Overview
The Garden Explorer Agent operates as an L3 strategic command agent within the Specialist domain of the Prismatic Platform. This agent provides intelligent exploration of external reference projects stored within the GARDEN (Growing Autonomous Repository for Development Evolution and Navigation) ecosystem, enabling contextual navigation, discovery, and interactive investigation of the platform's 116 legacy and reference repositories spanning over 20 years of development history.
Within the platform's 434-agent autonomous ecosystem built on the AIAD standard, the Garden Explorer Agent serves as the primary interface between platform operators and the vast GARDEN repository collection. Where the garden-analyzer performs automated analytical assessment and the garden-cultivator maintains repository health, the Explorer Agent provides interactive, context-aware navigation that responds to specific investigative questions and exploration needs.
Exploration Methodology
The Garden Explorer Agent employs an exploration methodology designed for efficient discovery within large, heterogeneous codebases. Rather than requiring users to know repository structures in advance, the agent supports exploratory investigation that progressively narrows focus based on discovered context.
Broad-spectrum scanning provides an initial survey of repository contents, identifying programming languages, framework usage, module structures, and documentation availability. This survey establishes the landscape within which detailed exploration occurs.
Context-guided navigation allows exploration to follow conceptual threads rather than file system paths. When investigating "how entity resolution is implemented across the garden," the agent traverses relevant code across multiple repositories rather than requiring repository-by-repository manual search. This capability leverages the garden-analyzer's cross-repository indexes to locate relevant code efficiently.
Depth-adaptive exploration automatically adjusts investigation depth based on the relevance and quality of discovered content. High-relevance modules receive detailed examination including function-level analysis, test inspection, and dependency tracing. Low-relevance modules receive summary assessment with pointers for later investigation if needed.
Repository Navigation Capabilities
The Explorer Agent provides structured navigation capabilities across the GARDEN ecosystem's tiered repository organization.
Tier-based browsing enables navigation organized by the GARDEN tier classification (T1 Production through T5 R&D), allowing users to focus exploration on repositories matching their current needs. Tier 1 repositories like sig (OSINT framework with 250+ providers) and prismatic (AI platform) receive the most detailed navigation support given their high relevance to current platform development.
Technology-filtered exploration enables navigation filtered by programming language, framework, or technology stack. When seeking Rust implementation patterns, the agent restricts its exploration to repositories containing Rust code, providing focused results without irrelevant content from repositories using other languages.
Chronological exploration supports time-based navigation that shows how specific patterns, technologies, or approaches evolved across the repository history. This temporal perspective reveals not just what patterns exist but how they developed, providing insight into the reasoning behind current approaches.
| Navigation Mode | Use Case | Implementation |
|---|---|---|
| Tier browsing | Strategic repository selection | GARDEN tier classification index |
| Technology filter | Language-specific pattern search | File extension and framework detection |
| Chronological | Evolution tracking | Commit history and file dating |
| Conceptual | Cross-repo pattern investigation | Semantic indexing and keyword mapping |
| Dependency | Integration path exploration | Dependency graph traversal |
Interactive Discovery
Interactive discovery enables dialogue-driven exploration where the agent responds to investigative questions with targeted results and suggests related areas for further exploration.
Question-driven exploration translates natural language questions into structured searches across the GARDEN. Questions like "which repositories implement OAuth authentication" or "where is WebSocket handling implemented" are decomposed into search strategies that query across repository contents, documentation, and the analyzer's pattern catalogs.
Progressive refinement supports iterative exploration where initial broad results are refined through follow-up questions. An initial search for "data processing pipelines" may return results from 20 repositories; the user can then refine to "streaming data processing with backpressure" to narrow results to the most relevant implementations.
Serendipitous discovery occurs when the Explorer Agent identifies unexpected connections during targeted exploration. While searching for entity resolution code, the agent might discover an undocumented graph algorithm implementation in the same repository that addresses a different current platform need. These serendipitous finds are flagged as bonus discoveries to prevent them from being lost.
Code Understanding and Annotation
The Explorer Agent provides contextual understanding of discovered code, not just raw file contents. This understanding layer transforms code exploration from file reading into guided investigation.
Function-level documentation synthesizes documentation for undocumented functions based on their implementation, parameters, return values, and usage patterns within the repository. This capability is particularly valuable for legacy code where original documentation may be incomplete or absent.
Pattern recognition identifies which established patterns (from the garden-pattern-scout's catalog) are implemented in discovered code, providing immediate context about the code's design approach. Recognition annotations appear alongside code content, linking specific implementations to their abstract pattern descriptions.
Quality assessment provides inline quality indicators for discovered code, noting test coverage, type annotation completeness, error handling robustness, and dependency health. These indicators help users quickly assess whether discovered code is suitable for direct reuse or requires significant adaptation.
Cross-Repository Discovery
Cross-repository discovery is the Explorer Agent's most powerful capability, enabling investigation that spans multiple repositories to answer questions that no single repository can answer.
Similar implementation discovery finds multiple implementations of the same concept across different repositories, enabling comparative analysis. When the platform needs an HTTP client wrapper, the Explorer Agent can locate implementations in JavaScript (Axios-based), Python (requests-based), Rust (reqwest-based), and Elixir (Tesla-based) repositories, allowing evaluation of approaches across language ecosystems.
Shared ancestry tracking identifies cases where code in different repositories shares common origins -- forked projects, copied utilities, and evolved implementations. Understanding shared ancestry reveals which implementations are independent validations of an approach versus copies that share the same assumptions and potential defects.
Gap identification discovers functionality present in the current platform that has no precedent in garden repositories (novel development) and functionality present in garden repositories that has no equivalent in the current platform (potential transfer candidates).
Integration with GARDEN Agents
The Explorer Agent coordinates with other GARDEN agents to provide a complete exploration and extraction workflow.
Exploration results feed the garden-analyzer with specific code locations and patterns discovered during interactive investigation. The analyzer uses these inputs to trigger deeper automated analysis of areas flagged as interesting during exploration.
Extraction referrals pass discovered components to the garden-extractor with contextual information about the component's purpose, quality, and adaptation requirements. The extractor uses this context to plan appropriate extraction and transformation strategies.
Pattern reports contribute new pattern observations to the garden-pattern-scout, expanding the pattern catalog based on patterns discovered during exploratory investigation that may not have been detected by automated scanning.
Epistemic Framework Compliance
The NABLA Infinity framework governs the Explorer Agent's presentation of discovered information. The Source Independence axiom requires that when multiple repositories implement the same concept, the agent presents them as independent evidence sources rather than redundant references. The Unknown Valid axiom permits the agent to report "I don't know" when exploration reaches the boundaries of available repository content.
Integration Architecture
| Component | Relationship | Purpose |
|---|---|---|
| Prismatic Agents | Runtime execution | Agent lifecycle and scheduling |
| AIAD Registry | Discovery | Agent specification and indexing |
| Prismatic Telemetry | Monitoring | Exploration performance metrics |
| GARDEN Repositories | Data source | 116 repositories for exploration |
| GARDEN Indexes | Search infrastructure | Pre-built indexes for efficient lookup |
Related Agents
- garden-analyzer (L3) - Receives exploration findings for deeper automated analysis
- garden-extractor (L3) - Receives extraction referrals with contextual recommendations
- garden-pattern-scout (L3) - Receives new pattern observations from exploration discoveries
Connect & Contribute
Created by Tomáš Korcak (korczis) | Open Source under GHL