Specific Code

Specific Code is an AI coding agent that runs in your terminal. It helps you understand a codebase, plan changes, edit files, run commands, debug failures, write tests, and keep work moving without leaving your project directory.

It is designed for developers who want an AI teammate that can work directly in a real repository instead of only answering questions in a chat window.

Why Use Specific Code?

Modern development work is full of context switching. You read unfamiliar files, search for call sites, patch code, run typechecks, inspect failures, update docs, and repeat. Specific Code gives an AI agent access to the same local project context you have so it can help with that full loop.

Use it when you want to:

• Explore a new codebase and understand how pieces fit together.
• Turn a bug report or stack trace into a concrete diagnosis.
• Implement small features without manually wiring every file yourself.
• Refactor code while preserving the surrounding project style.
• Generate or improve tests after a change.
• Update documentation, comments, and examples from the actual code.
• Run one-shot coding tasks from scripts, CI-like workflows, or your shell.

Specific Code keeps you in control. You can review what it does, steer it with prompts, choose models and providers, and configure how much permission the agent has.

Install

Install the npm package globally:

npm i -g specific-code@latest

Then run it from any repository:

specific

You can also run a one-off command without installing globally:

npx specific-code@latest run "explain this project and identify the main entrypoints"

The package exposes two binaries:

• specific for the normal command name.
• specific-code for the alternate command name.

Quick Start

1. Open a project directory.

cd path/to/your/project

2. Start the interactive terminal UI.

specific

3. Add a provider credential when prompted, or run provider setup directly.

specific providers login

4. Ask Specific Code to inspect, explain, or change your project.

Find where user authentication is implemented and explain the login flow.

Fix the failing checkout test, then run the relevant test command.

Add input validation to the signup form and update the related tests.

5. Review the changes and keep iterating.

Specific Code works best when you give it a specific goal, relevant constraints, and permission to verify the result with the commands your project already uses.

Common Workflows

Explore A Codebase

Ask broad questions when joining a project or touching an unfamiliar area.

Map the request lifecycle for the billing API and point me to the important files.

What code owns background job retries, and how are failures surfaced?

Fix Bugs

Paste an error, stack trace, failing test output, or reproduction steps.

This test is failing with the output below. Find the root cause and fix it.

Specific Code can search the codebase, inspect the implementation, make a minimal patch, and run the relevant verification command.

Build Features

Describe the desired behavior and any constraints.

Add a CSV export button to the orders page. Match the existing UI style and include tests.

For larger features, ask it to plan first, then implement step by step.

Refactor Safely

Use Specific Code for targeted refactors where preserving behavior matters.

Rename getCwd to getCurrentWorkingDirectory across the package and run typecheck.

Simplify this module without changing public behavior.

Run One-Shot Tasks

Use specific run when you want a non-interactive answer or patch from your shell.

specific run "summarize the project structure"

specific run "add tests for the date formatter and run them"

Continue a previous session when you want follow-up work in the same context:

specific run --continue "now update the README with the new behavior"

Models And Providers

Specific Code supports multiple AI providers through the underlying provider system. Start with:

specific providers login

List configured credentials:

specific providers list

Choose a model for a session:

specific --model provider/model

Or choose a model for one command:

specific run --model provider/model "review the latest changes"

Provider credentials can also be supplied through the environment variables supported by each provider. specific providers list shows active provider environment variables detected in your shell.

Permissions And Control

Specific Code is useful because it can operate on your local files, but you should still treat it like any other powerful developer tool.

• Run it from the repository you want it to work on.
• Review file changes before committing.
• Ask it to run the same typecheck, lint, or test commands you would run manually.
• Use narrower prompts for sensitive or high-risk changes.
• Keep secrets in environment variables or secret managers, not in prompts or committed files.

Interactive sessions let you steer the agent as it works. For automated usage, prefer precise specific run prompts and explicit verification requirements.

AI Training

Specific Code may ask whether you want to opt in to AI training by uploading prompts, agent traces, model outputs, tool calls, and tool outputs to Specific. This data helps improve the product and agent quality.

The default selected answer is yes. You can opt out in the prompt, or disable trace uploads with an environment variable:

SPECIFIC_AI_TRAINING_OPT_IN=0 specific

You can also opt in explicitly:

SPECIFIC_AI_TRAINING_OPT_IN=1 specific

The OPENCODE_AI_TRAINING_OPT_IN environment variable is also supported.

Command Reference

| Command | What it does | | --- | --- | | specific | Start the interactive UI in the current directory. | | specific path/to/project | Start the interactive UI in another directory. | | specific run "fix the failing tests" | Run a one-shot prompt. | | specific run --continue "apply the follow-up cleanup" | Continue the last session. | | specific run --file src/api.ts "review this file" | Attach files to a one-shot prompt. | | specific providers login | Add provider credentials. | | specific providers list | List configured credentials and active provider env vars. | | specific providers logout | Remove provider credentials. | | specific --help | Show CLI help. | | specific run --help | Show one-shot command help. |

Tips For Better Results

• Include the outcome you want, not just the task category.
• Mention the commands that validate success, such as npm run typecheck or bun test.
• Share relevant constraints, such as compatibility requirements or files to avoid.
• Ask for a diagnosis first when the cause is unclear.
• Keep large changes incremental so you can review and redirect easily.

Documentation And Community

Documentation is currently hosted on opencode.ai, and Specific resources are hosted on codewithspecific.com:

• Documentation: https://opencode.ai/docs
• Website: https://codewithspecific.com
• Discord: https://codewithspecific.com/discord
• npm package: https://www.npmjs.com/package/specific-code

License

MIT