Skip to content
Get packs

Blind Script Scorer Sub-agent

Scores a content script on a rubric in an isolated context, returning strict JSON for internal calibration.

Source Download skill
Repository
XBuilderLAB/cheat-on-content
Stars
7,443
Licence
MIT
Files
1 file

What it does

This is an internal helper that a main agent spawns as a separate sub-agent to score a single video script against a rubric (nine dimensions, each 0-5 with a confidence level and a one-line reason). It is designed to read only the script and the rubric notes, so its scores are not influenced by the conversation history or past performance data. It returns structured JSON that the main agent compares against its own judgement. It is not something an end user should invoke directly.

Great for

  • Developers maintaining the cheat-on-content toolkit who want unbiased calibration scores
  • Anyone studying how to isolate context in multi-agent scoring workflows

Skip it if

  • Ordinary users wanting feedback on a script: the skill refuses to run usefully outside its parent workflow and the README-style guidance says direct invocation is meaningless.
  • Anyone expecting a verified prediction of video performance: the scores are subjective model judgements, and the file itself says they are not ground truth.

Try asking

  • “Spawned by cheat-score via the Task tool with a script path and a rubric_notes path to get a blind 9-dimension score”
  • “Spawned by cheat-bump during calibration to re-score a pool of scripts without exposure to published results”
Step-by-step guide: how to use Blind Script Scorer Sub-agentInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 1 original files, the MIT licence and an attribution note.

    cheat-score-blind.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip cheat-score-blind.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip cheat-score-blind.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Only pass the script path and rubric notes path; extra paths trigger refusals by design.
  • Treat the sub-agent output as one input to compare against, not as the final decision.
  • If the rubric notes contain real performance numbers, the skill flags a contamination warning and lowers confidence; clean the rubric file first.

What's inside

  • SKILL.md

Redistributed unmodified from XBuilderLAB/cheat-on-content under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.