This skill is responsible for only one thing: optimizing one specific article under a fixed scorer.
What it is suitable for:
- Scoring a single article
- Revising the draft based on the improvement suggestions returned by the score
- Doing multiple rounds of version iteration
- Comparing the viral score of different versions
What it is not responsible for:
- Adjusting the scorer itself
- Calibrating the entire scoring system
- Batch sample evaluation
If the goal is to change the scoring Prompt rather than the article, you should switch to score-optimizer。
Working Scope
Applicable scenarios:
- The user says, "Optimize this article to a higher score"
- The user says, "Revise the draft according to fixed scoring rules"
- The user says, "Score a single article and rewrite it based on the suggestions"
- The user says, "Run single article optimization loop"
Not applicable scenarios:
- The user says, "Make this scoring system more accurate"
- The user says, "Batch evaluate a set of sample articles"
Required Reads
Read before starting:
Read again when you need to see the implementation:
Workflow
- First confirm that the current task is revising an article, not changing the scorer.
- Read
references/optimize_program.md, and confirm the experiment constraints. - Confirm the target article path.
- Use the single-article scoring script to produce a baseline score and improvement suggestions.
- Each round, only revise the article; do not touch the scoring rules.
- Save the iterative versions, then score them again for comparison.
Commands
Run in the skill root directory:
python3 scripts/run_single_score.py <article_path>
If a chart is needed:
python3 scripts/generate_optimize_chart.py
Guardrails
- By default, do not change
references/score_prompt.md - By default, do not change
scripts/run_single_score.py - Each round, prioritize changing only the article content
- If what the user wants is "training a judge," do not use this skill