AI Agent Tooling (.agent)
Relevant source files
The following files were used as context for generating this wiki page:
The AI Agent Tooling within KudosWall is a specialized subsystem designed to automate social proof management, content analysis, and marketing strategy. It serves as the "AI Core" of the platform, leveraging advanced prompt engineering and structured data extraction to handle high-fidelity text and video testimonial processing.
This system is integrated into the monorepo architecture, primarily residing within the web application layer while interacting with the core API and database packages to provide features like sentiment detection, automated case study generation, and spam filtering.
Sources: apps/web/src/lib/ai/prompts.ts:18-20, README.md:12-15, AGENTS.md:1-5
Core AI Capabilities and Persona
The AI Core is defined by a central system prompt that establishes its persona as an expert in content analysis, sentiment detection, and marketing strategy. It operates under strict guidelines to identify mismatches between user ratings and written feedback while maintaining the original customer "voice".
Operational Boundaries
- Objective Analysis: Focuses on deep analysis of text and video transcripts.
- Structured Output: Prioritizes high-precision JSON output for system integration.
- Tone Preservation: Maintains emotional intensity in translations and smart replies.
- Human-in-the-loop: Flags ambiguous content for human review rather than guessing.
Sources: apps/web/src/lib/ai/prompts.ts:24-45
AI Processing Pipeline
The AI system follows a structured flow from raw user input (text or video) to processed marketing assets. It utilizes Zod-validated schemas to ensure the AI SDK v6 outputs match the application's data requirements.
The diagram above illustrates the flow of a user testimonial through the AI Core processing various analysis and filtering tasks.
Sources: apps/web/src/lib/ai/prompts.ts:51-100, README.md:145-155
Technical Schemas and Prompts
KudosWall uses specific data structures for different AI tasks. These schemas define the contract between the LLM and the application logic.
AI Output Schemas
| Schema Name | Key Fields | Purpose |
|---|---|---|
analysis | sentiment, enthusiasm_score, themes, is_mismatch | Dashboard filtering and sorting. |
spam | is_legit, confidence, reason | Automated abuse prevention. |
feature_suggestion | feature_name, frequency_signal, marketing_angle | Product roadmap insights from bulk reviews. |
case_study | title, problem, solution, result | Transforming testimonials into marketing assets. |
translation | translated_text, language, cultural_notes | Localization while preserving intent. |
Sources: apps/web/src/lib/ai/prompts.ts:53-100
Prompt Engineering Strategies
The system utilizes several advanced prompting techniques:
- Few-Shot Prompting: Used in
ANALYZEandDETECT_SPAMto provide contextual examples of mismatches and toxic content. - Chain-of-Thought (CoT): Embedded in the
analysisschema via areasoningfield to improve transparency in sentiment detection. - Programmatic SEO (PSEO): Specific templates for generating competitor comparison pages and persona-based landing pages.
Sources: apps/web/src/lib/ai/prompts.ts:110-120, apps/web/src/lib/ai/prompts.ts:180-195
Specialized AI Modules
Video Suite Intelligence
For video testimonials, the AI tooling provides transcription refinement. This removes disfluencies (ums, uhs) while fixing run-on sentences, specifically keeping industry jargon and the author's personal voice for authenticity.
Sources: apps/web/src/lib/ai/prompts.ts:160-175
Growth and Marketing Engine
The AI generates strategic content aimed at "Generative Engine Optimization" (GEO). This involves refining testimonials to be highly "citable" by AI engines like Perplexity or ChatGPT by clarifying entity roles and converting qualitative wins into citable metrics.
Sequence showing the enrichment of testimonials for AI search engine visibility.
Sources: apps/web/src/lib/ai/prompts.ts:213-230, README.md:66-68
Skills and Standards
The AI Agent tooling is built using a decentralized skills architecture, tracked via a skills-lock.json file. This includes specialized logic for:
- AI-SEO: Search engine optimization specialized for generative AI.
- Copy-editing: High-precision text refinement.
- Competitor-profiling: Generating comparisons between KudosWall and competitors like Senja or Testimonial.to.
Sources: skills-lock.json:20-25, apps/web/src/lib/comparisons.tsx:35-50, AGENTS.md:15-20
Summary
The AI Agent Tooling in KudosWall is a robust system that automates the complex lifecycle of social proof. By combining structured JSON outputs with advanced prompt engineering, it enables the platform to act as a "Marketing Strategy" engine rather than just a storage tool. This automation extends from initial spam detection to high-value PSEO content generation, ensuring that every customer testimonial is maximized for conversion and search visibility.
Sources: apps/web/src/lib/ai/prompts.ts:24-40, AGENTS.md:50-55