broadcast-optimize

Broadcast Optimize

Use this skill when you want to polish a draft before calling broadcast-publish-*, or rank post-publish metrics.

Agent Instructions

• Optimizers are optional. The publishers do not require them.
• Configure the LLM via env: BROADCAST_KIT_LLM_PROVIDER ∈ {openai, anthropic, ollama} + matching API key. See docs/optimizers.md for the full env set.
• Draft input is a YAML/JSON file with at minimum platform (x|xhs|douyin) and body.
• The reviewer's publish_threshold is -5 in the bundled rubrics; a composite score below threshold means do NOT publish without revising.
• Treat content_brain.publish_decision == "hold" as a hard stop.
• engagement_score is pure math; no LLM, no env. HeavyRanker weights apply to X-style records; weighted composite scoring applies to any platform with engagement metrics.

Command Template

Structured diagnostic (dbskill-style):

broadcast-kit optimize content-brain --draft draft.yaml

Severity-weighted reviewer (10 dimensions per bundled rubric):

broadcast-kit optimize reviewer --draft draft.yaml --max-rounds 3

Generate N variants and pick the best:

broadcast-kit optimize variants --draft draft.yaml --n 3

Rank post-publish metrics:

broadcast-kit optimize engagement --metrics state/douyin/work/metrics/default/<date>.jsonl --scorer composite
broadcast-kit optimize engagement --metrics x_posts.jsonl --scorer heavy

See docs/optimizers.md for return shapes, Python-level usage, custom rubric format, and the source of every weight.