io/en-US/agent-dev-workflow)

    Chain the 25 skills in mattpocock/skills into one repeatable development flow: grill, spec, tickets, implement, code-review. Four worked scenarios, 31 known pitfalls, and a 14-day plan.

## Open Source

[TWSE MCP](https://github.com/taux-io/twse-mcp) 

An MCP server for the Taiwan Stock Exchange's open data — quotes, ETFs and 140+ public datasets — so AI applications can query it directly. We maintain it in the open.