Agentseed – Generate Agents.md from a Codebase

Parses source code files using Abstract Syntax Tree (AST) analysis to extract structural information including function signatures, class definitions, module exports, and dependency relationships. The tool traverses the AST to build a semantic map of the codebase architecture rather than relying on regex or simple text parsing, enabling accurate identification of public APIs and internal structure.

Transforms extracted codebase structure (functions, classes, exports, signatures) into a machine-readable Markdown document formatted specifically for AI agent consumption. The generator creates a structured Agents.md file that lists available functions, their parameters, return types, and usage context in a format optimized for LLM context windows and function-calling patterns.

Provides pluggable AST parser support for multiple programming languages, allowing the tool to analyze codebases written in JavaScript, TypeScript, Python, and potentially other languages. Each language uses its native or optimized parser (e.g., tree-sitter, Babel for JS/TS; ast module for Python) to ensure accurate structural extraction across heterogeneous codebases.

Monitors codebase changes (file additions, deletions, modifications) and regenerates only affected sections of Agents.md rather than re-parsing the entire codebase. Uses file hashing or modification timestamps to detect changes, then re-parses only modified files and updates the corresponding entries in the generated documentation.

Extracts detailed function signatures including parameter names, types, default values, and return types from parsed code. For typed languages (TypeScript, Python with type hints), captures full type information; for untyped languages, infers or documents parameter positions. Generates structured metadata that includes arity, parameter order, and type constraints.

Analyzes import statements and require() calls to build a dependency graph showing which modules depend on which others. Tracks both internal dependencies (within the codebase) and external dependencies (third-party libraries). Generates metadata about module relationships that helps agents understand code organization and potential side effects.

Parses docstrings, JSDoc comments, and inline comments from source code and includes them in the generated Agents.md. Extracts structured documentation (parameter descriptions, return value docs, examples) from comment blocks and associates them with the corresponding functions. Handles multiple comment formats (JSDoc, Python docstrings, Markdown comments).

Identifies which functions, classes, and variables are exported from modules (public API) versus internal/private. Analyzes export statements, visibility modifiers (private, protected, public), and naming conventions to determine what should be exposed to agents. Filters out internal implementation details and focuses on the public contract.

Generates Agents.md in a standardized, machine-readable format with consistent structure for function listings, parameter documentation, and type information. Validates the generated Agents.md against a schema to ensure it's properly formatted and can be reliably parsed by AI agents. Supports multiple output formats (Markdown, JSON, YAML) for different agent consumption patterns.

Provides a command-line interface for running codebase analysis and Agents.md generation with support for configuration files (.agentseedrc, agentseed.config.js, etc.). Allows users to specify source directories, output paths, language support, filtering rules, and documentation options via CLI flags or config files. Integrates with standard development workflows and CI/CD pipelines.