AI Hands

AI Hands is a personal collection of agent skills and reusable engineering rules. It is meant to be read by coding agents, copied into agent environments, and used as a shared operating manual for recurring product, engineering, and workflow tasks.

The repository is intentionally lightweight: the Markdown files are the product. There is no application runtime, build step, or package manager setup for the root repository.

Repository Layout

.
├── README.md
├── rules/
│   ├── main.md
│   ├── context7.md
│   ├── folders.md
│   ├── git.md
│   └── web/
└── skills/
    ├── README.md
    ├── design/
    ├── engineering/
    ├── forthright/
    ├── marketing/
    ├── note-taking/
    ├── personal/
    ├── productivity/
    ├── seo/
    └── tools/

Skills

Most skills live under skills/<section>/<skill-name>/SKILL.md. Nested skill sets can live deeper; for example, UltraSkills uses skills/engineering/UltraSkills/skills/<skill-name>/SKILL.md, while Superpowers uses skills/engineering/superpowers/<skill-name>/SKILL.md. A skill can also include nearby reference documents, templates, assets, or scripts when the workflow needs more than one file.

Start with skills/README.md, then open the relevant section README:

Section	Use when	Index
Design	Design critique, high-fidelity prototypes, interactive demos, slides, animations, image generation, brand systems, or frontend taste workflows.	skills/design/README.md
Engineering	Research, code review, wayfinding, code planning, architecture, debugging, implementation, TDD, security, CI, deployment, macOS, web, UltraSkills staged research/review workflows, or Superpowers process workflows.	skills/engineering/README.md
Forthright	Agent-to-agent handoffs and compact internal operational writing.	skills/forthright/README.md
Marketing	Growth, positioning, acquisition, conversion, lifecycle, revenue, and launch work.	skills/marketing/README.md
Note-taking	Second Brain retrieval, capture, handoffs, reviews, and Obsidian vault note work.	skills/note-taking/README.md
Personal	User-specific writing, Obsidian, and recipe/cookbook workflows.	skills/personal/README.md
Productivity	General workflow control: handoffs, teaching, grilling, and writing better skills.	skills/productivity/README.md
SEO	SEO/GEO research, technical audits, content, schema, authority, monitoring, and reporting.	skills/seo/README.md
Tools	Direct operation of documents, PDFs, Obsidian CLI, and Playwright/browser automation.	skills/tools/README.md

Skill routing convention

Each section README separates:

• User-invoked skills — reachable only when explicitly called (disable-model-invocation: true).
• Model-invoked skills — model- or user-reachable, with trigger-oriented descriptions so agents can select them automatically.
• Imported or nested skill sets — collections such as UltraSkills or Superpowers; use their linked paths plus the section README when a skill name overlaps with another installed skill.

Hermes installation note

For Hermes Agent, install or point skills.external_dirs at this repository's skills/ directory. Hermes users should reload/start a new session after the repo update so nested SKILL.md files, including UltraSkills and Superpowers, are discoverable. If a nested skill has the same name as another installed skill, use the linked README path to choose the intended version.

Rules

Rules are reusable implementation guidance for agents. They are broader than skills: a skill defines a workflow, while a rule defines standing project preferences or architecture patterns.

File	Purpose
rules/main.md	Entry point that lists the rule set agents should load.
rules/context7.md	Requires Context7 for current library, framework, SDK, API, CLI, and cloud-service documentation.
rules/folders.md	Keeps new project code in subfolders instead of the repository root.
rules/git.md	Defines default Git and GitHub workflow expectations.
rules/credits-and-limits.md	Architecture guidance for credits, limits, account balances, and spend reservations.

Web-specific rules live in rules/web/:

File	Purpose
rules/web/analytics.md	Consent-aware analytics setup with Plausible, Google Tag Manager, and Yandex.Metrika.
rules/web/api.md	API implementation guidance.
rules/web/async-jobs-and-workers.md	Background job, worker, retry, idempotency, and observability patterns.
rules/web/auth.md	Google OAuth authentication architecture and security requirements.
rules/web/frontend.md	Frontend stack and design expectations.
rules/web/lemonsqueezy.md	Lemon Squeezy checkout, webhook, entitlement, and local data model guidance.
rules/web/testing.md	Full-stack testing strategy with Vitest and Playwright expectations.

Using This Repository

Use the repository as a source of truth for agent behavior:

1. Start at skills/README.md when choosing a skill section.
2. Open the chosen section README to choose between related skills.
3. Point the agent at the relevant SKILL.md once a match is clear.
4. Load rules/main.md when starting a project that should follow these defaults.
5. Copy or symlink selected skills and rules into the agent environment if your tool expects skills in a specific local directory.
6. Keep section README files updated whenever you add, rename, move, or remove a skill.

Adding A Skill

Use the writing-great-skills workflow. At minimum, every skill should include:

• A SKILL.md file.
• Frontmatter with name and description.
• A description that says exactly when the agent should use the skill.
• Concise instructions in the main file.
• Supporting references, templates, assets, or scripts only when they keep the main skill easier to read.

Recommended structure:

skills/<section>/<skill-name>/
├── SKILL.md
├── references/
├── templates/
└── scripts/

Maintenance Notes

• Keep instructions concrete and current.
• Prefer short skill files with links to focused references.
• Keep section README files as routing indexes, not full manuals.
• Avoid time-sensitive claims unless the skill also tells the agent how to verify them.
• Do not commit or push changes unless the user explicitly asks for it.
• When a rule depends on current third-party documentation, use Context7 instead of relying on memory.