Smithery is now a part of Arcade.dev! Read more in our announcement here
Triggers when reading or writing documents exceeding 20,000 tokens (~80,000 characters or ~15,000 words)...
This skill INTERRUPTS work when encountering or creating documents that exceed 20,000 tokens. Large documents consume excessive context window, reduce response quality, and risk truncation.
| Metric | Threshold | Action |
|---|---|---|
| Characters | > 80,000 | STOP and consult user |
| Words | > 15,000 | STOP and consult user |
| Lines | > 2,000 | STOP and consult user |
| Tokens (estimated) | > 20,000 | STOP and consult user |
Token estimation: ~4 characters per token for English text.
When detecting a document at or approaching these thresholds:
Do not continue reading or writing. Do not attempt to process the full document.
State clearly:
Ask the user how to proceed with these choices:
"I'll read only the specific sections relevant to your request. Tell me what you're looking for."
"I'll read the document in chunks and provide a summary of its structure and key content."
"I can split this into smaller, focused files using the /bmad:core:tools:shard-doc command (splits on level-2 headings by default)."
"I'll read the full document, but be aware this may consume significant context and affect response quality."
"I'll write the document in phases, checking in with you after each major section."
"I'll create a detailed outline for your approval before writing any content."
"I'll create multiple smaller documents organized by topic/section instead of one large file."
"I'll write the structure with placeholders, then we fill in sections incrementally."
Claude detects large document:
STOP — This document is approximately 25,000 tokens (~100,000 characters).
This exceeds the 20,000 token threshold. Processing the full document may:
- Consume excessive context window
- Reduce response quality for follow-up questions
- Risk truncation or incomplete processing
How would you like to proceed?
- Targeted extraction — Tell me what specific information you need
- Summary mode — I'll provide a structural overview
- Shard it — Split into smaller files by section
- Proceed anyway — Read the full document (not recommended)
For BMAD documents specifically:
/bmad:core:tools:shard-doc to split large docs by H2 sections