Code-Assistant (ca) A developer focused AI integration

Overview

Code-Assistant (ca) is a CLI tool designed to assist developers by integrating AI-powered modifications into their code workflow. It supports file refactoring, summaries, commit tracking, and LLM customization via gollm.

This tool is designed to work seamlessly within a Git-tracked or non-Git project, allowing developers to iteratively improve their code while maintaining full control over changes.

⚠️ Warning: Code-Assistant (ca) is still under development. Expect changes and improvements as the project evolves.

---

Key Features

✅ AI-Powered Code Editing – Modify code using prompts.
✅ Per-File or Batch Processing – Apply changes to one or multiple files.
✅ LLM Customization – Use gollm for local or remote LLM models.
✅ Git Integration (Optional) – Track commits and rollback changes.
✅ Session-Based Workflow – Log AI interactions for review and undo.
✅ Summarization Mode – Review code without modifying files.
✅ Configurable Defaults – Customize behavior via environment variables.

---

Command Specification

Command	Description
ca start-session "<description>"	Start a new coding session.
ca code "<prompt>" [--files file1 file2] [--per-file] [--dry-run] [--commit]	Apply AI modifications.
ca review	Show session progress and diffs.
ca commit "<message>"	Commit current changes to Git.
ca rollback --step N	Undo a specific AI-modified step.
ca replay-step --step N [--prompt "<new prompt>"] [--files file1 file2]	Modify a previous AI change.
ca end-session	Archive session to historical storage.
ca code "<prompt>" --summary [--output file] [--store-session]	Generate an analysis instead of modifying files.
ca config llm [--set model=gpt-4] [--list]	Manage LLM configuration.

---

Implementation Checklist

Basics

• Create stub commands
• Add make file so it can build and do other commands
• Add start-session and end-session command
• code command work
  • Initial work to do simple prompt againt one file
  • Add ability to do multiple files
  • Add dry-run
  • Add ability to do a commit at the end and save that info in file
• review command work
  • More TBD
• rollback command work
  • More TBD
• Add the summary mode into code command

File Tracking & Storage

• Implement .ca_session.json for tracking prompts, modified files, and steps.
• Store created, modified, and deleted files for each AI step.
• Track Git state (pre-commit and post-commit hashes).

---

Extensions Features

• Code Review & History

  • ca replay-step --step N [--prompt "<new prompt>"] [--files file1 file2] - Re-run a specific AI step with modifications.

• AI-Powered Summaries

  • ca code "<prompt>" --summary - Generate an AI analysis instead of modifying files.
  • --output file - Save the AI summary to a separate file.
  • --store-session - Store the summary output in the .ca_session.json session file.

• Git Integration

  • Detect and track files under Git.
  • Store pre-change and post-change commit hashes in .ca_session.json.
  • ca rollback --step N - Restore files to their previous Git state.

• Configuration Management

  • ca config llm --set model=gpt-4 - Configure the LLM model for AI processing.
  • ca config llm --list - List available LLM models.
  • Support CA_LLM_MODEL, CA_STORE_SUMMARY, and other environment variables.

---

CLI Behavior & UX

• Use urfave/cli for handling command-line arguments.
• Implement global config defaults (CA_DEFAULT_MODE, CA_STORE_SUMMARY).
• Ensure --summary, --per-file, and --batch behavior is intuitive.
• Allow per-command overrides for global config settings.

---

LLM Integration (Using gollm)

• Integrate gollm for AI processing.
• Support local LLM models (e.g., LLaMA, Mistral).
• Support remote APIs (e.g., OpenAI GPT).
• Allow per-command model selection (--llm-model <model>).
• Allow configuring temperature and system prompts.

---

Error Handling & Safety

• Prevent AI modifications on untracked Git files (unless overridden).
• Display warnings when AI output exceeds token limits.
• Implement proper rollback in case of errors.
• Ensure commands fail gracefully when necessary.

---

Usage Specification

1️⃣ Start a New Session

ca start-session "Refactoring legacy code"

• Starts tracking AI modifications in .ca_session.json.

2️⃣ Apply AI Modifications

Default (Batch Mode)

ca code "Convert functions to use named arguments" --files main.go utils.go

• Modifies all specified files at once.

Per-File Mode

ca code "Refactor functions" --files main.go utils.go --per-file

• Processes each file separately to avoid token limits.

Dry Run (Preview Changes)

ca code "Upgrade to Rails 7" --dry-run

• Shows the AI-generated diff but does NOT modify files.

Auto-Commit After AI Changes

ca code "Improve error handling" --commit

• Commits the AI-modified files after applying changes.

3️⃣ Review AI Changes

ca review

• Displays session steps and modified files.
• Shows Git commit history (if tracking with Git).

4️⃣ Undo AI Modifications

Rollback a Specific Step

ca rollback --step 2

• Restores files to pre-step commit.

Replay a Step with a New Prompt

ca replay-step --step 2 --prompt "Use named structs instead of tuples"

• Updates the step without creating a new one.

5️⃣ Generate AI Summaries

Default (On-Screen)

ca code "Review changes for potential bugs" --summary

• Shows an AI-generated analysis without modifying files.

Save Summary to a File

ca code "Analyze security vulnerabilities" --summary --output security_review.txt

• Saves output to ``.

Store Summary in Session

ca code "Review for bugs" --summary --store-session

• Saves **analysis inside **``.

6️⃣ Commit & End the Session

ca commit "Completed function refactor"
ca end-session

• Moves the session to historical storage.

---

LLM Integration (``)

Configure LLM Defaults

ca config llm --set model=llama-3 --set temperature=0.7

• Sets the LLM model and temperature globally.

List Available Models

ca config llm --list

• Shows supported local/remote models.

Override LLM for a Single Run

ca code "Optimize SQL queries" --llm-model gpt-4

• Uses gpt-4 for this command without changing global settings.

---

Getting Started

1️⃣ Install Code-Assistant

go install github.com/chrisrob11/codeassistant@latest

Or Clone & Build

git clone https://github.com/chrisrob11/codeassistant.git
cd codeassistant
go build -o ca
mv ca /usr/local/bin/

2️⃣ Configure Your LLM

export CA_LLM_MODEL="llama-3"
export CA_LLM_TEMPERATURE=0.7

3️⃣ Start Using It

ca start-session "Refactoring project"
ca code "Improve function names" --per-file
ca review
ca commit "Refactored functions"
ca end-session

---

Final Thoughts

💡 This CLI is flexible, AI-powered, and integrates well with Git.
🚀 It allows both direct file modification and AI-powered analysis.
🔧 LLM settings are fully configurable via ``.

📌 This is currently under development—track progress with the implementation checklist! 🚀