claude_console

KO | EN

gitlite — search

#javascript #hacktoberfest #react #typescript #go #rust #deep-learning #java #nodejs #windows #awesome #cli #pytorch #ios

(★ 20)

No description available.

claude_console Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Claude Console

Pair with [Claude Code](https://claude.ai/claude-code) in your Rails console. No quoting needed.

```ruby
claude help me debug this user's deliverables
claude write a scope to find users with expired trials
claude why does User.find(42) have no campaigns
```

Claude runs Ruby code directly in your console session and sees the results — just like pair programming.

## Installation

Add to your Gemfile:

```ruby
gem "claude_console", group: :development
```

Requires [Claude Code](https://claude.ai/claude-code) installed and authenticated on the machine.

## How It Works

Open `rails console` and type `claude` followed by your request:

```
irb(main):001> claude find the user with email "foo@bar.com" and show their campaigns
Asking Claude...

Here's what I found:

>> Running:
   user = User.find_by(email: "foo@bar.com")

=> #<User id: 42, email: "foo@bar.com">

>> Running:
   pp user.campaigns.select(:id, :name, :status)

[#<Campaign id: 1, name: "Summer 2026", status: "active">,
 #<Campaign id: 2, name: "Fall Launch", status: "draft">]
```

Claude sees your console history for context, runs code in your session, and can iterate on results.

## Safety Gate

Before any Ruby block Claude proposes is `eval`'d, it's parsed and statically analyzed (via [Prism](https://github.com/ruby/prism)). When a flagged operation is detected — `destroy_all`, `update_all`, `delete_all`, `FileUtils.rm_rf`, raw destructive SQL, shell-outs, mailer dispatches, etc. — you're prompted before it runs:

```
⚠️  Gate flagged 1 operation:
   line 1: destroy_all  (always_confirm)
      User.where(plan: "free").destroy_all
   environment: production

[y] run   [n] skip block   [e] edit   [a] abort response
>
```

- **`y`** — run the block
- **`n`** — skip this block, continue with the rest of Claude's response
- **`e`** — open the code in `$EDITOR` for a quick edit before running
- **`a`** — stop processing the entire response

Findings come from a real Prism AST walk, not regex on source — so `array.delete(x)` doesn't trip the file-destructive gate, but `File.delete(path)` does.

### Gate policies

| Policy      | When prompts                                                                 |
|-------------|------------------------------------------------------------------------------|
| `:standard` (default) | Always-destructive / shell / file / raw SQL / mailer: always. Mutating record ops (`save!`, `update`, `create!`, …): only in `production`/`staging`. |
| `:strict`   | All flagged tiers in every environment, including dev/test.                   |
| `:off`      | Never (escape hatch).                                                         |

## Error Feedback

When Claude's code raises during `eval`, the error class, message, and a trimmed backtrace are automatically sent back to Claude in one follow-up turn so it can correct course — without you needing to copy/paste the error. Capped at one follow-up per request to prevent loops. Disable with `ClaudeConsole.error_feedback = false`.

## Session Reset

```
> claude /reset
(session reset — next `claude` call starts fresh)
```

`claude /reset`, `claude /clear`, or `claude --reset` all clear the resumed session so your next call starts with the fresh system prompt and environment context.

## Configuration

```ruby
# config/initializers/claude_console.rb

# Custom system prompt
ClaudeConsole.system_prompt = "You are a database expert..."

# Path to claude binary
ClaudeConsole.cli_path = "/usr/local/bin/claude"

# Pick a specific model — passes --model to the claude CLI
ClaudeConsole.model = "claude-haiku-4-5"

# Safety gate policy: :standard (default), :strict, or :off
ClaudeConsole.gate = :standard

# Whether the gate prompts on ActionMailer dispatches
ClaudeConsole.gate_mailers = true

# Auto-send eval errors back to Claude for correction
ClaudeConsole.error_feedback = true

# Print the assembled prompt + CLI invocation before each call
ClaudeConsole.verbose = false

# Additional env vars to clear (for auth isolation)
ClaudeConsole.clear_env_vars = ["MY_CUSTOM_API_KEY"]
```

## How Auth Works

Claude Console shells out to `claude -p`. It clears env vars like `ANTHROPIC_API_KEY` and `CLAUDE_CONFIG_DIR` from the subprocess so Claude Code uses its own stored credentials — not keys from your Rails environment.

## License

The gem is available as open source under the terms of the [MIT License](LICENSE.txt).

claude_console

claude_console Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

claude_console

claude_console Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README