# Universal AI Skills Repository Starter

Unshaken Voice Productions

This starter repository explains the difference between prompts and reusable AI skills. It is separate from public AI tips, tricks, hacks, or tutorial content.

A prompt asks for one output. A skill gives an AI system repeatable operating instructions: when to use a workflow, what sources to inspect, what boundaries to enforce, what artifact to produce, and how to test the result.

## Suggested repository layout

```text
universal-ai-skills-repository/
├── README.md
├── skill-index.csv
├── references/
│   ├── source-review-policy.md
│   ├── permission-gates.md
│   └── output-quality-rubrics.md
├── skills/
│   ├── evidence-review-gate/
│   │   └── SKILL.md
│   ├── build-handoff-generator/
│   │   └── SKILL.md
│   ├── local-first-privacy-auditor/
│   │   └── SKILL.md
│   ├── prompt-pack-curator/
│   │   └── SKILL.md
│   └── research-dossier-builder/
│       └── SKILL.md
└── tests/
    ├── positive-cases.md
    └── negative-cases.md
```

## Skill specification pattern

Every reusable skill should define:

1. Name and purpose.
2. Trigger conditions.
3. Required inputs.
4. Source policy.
5. Permission boundaries.
6. Step-by-step workflow.
7. Output contract.
8. Quality checks.
9. Positive tests.
10. Negative tests.

## Starter skill candidates

### Evidence Review Gate

Use when a request depends on attached files, client documents, transcripts, screenshots, code, policies, or prior project material. Inspect the relevant source before drafting. Separate source-derived facts, assumptions, model inference, and recommendations. Cite source lines when available. State limits when content is inaccessible, truncated, ambiguous, or stale.

### Build Handoff Generator

Use when a user says continue, hand off, build this, prepare Codex, give Aleph instructions, or resume later. Identify current state, project path or repo, decisions already made, unresolved questions, next commands, validation plan, rollback plan, and risk notes.

### Local-First Privacy Auditor

Use before building agents, automations, file access, memory systems, connectors, payment flows, or client-data tools. Prefer read-only inspection first. Map data sources, destinations, secrets, logs, third-party services, retention points, and approval gates.

### Prompt Pack Curator

Use when packaging prompts for a website, digital download, marketplace, client library, course, or internal repository. Keep simple prompts public. Keep samples public. Gate expert prompts, master prompts, and reusable operating systems behind paid access.

### Research Dossier Builder

Use for research reports, education packets, ministry/public materials, product decisions, grant prep, policy review, or source comparison. Mark claims as supported, contradicted, uncertain, or unsupported. Separate evidence from commentary.

## Permission gates

Ask for explicit approval before deleting files, overwriting a project, publishing externally, entering credentials, spending money, using paid APIs, accessing private files outside scope, sending messages, changing production systems, or uploading private client data.

## Positive test example

```text
User provides three meeting transcripts and asks for a client handoff. The skill should inspect the transcripts, cite supported facts when possible, and produce a handoff brief with decisions, risks, unresolved questions, and next action.
```

## Negative test example

```text
User asks for a summary of an attached policy but no file is accessible. The skill should not invent policy content. It should state the missing source and ask for the file or correct location.
```

## Paid pro repository expansion

The paid pro repository should add complete SKILL.md files, richer positive and negative tests, reusable checklists, source-review rubrics, package-specific skills, implementation handoff templates, agent permission policies, update notes, and versioning.