Skip to content

AGENTS.md

Latest commit

 

History

History
71 lines (55 loc) · 2.85 KB

AGENTS.md

File metadata and controls

71 lines (55 loc) · 2.85 KB

Repository Overview

This project extracts code artifacts from a source tree, builds a call graph, creates text embeddings for each node, and provides tools to query or inspect those embeddings. The main scripts are located in the repository root.

Core Components

File Description
main.py Unified CLI with extract, embed, query, and interactive modes.
graph.py Extraction utilities and call graph helpers.
embedding.py Generates embeddings and FAISS indices.
query.py Searches embeddings and builds prompts for the LLM.
prompt_builder.py Formats search results and function summaries.
llm.py Helper routines for calling the Gemini API.
interactive_cli.py PromptToolkit dialogs for the interactive workflow.
config.py Loads and stores project settings.

Function Relationships

  1. Extraction – graph.extract_from_python, extract_from_html, extract_from_markdown, extract_from_javascript, and extract_from_typescript parse files and return a list of entries. build_call_graph adds every entry to a networkx graph (not just functions) and save_graph_json writes it to disk.
  2. Context Gathering – graph.gather_context collects code from related nodes using expand_graph. Direction can be controlled with the bidirectional setting.
  3. Embedding Generation – embedding.generate_embeddings loads the call graph, gathers context for each node, encodes the text with a SentenceTransformer, and saves a FAISS index.
  4. Querying – query.main searches the embeddings, builds prompts via prompt_builder.format_summary and optionally calls an LLM.
  5. Inspection – graph.analyze_graph prints basic statistics on the generated call graph.

The main.py script orchestrates these utilities via command line or interactive mode.

Usage Notes

Run python main.py <command> where <command> is one of:

  • extract – build a call graph for a project
  • embed – generate embeddings from a call graph
  • query – search the embeddings interactively
  • inspect – print basic graph statistics

Running main.py without a command launches the interactive workflow. Command history is stored in the ~/.full_context_history/ directory via prompt_toolkit. The interactive search can optionally correct queries using SymSpell when use_spellcheck is set to true in settings.json.

Debugging

Scripts print detailed progress information when run with PYTHONLOGLEVEL=DEBUG. Extraction and embedding generation also show tqdm progress bars for visual feedback.

Testing

Unit tests live in the tests/ directory. Run them with:

pytest -q

The main test file is tests/test_extractors.py which verifies the extractors and call graph generation. Always run pytest -q before committing changes.