ctxeng v0.1.8 — Production context engineering (RAG, redaction, tracing, CI)

Highlights

This release adds production-grade context management for large codebases:

• Secrets & PII redaction (default on): masks common API keys/tokens, passwords, emails, and private keys before token counting, tracing, and output.
  Disable with: --no-redact

• RAG / chunk-level retrieval: --rag switches from whole-file inclusion to selecting the most relevant chunks for the query.

  • Uses embeddings if sentence-transformers is installed
  • Falls back to lexical retrieval otherwise

• AST skeleton mode (Python): --skeleton outputs a high-level outline (imports, classes, methods, function signatures) instead of full bodies.

• Local observability / tracing: --trace writes JSONL traces under .ctxeng/traces/ and includes trace_id + trace_path in metadata.

• CI support: new ctxeng ci subcommand for pipeline-friendly context generation (always writes to --output).

• Context snapshots: --snapshot saves a versioned bundle under .ctxeng/snapshots/<id>/ (output + manifest).

• Few-shot examples: --fewshot injects examples from .ctxeng/examples/ into the context.

Security & repo hygiene

• Respects .gitignore + .ctxengignore by default.
• Adds --allow / --deny path filters for access-control style scoping.

Output

• Improved structure with metadata + included-file index (XML + Markdown).

VSCode extension

The VSCode extension is disabled in this repo version because it is still under development. Please use the ctxeng CLI / Python package for now.

Upgrade

pip install -U ctxeng

v0.1.6 — VS Code Extension + Streaming AI + Smart Context

🚀 CtxEng AI v0.1.6

This release marks a major milestone — introducing the VS Code extension along with powerful improvements in context engineering and AI interaction.

---

✨ New Features

🧠 VS Code Extension (NEW)

• Full sidebar UI for interacting with your codebase
• Build context directly inside VS Code
• Seamless integration with AI providers

⚡ Streaming AI Responses

• Real-time response streaming (like ChatGPT)
• Faster and more interactive experience

📄 Context Transparency

• View selected files used for context
• Understand how context is built

🧭 Clickable File Navigation

• Open relevant files directly from results
• Faster debugging and exploration

🔍 Diff View for Code Suggestions

• Compare original code vs AI suggestions
• Improve code confidently

📊 Metrics Dashboard

• Track:
  • Tokens used
  • Estimated cost
  • Number of files selected

---

🛠 Improvements

• Better file relevance scoring
• Improved CLI performance
• Cleaner UI and interaction flow
• Enhanced error handling

---

💡 Why this matters

LLMs don’t fail because of bad prompts —
they fail because of bad context.

CtxEng focuses on solving this core problem.

---

🔗 Try it

VS Code Marketplace:

https://marketplace.visualstudio.com/items?itemName=saeemabkari6.ctxeng-ai

under development

GitHub:

https://github.com/Sayeem3051/python-context-engineer

---

🙌 Feedback

If you find this useful, consider:

• ⭐ Starring the repo
• Reporting issues
• Sharing feedback

---

🚀 What’s next

• "Why this file selected" explanations
• Local LLM integration
• More context optimization features

---

Built with ❤️ for developers.

v0.1.5: feat: add semantic similarity scoring

🚀 What’s New in ctxeng

This release introduces powerful automation, smarter relevance scoring, improved metadata, and important fixes to enhance the overall developer experience.

---

✨ New Features

🔄 Auto Context Rebuild (watch mode)

• Added ctxeng watch command to automatically rebuild context when files change
• Designed for a smoother development workflow without manual re-runs

Key capabilities:

• ⏱️ Debounced rebuilds to prevent excessive processing
• 📁 Optional output file support
• 🛑 Graceful shutdown using Ctrl+C

Example usage:

ctxeng watch

v0.1.4- Automatically rebuild context

🚀 What’s New in ctxeng

This release introduces powerful automation features, improves package metadata, and resolves minor issues to enhance overall developer experience.

---

✨ New Features

🔄 Auto Context Rebuild (watch mode)

• Added ctxeng watch command to automatically rebuild context when files change
• Designed for a smoother development workflow without manual re-runs

Key capabilities:

• ⏱️ Debounced rebuilds to prevent excessive processing
• 📁 Optional output file support
• 🛑 Graceful shutdown using Ctrl+C

Example usage:

ctxeng watch

v0.1.3 — Import Graph + Cost Estimates

What's New

✨ Import Graph Analysis

• ctxeng now automatically pulls in locally imported Python modules
• If auth.py is relevant, files it imports get included automatically
• Configurable depth (default: 1 hop) and score decay (default: 0.7)
• Works via static AST parsing — no runtime execution
• New API: build_import_graph(), expand_with_imports()
• New CLI flags: --import-graph-depth N, --no-import-graph

💰 Cost Estimates

• ctx.cost_estimate shows rough USD input cost for known models
• Shown automatically in ctx.summary():
  Est. cost: ~$0.037 (claude-sonnet-4)
• New public API: estimate_cost(tokens, model)
• New CLI flag: --no-show-cost to hide from output
• Supported: Claude, GPT-4o, GPT-4, Gemini, Llama

Previous releases

• v0.1.2 — .ctxengignore support, pathspec dependency
• v0.1.0 — Initial release

Installation

pip install --upgrade ctxeng

Full Changelog

https://github.com/Sayeem3051/python-context-engineer/blob/main/CHANGELOG.md

v0.1.0 — Initial Release

First public release of ctxeng.

• ContextEngine and ContextBuilder APIs
• AST-aware, git-aware relevance scoring
• Smart token budget optimization
• Claude, OpenAI, LangChain integrations
• CLI: ctxeng build / ctxeng info
• Zero required dependencies