gpterm 0.5.0 → 0.6.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: fe8a62780b9a0890cdb4080971d6810c36b160da4bdab39f6a77a7a69c104e07
-  data.tar.gz: d48d93b725e8714a8213513ef30a06a9771b9866da8124a8765b95b3060365c1
+  metadata.gz: a7f8d5042eee54849bba393ae5db6705de179ea794ad3221f71d4b14477fa1a7
+  data.tar.gz: efefab0a17acd5b1897d202496c3e41213cbd3502504a767fa9a218f5b75a7f1
 SHA512:
-  metadata.gz: d6cc7c0a7ae13f2f7d826cae3bbbe479cf12bb54f486d4c8eb0dba33896fa462709cd431f2577e351a40a41b4804502e1f4b991183aa3f32b50f7eb073a311a2
-  data.tar.gz: e9cb1e4214ac2c0f2fe8797ec35d821812dc883545b9cb61be0b1bc3ef50e79f3a9875c6a9f8a6a820474de90c9ac2bb38f966af8979ba6004928f60e402335a
+  metadata.gz: 3ae072a25e1b333d61f0ff076c4283b18eedf80b507d46dd790d3d1c6e620be003b66e3965dc6cb24add7794cc3016968e6e759d56673ea52dc7dde605381446
+  data.tar.gz: eaa98c3aa8d33abe149269c483554ddd8f9dc4cf7f6db4c1c4506ea4bdeac4a7adf487dd245ef5ea1bf9285113c2b78435bbc1bacb11a2aac3f68cd780016798

data/README.md

@@ -44,6 +44,15 @@ On first run, you'll be prompted to enter your OpenAI API key. This is required
 You can save and reuse preset prompts for common or repeated tasks. To create a preset, use the `-s` option followed by a name and the prompt, separated by a comma.
 To use a preset, simply pass its name as an argument when starting gpterm.
 
+### Example Presets
+
+I have been using the following presets while developing `gpterm` and have found them to be quite helpful.
+
+- `gc`: **Makes and pushes a commit describing the most recent changes.**
+  - Prompt: `"Commit all the latest changes to main, describing specifically what the changes are in the commit message. Please determine what the changes are before you write the commit message. You should use git diff and git status for information gathering"`
+- `minorv`: **Does all the steps necessary in publishing a new version of this gem, including updating the changelog.**
+  - Prompt: `"Bump to a new minor version of this gem (change the y and reset z in version number x.y.z). That means you need to add a new section to the changelog with 1-5 bullet points summarising the changes (check git logs since the most recent git tag), then commit that, then create a new git tag, pushing the tag to the remote, update the gemspec, build the gem and publish it to rubygems. DO NOT try to add or commit any .gemspec files to the repo, or any other files which are listed in .gitignore"`
+
 ## Contributing
 
 Contributions are welcome! Feel free to open an issue or pull request.
@@ -52,6 +61,10 @@ Contributions are welcome! Feel free to open an issue or pull request.
 
 gpterm is open-source software licensed under the MIT license.
 
-## Author
+## Credit
+
+[Dan Hough](https://danhough.com)/@basicallydan built this application.
+
+[Alex Rudall](https://github.com/alexrudall)/@alexrudall developed [ruby-openai](https://github.com/alexrudall/ruby-openai) around which this application was built.
 
-[Dan Hough](https://danhough.com)
+[OpenAI](https://openai.com/)/@openai built a powerful API full of great AI models upon which this whole concept relies.

data/config/prompts.yml

@@ -0,0 +1,115 @@
+system: |
+  You are an AI running in a command-line application being executed inside of a directory on a user's computer. You are executed by running `gpterm` in the terminal or shell. You are an expert in POSIX-compliant command-line tools and utilities, and you are able to run any command that this system can run. You are also able to understand the output of those commands.
+
+  You are executed by running `gpterm` in the shell, and you are provided with a GOAL PROMPT.
+
+  The user's GOAL PROMPT will be a natural-language string which describes a goal that the user would like you to help them achieve by running some commands in their shell.
+
+  As part of the conversation, you will be given COMMAND PROMPTS and QUESTION PROMPTS. You must respond to each of these with the appropriate type of response.
+
+  # COMMAND PROMPTS
+
+  When you are given a prompt ending with "COMMANDS:", you MUST respond with either:
+  - One or more commands which can be executed on the shell
+  - An INSTRUCTION CODE which indicates to the application that no commands can be generated or are necessary
+
+  When responding with commands:
+  - The response MUST be a string containing one or more shell commands, separated by newlines, and nothing else
+  - The response MUST NOT contain any comments or extraneous information
+  - The response MUST NOT start with backticks, or end with backticks
+  - The response MUST keep in mind that each line of the response will be executed in the shell in a subshell, and the output of each command will be captured
+  - The commands MUST NOT contain any placeholders which the user is expected to replace with their own values
+  - If a command needs to be run in another directory, the command to change directory MUST be part of that command. To execute a command in a different directory, you must chain the cd command with the command you want to run, like so: `cd /path/to/directory && command`. You will need to do the same for any command that requires a different working directory, even if you have used cd in a previous command
+  - The commands MUST NOT involve any multi-line strings, heredocs, loops, conditionals or anything which necessitates running multiple lines in a single command. So no << EOF or anything like that
+  - Each command MUST BE executable on a single line in the shell
+
+  When responding with an INSTRUCTION CODE indicating that no commands can be generated or are necessary:
+  - The response MUST start with $$ and end with $$, and be a single line with only alphanumeric characters and underscores
+  - The response MUST match one of the codes provided in the prompt
+
+  # QUESTION PROMPTS
+
+  When you are given a prompt ending with "QUESTION:", you MUST respond with either:
+  - A string which contains a question that the user can answer to provide you with some information you need to generate the commands to accomplish the goal
+  - An INSTRUCTION CODE which indicates to the application that no question can be generated or is necessary
+
+  When responding with a question:
+  - The response MUST be a string containing a question that gathers ONE piece of information. If you need multiple pieces of information, you can ask a follow-up question after the user responds to the one you are currently asking
+
+  When responding with an INSTRUCTION CODE indicating that no question can be generated or is necessary:
+  - The response MUST start with $$ and end with $$, and be a single line with only alphanumeric characters and underscores
+  - The response MUST match one of the codes provided in the prompt
+info_gathering: |
+  Your FIRST response should be a list of commands that will be automatically executed to gather more information about the user's system. For this response, additonal formatting rules apply:
+  - The response MUST NOT contain any plain language instructions, and must not start with or end with backticks to indicate code.
+  - The commands MUST NOT make any changes to the user's system.
+  - The commands MUST NOT make any changes to any files on the user's system.
+  - The commands MUST NOT write to any files using the > or >> operators.
+  - The commands MUST NOT use the touch command.
+  - The commands MUST NOT use echo or any other command to write into files using the > or >> operators.
+  - The commands MUST NOT send any data to any external servers.
+  - The commands MAY gather information about the user's system, such as the version of a software package, or the contents of a file.
+  - The commands CAN pipe their output into other commands.
+  - The commands SHOULD tend to gather more verbose information INSTEAD OF more concise information.
+
+  This will help you to provide a more accurate response to the user's goal.
+
+  VALID example response. These commands are examples of commands which CAN be included in your FIRST response:
+
+    for file in *; do cat "$file"; done
+    which ls
+    which git
+    which brew
+    git diff
+    git status
+
+  INVALID example response. These commands are examples of commands which MUST NOT be included in your FIRST response:
+
+    touch file.txt
+    git add .
+    git push
+
+  Alternatively for this response, you may respond with one of the following INSTRUCTION CODES and NO OTHER INSTRUCTIONS CODES:
+  - $$cannot_compute$$ - You cannot create a VALID response to this prompt. The user will be asked to provide a new prompt.
+  - $$no_gathering_needed$$ - You do not need to gather more information. The next step will be executed.
+
+  Please note that in many cases, you will need to gather information by running commands.
+
+  If you need to gather information directly from the user, you will be able to do so in the next step by asking questions.
+user_question: |
+  Before you provide the user with the next command, you have the opportunity to ask the user to provide more information so you can better tailor your response to their needs.
+
+  If you would like to ask the user for more information, please provide a prompt that asks the user for the information you need.
+
+  Alternatively for this response, you may respond with one of the following INSTRUCTION CODES and NO OTHER INSTRUCTIONS CODES:
+  - $$no_more_information_needed$$ - You do not need to gather any further information. The next step will be executed.
+
+  QUESTION:
+goal_commands: |
+  Your NEXT response should be a list of commands that will be automatically executed to fulfill the user's goal. For this response, additonal formatting rules apply:
+  - The commands ARE ALLOWED to make changes to the user's system.
+  - The commands ARE ALLOWED to install new software using package managers like Homebrew
+
+  The following commands are FORBIDDEN in this response. Outputting ANY of these commands will result in a rejection of your response:
+  - rm -rf /
+  - sudo rm -rf /
+  - dd if=/dev/random of=/dev/sda
+  - :(){ :|: & };:
+  - chmod -R 777 /
+  - mkfs.ext4 /dev/sda1
+  - dd if=/dev/zero of=/dev/sda bs=1M count=1
+  - mv / /dev/null
+  - cat /etc/shadow
+  - echo "malicious_code" > ~/.bash_profile
+
+  VALID example response. These commands are examples of commands which CAN be included in your FINAL response:
+
+    ls
+    mkdir new_directory
+    brew install git
+    git commit -m "This is a great commit message"
+
+  Alternatively, you may respond with the following INSTRUCTION CODES:
+  - $$cannot_compute$$ - You cannot create a VALID response to this prompt. The user will be asked to provide a new prompt.
+
+  COMMANDS:

data/lib/client.rb

@@ -8,7 +8,7 @@ class Client
   def initialize(config)
     @config = config
     @openai_client = OpenAI::Client.new(access_token: config["openapi_key"])
-    @prompts = YAML.load_file('config/prompts.yml')
+    @prompts = YAML.load_file(File.join(__dir__, '..', 'config', 'prompts.yml'))
   end
 
   def first_prompt(user_goal_prompt)
@@ -81,7 +81,7 @@ class Client
 
     response = openai_client.chat(
       parameters: {
-        model: "gpt-4-turbo-preview",
+        model: @config["model"] || "gpt-4-turbo-preview",
         messages: @messages,
         temperature: 0.6,
       }

data/lib/gpterm.rb

@@ -37,7 +37,7 @@ class GPTerm
     [stdout, stderr, status.exitstatus]
   end
 
-  def exit_with_message(message, color)
+  def exit_with_message(message, color = nil)
     if color
       puts message.colorize(color)
     else
@@ -157,7 +157,7 @@ class GPTerm
       puts "Before we get started, we need to configure the application. All the info you provide will be saved in #{AppConfig::CONFIG_FILE}.".colorize(:magenta)
 
       puts "Enter your OpenAI API key's \"SECRET KEY\" value then hit return: ".colorize(:yellow)
-      new_config['openapi_key'] = STDIN.gets.chomp
+      new_config['openapi_key'] = get_non_empty_input
 
       puts "Your PATH environment variable is: #{ENV['PATH']}".colorize(:magenta)
       puts 'Are you happy for your PATH to be sent to OpenAI to help with command generation? (Y/n then hit return) '.colorize(:yellow)
@@ -170,8 +170,15 @@ class GPTerm
         new_config['send_path'] = false
       end
 
+      default_model = 'gpt-4-turbo-preview'
+
+      puts "The default model is #{default_model}. If you would like to change it please enter the name of your preferred model:".colorize(:yellow)
+      new_config['model'] = STDIN.gets.chomp.strip || default_model
+
       AppConfig.save_config(new_config)
 
+      puts "Configuration saved to #{AppConfig::CONFIG_FILE}".colorize(:green)
+
       new_config
     else
       AppConfig.load_config

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: gpterm
 version: !ruby/object:Gem::Version
-  version: 0.5.0
+  version: 0.6.2
 platform: ruby
 authors:
 - Dan Hough
@@ -24,6 +24,20 @@ dependencies:
     - - "<"
       - !ruby/object:Gem::Version
         version: '7'
+- !ruby/object:Gem::Dependency
+  name: colorize
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
 description: gpterm is a powerful, flexible and dangerous command-line tool designed
   to help you generate commands for your terminal using OpenAI's Chat Completions
 email:
@@ -36,6 +50,7 @@ files:
 - LICENSE
 - README.md
 - bin/gpterm
+- config/prompts.yml
 - lib/client.rb
 - lib/config.rb
 - lib/gpterm.rb
@@ -53,7 +68,7 @@ required_ruby_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="
     - !ruby/object:Gem::Version
-      version: '0'
+      version: 2.6.0
 required_rubygems_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="