ContactDownload
LanguageEN简中

ModelDial Agent Skill

Ask your agent which model to use

Ask which models lead, how their results have changed, and which model to use for your main or worker agent.

See a live answerInstall the Skill

Current results

See the answer before you install

Checking latest

Published after verification

Agent SkillREST API

Install once, then ask directly

Send the prompt below to a tool that supports Agent Skills, or install from the terminal. The Skill reads public results and requires no account or API key.

Install prompt
Install the ModelDial Skill from https://modeldial.com/modeldial-skill/SKILL.md. Review the linked files before writing to the skills directory.
Prefer a terminal command?
Manual install for Codex, Gemini CLI, Copilot, or OpenCode
bash <(curl -fsSL https://modeldial.com/modeldial-skill/install.sh) --target agents

  1. Install into the directory your agent already uses

    The installer checks the files and updates an existing ModelDial Skill if needed.

  2. Start a new conversation

    Most tools discover Skills when a session starts. Reopen the tool or begin a fresh task after installation.

Verify with one question

Which model and effort level lead ModelDial Radar? What changed in the latest comparable Backend & Testing results? Include the results and source links.

A successful answer includesThe model and effort level, test date or batch, supporting scores, and links to the ModelDial results.

If the data cannot be readThe Agent must stop and say so. It must not scrape the webpage, reuse an older batch, or fill the gap from memory.

Ask for the decision you need

The Skill looks up current rankings, changes between comparable tests, and model suggestions for different agent roles.

  • 01
    Current ranking
    Which configuration leads the current Radar, and why?
  • 02
    Recent regression signal
    Did GPT-5.6 Sol High show a regression signal?
  • 03
    Your configuration
    I use GPT-5.6 Sol with XHigh reasoning effort through the official login route. Is it worth switching for balanced use?
  • 04
    Agent configuration
    How should I configure my main and worker agents from the latest ModelDial results? Start with the balanced profile, then show quality and value alternatives.

Evidence first, recommendation second

The Agent uses published rankings and comparison rules to answer. If there is not enough data, it says what is missing.

  1. Compare only matching batches

    Recent movement is available only when the comparison standard matches.

  2. Start from your configuration

    If the model, effort, or route is missing, the Agent asks before turning the top row into personal advice.

  3. Treat a drop as a signal

    One lower comparable score can justify another check, but not a claim that a provider changed the model.

  4. Keep acceptance with the main agent

    The worker gets bounded execution. Requirements, integration, and final acceptance stay with the main agent.

What the Skill will not claim

It will not rerank the data

Publisher order is the ranking. The Skill does not rebuild it from score or provider names.

It will not declare a hidden model change

A lower comparable score is a signal, not proof that a provider changed or degraded a model.

It will not invent personal fit

Without your exact configuration, the Skill reports the field or asks one short follow-up question.

It will not present estimates as a bill

Reference cost describes the controlled batch, not your invoice or actual usage charge.

It will not claim the pair was tested together

Role profiles use configurations evaluated independently across the latest published axis results, not a paired-agent benchmark.

Agent SkillREST API

Read the same Radar directly

Use compact JSON for current rankings. For audits, choose the Backend & Testing index or the three-capability publication index.

  • GET / JSON

    Current compact ranking

    /api/v1/radar/latest.json

    Default overall order with all three axis scores and weights, plus the compatible backend ranking.

  • GET / JSON

    Agent configuration profiles

    /api/v1/radar/agent-profile.json

    Balanced, quality, and value role assignments from published decision tags; not a paired-agent benchmark.

  • GET / JSON

    Latest comparable changes

    /api/v1/radar/changes.json

    Backend rank, score, time, and recommendation changes after a protocol compatibility check; cost deltas appear only when the pricing snapshot matches.

  • GET / JSON

    Backend & Testing publication index

    /api/v1/radar/index.json

    Backend batch identity, publication time, protocol fields, hashes, and complete snapshot links.

  • GET / JSON

    Capability publication index

    /data/benchmark-snapshots/index.json

    Latest verified batch identity and archive link for each capability and the weighted overall ranking.

  • GET / JSON

    Complete Backend & Testing snapshot

    /data/reference-snapshots/latest.json

    Question-level Backend & Testing results and provenance fields for audit work.

Open Skill, attributed data

Skill instructions use the MIT License. Published Radar data uses CC BY 4.0 and should identify ModelDial and the source batch when practical.

Data licenseEvaluation methodSHA-256