Prompt Library and Skills

Prompts and recovery strategies that work with the VibeReady AI framework.

How it works: AGENTS.md provides core rules. Scoped rules auto-load based on the files you edit. Agent Skills handle multi-step workflows. You describe what you want, and the framework handles the rest.

---

Agent Skills

For multi-step workflows, use Agent Skills. Skills use keyword-based discovery — the AI tool matches your request against keywords in each skill’s description. If you say “build a user profile page,” the tool may auto-discover the new-feature skill because “build” is a keyword. But this is unreliable across tools and sessions.

Always invoke by name for guaranteed results:

"Use the new-feature skill to build a user profile page."   -> guaranteed
"Build a user profile page."                                  -> might skip the skill

When the skill is invoked by name, the AI loads the full workflow (pre-flight checks, three-gate system, TDD steps, completion checklist). Without it, the AI uses its general knowledge and may skip project patterns.

Skill	When to Use	Invoke With
specify	Defining WHAT: new features, complex bugs, significant changes	”Use the specify skill”
plan	Defining HOW: decomposing any complex change into tasks	”Use the plan skill”
new-feature	Building a new feature end-to-end	”Use the new-feature skill”
new-api-route	Adding an API endpoint with auth + validation	”Use the new-api-route skill”
fix-bug	Investigating and fixing a bug	”Use the fix-bug skill”
refactor	Refactoring code safely	”Use the refactor skill”
add-test	Adding tests for existing code	”Use the add-test skill”
stack-swap	Swapping a technology in the stack	”Use the stack-swap skill to replace Prisma with Drizzle”
code-review	Reviewing code for architecture and patterns	”Use the code-review skill”
design-review	Reviewing UI for accessibility and design system	”Use the design-review skill”
manual-qa	Smoke-testing a feature in the browser after building it	”Use the manual-qa skill”
qa	Analyzing test quality and coverage gaps	”Use the qa skill”
security-reviewer	Security audit with OWASP checklist	”Use the security-reviewer skill”
readme-updater	Updating module documentation	”Use the readme-updater skill”

Skills follow the Agent Skills open standard and are stored in a dedicated skills directory within the project.

Recommended Flow for Complex Changes

For changes that span multiple files or require architectural decisions:

1. "Use the specify skill to create a PRD for [change]."        (WHAT)
2. "Use the plan skill to decompose PRD-[name] into tasks."     (HOW)
3. "Use the [skill] skill to implement Task 1: [title]."        (BUILD)
   (new-feature, new-api-route, fix-bug, or refactor -- repeat for each task)
4. "Use the manual-qa skill to verify the feature in the browser." (MANUAL QA)
5. "Use the code-review skill to review the implementation."    (REVIEW)
   (also: design-review for UI changes, qa for test coverage)
6. "Use the security-reviewer skill to audit [change]."         (REVIEW)
7. "Use the readme-updater skill to update READMEs."            (DOCUMENT)

Shortcuts for simpler work:

• Simple feature — skip to step 3 with new-feature
• Simple bug — skip to step 3 with fix-bug
• Large refactoring — skip step 1, start with plan (step 2)

For the full lifecycle walkthrough, see PRD Workflow.

---

Plan Mode

Always use plan mode for complex tasks. It ensures AGENTS.md and all scoped rules load before the AI begins working.

Tool	Activation
Claude Code	Shift+Tab
Cursor	/plan or Composer
Windsurf	Cascade
Gemini CLI	/plan

---

How Context Loading Works

The framework uses a layered approach:

1. AGENTS.md (always loaded) — Core rules: multi-tenancy, TDD, DRY, security
2. Scoped rules (auto-loaded per tool) — Relevant patterns based on files you edit
3. Agent Skills (invoked on demand) — Multi-step workflow instructions

Each AI tool reads AGENTS.md automatically through its native mechanism or via a symlink created during setup. Scoped rules load from each tool’s dedicated rule directory. You do not need to know which rules exist — they load automatically based on the files you touch.

---

Simple Task Prompts

For tasks too small for a full skill workflow, just describe what you want. AGENTS.md and scoped rules load automatically — no special prompt needed.

Task	Example Prompt
Build Feature	Build {feature description}.
Fix Bug	Fix {bug description}.
Research Code	Explain how {feature} works.
Add Tests	Add tests for {feature}. Check existing tests for patterns.
Refactor	Refactor {what to refactor}.
Use External Lib	Implement {feature} using {library}. Use Context7 for docs.

For multi-step changes, use the matching Agent Skill instead (see the table above).

---

Recovery Prompts

When AI skips the framework or ignores project conventions, use these prompts to get back on track:

Problem	Prompt
Skipped framework entirely	Stop. Follow AGENTS.md first.
Not checking READMEs	Stop. Check README.md files in the target area first.
Not using TDD	Stop. Write a failing test first, then implement.
Using generic patterns	Stop. Check existing code for project patterns.
Wrong library usage	Use Context7 for {library} docs first.
Ignoring scoped rules	What scoped rules loaded? Check the tool's rules directory.

---

Related Documentation

• AI Development Framework — Full architecture and component overview
• PRD Workflow — When and how to use PRDs
• Build Your First Feature — Hands-on quick start
• Troubleshooting — Fix framework issues