Shakehand
Security response workspace with a laptop, notebooks, and analysis tools.

AI literacy and security reasoning

Shakehand

A practical index for learning how to ask better AI questions, evaluate answers, and design phishing defenses that bind claims to evidence instead of confidence alone.

Browse the index View the technical model

Section index

Two attached guides, organized into one landing page.

The page groups the source material into learner-facing prompt guidance and technical security architecture. Each section is written so a reader can scan the idea, choose a path, and move into the details without opening a separate document.

01

Prompt Confidence Framework

Goal, context, task type, success criteria, and the prompt map for common AI work.

 02

Grounded Phishing Detection

Evidence-conditioned tree search, typed claims, calibrated risk, and fail-closed behavior.

 03

Learning Path

A five-level progression from simple questions to evaluated, iterative AI workflows.

 04

Templates and Rubrics

Reusable prompt structures, quality checks, and workshop sessions for teams.

Prompt confidence

Useful prompting starts with clear thinking.

A strong prompt tells the AI the goal, the context, the task type, and the success criteria. Shakehand frames prompting as structured communication, not a list of magic words.

Role Context Task Constraints Format Quality
Illustration of a structured report and checklist.

Prompt map

Choose the kind of thinking before writing the prompt.

Understand

Fact-finding, research support, summaries, explanations, and comparisons.

Improve

Editing, tone shifts, critique, quality checks, and revision loops.

Organize

Classification, extraction, structuring, tables, checklists, and action items.

Decide

Scenario planning, prediction, data analysis, decision support, and tradeoffs.

Design

Plans, workflows, lessons, workshops, rubrics, and implementation roadmaps.

Create

Examples, activities, titles, scripts, analogies, and creative alternatives.

Grounded detection

Phishing decisions should be evidence-conditioned.

The technical deep dive proposes adaptive tree search over explicit hypotheses. The system decides what to check next based on current evidence, then binds every factual claim to a machine-checkable evidence record.

Search loop

  1. Run deterministic tier-zero checks.
  2. Seed a risk belief and evidence store.
  3. Select the next hypothesis branch with UCT.
  4. Gather high-information evidence or refine the assessment.
  5. Return the best verdict before the deadline.

Grounding rule

Factual claims must cite an evidence ID and pass typed comparison against the store. Unsupported claims trigger investigation or receive a heavy reward penalty.

Risk model

Cryptographic, registrar, reputation, and language signals combine as calibrated log-odds. Strong infrastructure evidence outweighs weak prose-level impressions.

Implementation notes

Design the system to fail closed under pressure.

Search is reserved for uncertain or high-stakes messages. It runs under a latency cap, stops early on conclusive evidence, caches repeated infrastructure checks, and treats budget exhaustion as a reason to quarantine or escalate.

Evidence store Typed claims Reward verifier Latency budget Information gain Fail-closed policy

Learning path

Build confidence through progressive control.

The guide moves learners from asking simple questions to evaluating AI output with explicit criteria. The important shift is from receiving answers to steering and checking the work.

  1. AskStart with a direct question.
  2. Add audienceSay who the answer is for and why it matters.
  3. Add formatRequest bullets, tables, checklists, scripts, or matrices.
  4. Add criteriaDefine accuracy, usefulness, tone, level, and constraints.
  5. IterateReview, revise, simplify, add examples, and check weak spots.

Reusable tools

Templates for repeatable AI work.

Shakehand turns the guide into a practical prompt library: explain, summarize, edit, classify, extract, plan, compare, critique, and create. Each template keeps the human responsible for judgment by making assumptions and quality criteria visible.

Explain [topic] for [audience]. Extract [elements] into [format]. Compare [options] using [criteria]. Review this work for [risks].

Static site ready

Designed for Cloudflare Pages at shakehand.co.

The page is intentionally static: one HTML file, one stylesheet, and local image assets. That keeps hosting simple, fast, and easy to cache at the edge.

hello@shakehand.co Back to index