Cost-Aware LLM Pipeline Design skill for Claude and Codex

Cost-Aware LLM Pipeline Design as an AI skill. It asks for The feature, Request volume, Monthly budget and 3 more, then does the rest. Free to install in…

Install it once and Claude or Codex runs it for you. When your request matches, your AI loads the skill, asks for anything it is missing, and works through the template itself. Prefer to fill it in by hand? See the full prompt template.

Download the skill (.zip) · View SKILL.md

When it is used

Use when a feature built on a language model needs a pipeline design that controls cost with caching, model routing, token budgets, and eval gates. Do not use for writing the prompts themselves or choosing between model vendors on benchmarks. Related topics: AI, coding.

What it will ask you for

• The feature (longer free text)
• Request volume (free text, e.g. "Requests per day")
• Monthly budget (free text, e.g. "Amount in dollars")
• Quality bar (longer free text)
• Latency target (free text, e.g. "under 3 seconds")
• Levers to consider (choose any: Prompt caching, Response caching, Model routing, Token budgets, Batching, Smaller context, Fallbacks)

How it works

1. Sketch the pipeline steps and estimate input and output tokens for each, so cost per request is a number.
2. Apply caching and context trimming where the same prefix or answer repeats, and estimate the saving.
3. Define routing rules that send easy requests to a cheaper path and escalate on a stated confidence signal.
4. Set a hard token cap per step and a fallback for timeouts and refusals.
5. Define an eval set and release gates on quality and cost, then summarise before and after cost in a table.

What you get

A text diagram of the pipeline, a before and after cost table, the routing rules, and the eval gate definition with thresholds.

What it watches out for

• Routing by prompt length alone, which sends hard short questions to the weak model.
• Caching responses that contain user-specific data and serving them to another user.
• Cutting cost with no eval set, so quality drops unnoticed until users complain.

How to install

Claude.ai and the Claude desktop app

1. Download the skill .zip.
2. Open Settings, then Capabilities, and find Skills.
3. Choose Upload skill and select the .zip.
4. Ask Claude for help with the task. It loads the skill when your request matches.

Claude Code

1. Download the skill .zip.
2. Unzip it into ~/.claude/skills/ (or .claude/skills/ inside a project).
3. Start a new session. Claude Code loads the skill when your request matches.

Codex

1. Download the skill .zip.
2. Unzip it into ~/.agents/skills/ (or .agents/skills/ inside a project).
3. Codex picks it up on its own. Type $ and the skill name to use it, or just ask for the task.

Cursor

1. Download the Cursor rule (.mdc).
2. Save it in .cursor/rules/ inside your project.
3. Cursor's agent applies the rule when your request matches its description.

AGENTS.md (Codex, Cursor, and other coding agents)

1. Copy the AGENTS.md section.
2. Paste it into AGENTS.md at the root of your repository, creating the file if needed.
3. Coding agents that read AGENTS.md follow it when your request fits.