Prompt And Pipeline Optimizer

• observation versus inference separation
• time and action granularity
• evidence traceability
• controlled improvement suggestions

Output Requirements

• Separate diagnosis into chunk prompt, report prompt, and pipeline.
• When proposing rewrites, return copy-ready prompt text.
• Explain which symptom each rewrite is intended to fix.
• Include a compact experiment plan with success criteria.

Guardrails

• Do not assume prompts are the only cause of weak outputs.
• Keep rewrites specific to work-video analysis, not generic AI-writing advice.
• Prefer small, testable revisions over a full rewrite unless the current prompt is structurally broken.40:["$","$L48",null,{"content":"$49","frontMatter":{"name":"prompt-and-pipeline-optimizer","description":"Improve chunk-caption prompts, report-synthesis prompts, and first-pass video-analysis workflow design. Use when optimizing prompt quality, tuning system prompts, reducing overclaiming, improving chunk captions, improving report synthesis, rerun planning, pipeline diagnosis, プロンプト改善, キャプション改善, or report-generation experiments.","argument-hint":"Provide the current chunk-caption prompt, the final report prompt, representative weak outputs, and whether you want prompt rewrites, failure diagnosis, or an experiment plan."}}]