Cost Estimator (Pre-flight)
Before user commits 30 min + $2.40 to /article, show them the estimate.
When to invoke
/cost-estimate— defaults to listicle 6000w/cost-estimate listicle 6000/cost-estimate --brief workspace/abc/brief.json- User asks: "how much for a pillar page?", "what does an article cost?", "is this gonna be over budget?"
- Auto-trigger via L1 SKILL.md before phase-research starts (mandatory pre-flight)
How to invoke
bash# By format + word count python -m scripts._core.cost_estimator --format listicle --words 6000 # From an existing brief.json python -m scripts._core.cost_estimator --brief workspace/abc123/brief.json --json # With cheaper config python -m scripts._core.cost_estimator \ --format how-to-guide --words 4500 \ --images 2 --image-quality medium --batch # Machine-readable python -m scripts._core.cost_estimator --format pillar-page --json
Output (human-readable)
━━━ /article cost estimate ━━━━━━━━━━━━━━━━━━━━━━━━━ Format: listicle (target 6000 words) Images: 6 × high-quality 1024² (batch) ← count = image_policy.DEFAULT_IMAGE_COUNT (format rows may set fewer) Drafter / pipeline model: claude-opus-4-7 / claude-opus-4-7 Cache hit rate assumed: 30% Breakdown: Research (Tavily ×8 adv search + 12 extracts): $0.1664 Drafting (6000w via claude-opus-4-7, 21,000 in + 8,400 out + 9,000 cached): $0.3360 Pipeline stages (13 stages, 124,000 in + 13,500 out via claude-opus-4-7): $0.6469 Images (4 × gpt-image-2 high 1024² batch): $0.4220 Overhead buffer (10%): $0.1571 TOTAL: $1.73 80% range: $1.30 – $2.25 ✓ Within budget.
Typical costs by format (rough)
| Format | Words | Est. cost (high+batch) |
|---|---|---|
| news-analysis | 1500 | ~$1.40 |
| definition | 3500 | ~$2.00 |
| how-to-guide | 4500 | ~$2.40 |
| product-review | 4500 | ~$2.40 |
| comparison | 4500 | ~$2.50 |
| case-study | 5500 | ~$3.10 |
| listicle | 6000 | ~$1.73 |
| pillar-page | 6500 | ~$3.40 |
Levers to lower cost
- Use Batch API for images — 50% off (default; toggle with
--no-batch) - Lower image quality —
mediumcuts image cost 4×,lowcuts 35× - Fewer images —
--images 2instead of 4 - Cheaper drafter model —
--drafter-model claude-sonnet-4-6cuts ~60% - Cheaper pipeline model —
--pipeline-model claude-haiku-4-5cuts ~80% - Smaller word count — costs scale ~linearly with words
Budget check semantics
approved — Within per-article + daily limits
needs_approval — Above 50% of per-article budget OR 80% daily
blocked — Exceeds per-article OR would push daily over limit
If blocked, suggest lowering image quality OR shortening word count OR raising
budget in ~/.xuanran-seo/config.yaml.
What this skill does NOT do
- ❌ Actually run /article (that's L1 SKILL.md)
- ❌ Calibrate estimates from past runs (future closed-loop work — Step P2-真)
- ❌ Lock in pricing (estimates can be ±25%)
Pre-flight integration
L1 SKILL.md should call this BEFORE phase-research kicks off:
bashpython -m scripts._core.cost_estimator --brief workspace/$TASK/brief.json --json
If total_usd > per_article_limit: pause and ask user before continuing.
