recon - Gather background information to prompt LLMs

LLMs like Claude and ChatGPT can be extremely useful. But gathering up all the background information they need to provide appropriate answers can be painful. That's where recon comes in.

Running a simple command like recon --files ./docs --prompt "What are the addresses for all of our internal servers?" | llm beats manually looking through everything in the docs folder.

Installation

• global: npm install -g @joshuacc/recon - This will add a recon command for you to use anywhere on your system
• local: npm install --save-dev @joshuacc/recon - This will add a recon command that you can use within npm scripts or by referencing ./node_modules/.bin/recon

Basic Usage

The recon command gathers information from various sources and combines it into a single prompt for use with an LLM. You can paste it to your clipboard for use with ChatGPT, write it to a file, or send it via stdout to combine it with other tools.

Output

The recon command will output a text prompt in one of three ways:

• Clipboard: If the --clipboard flag is provided, the prompt will be copied to your clipboard.
• File: If the --output flag is provided, the prompt will be written to the specified file.
• Stdout: Recon will automatically detect if it is being piped to another command and will output the prompt to stdout in that case. This can be useful for combining with other tools, like llm. Example: recon --files ./docs --prompt "How do I debug docker problems for this project?" | llm.

Specifying what you want from the LLM (prompt)

Example: recon --files ./docs --prompt "What are the addresses for all of our internal servers?"

In addition to the background information that recon gathers, you can optionally provide a prompt for the LLM which will be included in the final text.

Gathering files

Example: recon --files ./docs

You can specify files for recon to gather in several ways.

• Directories: You can provide the path to a directory, and recon will recursively search through all the files in that directory.

• Globs: Globs are like wildcards, and can be used to specify multiple files or directories. For example, recon --files ./docs/**/*.md will gather information from all markdown files in the docs directory and its subdirectories.

• Files: You can specify a single file, and recon will gather information from that file. For example, recon --files ./docs/server_info.md.

• Comma Separated: To provide multiple sources, you can provide them as a comma-separated list. For example, recon --files ./docs,./src/**/*.tsx.

• Exclusions: You can exclude files or directories by prefixing them with a !. For example, recon --files ./docs,./src/**/*.tsx,!./src/secret.tsx.

NOTE: recon excludes some files and directories by default. These are .git, node_modules, and others. To see the complete list, see src/defaultExclusions.ts.

Gathering urls

Example: recon --urls https://example.com

To gather information from a website, use the --urls option followed by the URL. recon will send a GET request to the URL and place the response into the prompt.

Gather from multiple sources

Example: recon --files ./docs --urls https://example.com

You can gather information from multiple sources by providing multiple --files and --urls options.

Creating recon commands as shortcuts

If you have a specific set of sources you want to gather information from frequently, you can define your own recon commands.

Create a .recon.config.mjs file in the root of your project or in your home directory. For example:

export default {
  commands: {
    // The key is the name of the command
    docs: {
      gather: {
        files: ['./docs'],
        urls: ['https://example.com/some-more-docs']
      }
    }
  }
};

You can then run your command with recon docs and pass in any additional arguments, such as a prompt with custom directions. For example: recon docs --prompt "How do I reset my local db?". If you pass in more --files or --urls options, they will be merged with the ones defined in the command.

Config merging

If you have a .recon.config.mjs file in your home directory and one in the root of your project, the two will be merged. The project-level config will take precedence over the home-level config. Though it will print a warning if there are conflicting command definitions.

Advanced usage

Creating your own recon agents

You can create your own recon agents by implementing the ReconAgent interface.

Configuration format

// This is a hypothetical recon agent for accessing database information
import myDbAgent from './myDbAgent';

export default {
  agents: [myDbAgent]
  commands: {
    // The key is the name of the command
    growth: {
      // The default prompt for this command
      prompt: "How are we doing on our growth goals?",
      // The `gather` object must be provided, and must have at least one key
      // specifying an agent to use
      gather: {
        // The default set of files to gather information from
        files: ['./docs/business-plan.md'],
        // The default set of urls to gather information from
        urls: ['https://example.com/'],
        // This will be used by MyDbAgent to gather more information for the prompt
        myDb: {
          query: 'SELECT COUNT(*) FROM users',
        },
        // Additional notes to pass along with the prompt. This is the built in `notes` agent
        notes: `REMEMBER: All replies should be in business-speak. The more synergy, the better.`,
      }
    },
  }
};

For full details on the configuration format, see src/config.ts.

Using the function agent

The function agent allows you to run one or more functions to gather information without having to write a full-blown reusable recon agent. This can be useful for quick one-off tasks.

export default {
  commands: {
    growth: {
      gather: {
        function: async () => {
          return [{
            tag: 'stripe-report',
            attrs: {
              date: Intl
                .DateTimeFormat('en-US', { dateStyle: 'short' })
                .format(new Date())
            },
            content: await getStripeReport();
          }];
        }
      }
    },
  }
};

Future features

• Additional built-in agents
• Giving custom agents the ability to plug their own flags into the CLI. e.g. recon --agent:db "SELECT COUNT(*) FROM users"
• Improved documentation