Doubleword Skill

The doubleword skill helps AI agents such as Claude Code, Cursor, Codex, Hermes and OpenClaw run cost-aware LLM inference on Doubleword.ai. It gives agents a practical operating procedure for choosing between realtime, async, and 24-hour batch inference, validating local JSONL payloads, submitting jobs through the dw CLI, and retrieving results.

Doubleword.ai provides high-performance inference for generation, extraction, classification, OCR, embeddings, evals, and large data-processing workflows.

Installation

Install with the Skills CLI:

npx skills add https://github.com/doublewordai/doubleword-skill

Or clone into your agent skills directory:

git clone https://github.com/doublewordai/doubleword-skill

Updating

npx skills update doubleword

Or, if you installed via git:

git pull

Some agents, like Hermes and OpenClaw, may use a different skills path than the default Skills CLI targets. If the one-liner does not land where your harness loads skills, clone or copy this repository into that harness skills directory instead.

Get a Doubleword.ai Account

You need a Doubleword.ai account and API key to use this skill.

If you do not already have an account, sign up at doubleword.ai. Signup is free and includes free tokens so you can try the platform before adding paid usage.

After creating an account, create or copy an API key and expose it to your agent environment as:

export DOUBLEWORD_API_KEY="your_api_key_here"

The skill supports two authentication readiness paths. With browser login, it verifies identity and organization with:

dw whoami

For headless/API-key login, dw whoami may fail because API-key credentials do not include admin API access. In that mode, the skill validates JSONL payloads locally and uses a non-interactive, token-limited realtime request, such as MODEL="${MODEL:-openai/gpt-oss-20b}"; dw realtime "$MODEL" "Reply with OK." --temperature 0 --max-tokens 2 --no-stream, as the inference-authentication probe before uploading or submitting jobs.

What This Skill Does

• Routes small interactive prompts to realtime inference.
• Sends same-session background workflows to async inference.
• Uses 24-hour batch inference for large jobs when cost savings matter more than immediate completion.
• Validates JSONL payloads locally before upload.
• Encourages deterministic request IDs so batch results can be joined back to source data.
• Avoids long idle polling loops that are unsafe for agents like Hermes.
• Points agents to detailed command, model, and pricing references only when needed.

Repository Layout

SKILL.md
references/
  cli-recipes.md
  models-and-pricing.md

This repository is packaged as a single skill, so SKILL.md lives at the repository root.

The skill uses progressive disclosure:

• Level 0: skill list metadata from SKILL.md