modbender Preread Phase — Semantic Chunking
You are performing the preread phase of a sequential reading session. Your job is to analyse the source material, split it into semantic chunks, and prepare the session for reading.
Inputs
You will be given:
SESSION_ID— the session identifierSOURCE_FILE— path to the source text fileBASE_DIR— path to the sequential_read skill directory
Procedure
1. Read the Source File
Read the entire source file using the read tool. Understand its structure, length, and type (novel, essay, article, poetry, non-fiction, etc.).
2. Decide Chunking Strategy
Choose an appropriate chunking approach based on the material:
| Material Type | Default Approach |
|---|---|
| Novel with chapters | One chunk per chapter (combine very short chapters) |
| Novel without chapters | Scene breaks or ~300-400 line segments at paragraph boundaries |
| Essay/article | Section by section, or argument-by-argument |
| Non-fiction with chapters | One chunk per chapter |
| Poetry | Stanza groups or poem-by-poem in a collection |
| Short story | 2-5 chunks based on narrative movement |
Key constraints:
- Chunks must contain the complete original text — no abridging, no summarising
- Chunks must be in order
- Each chunk should be a meaningful unit of reading (not mid-sentence or mid-paragraph)
- Consider reading rhythm: cliffhanger endings might pair better with the next section
3. Determine if Structural Fallback is Needed
If the source file is too large to fit entirely in your context (roughly >200,000 characters), use the structural chunker as a fallback:
python3 {BASE_DIR}/scripts/chunk_manager.py structural-chunk {SESSION_ID} {SOURCE_FILE}
This will automatically split on chapter markers, scene breaks, and paragraph boundaries (targeting 300-400 lines per chunk) and save all chunks to the session.
Skip to step 5 if using the structural fallback.
4. Save Chunks (Semantic Approach)
For each chunk you identify:
Write the chunk text to a temp file:
# Write chunk text to temp file (use the read tool to get the text, write tool to save it)Save via chunk_manager:
python3 {BASE_DIR}/scripts/chunk_manager.py save {SESSION_ID} {N} \ --text-file /tmp/chunk_N.md \ --meta '{"tone":"<tone>","intensity":"<low|medium|high>","themes":["<theme1>","<theme2>"],"adjacent_relationship":"<relationship to previous/next chunk>"}'Metadata fields:
- tone: descriptive string (e.g., "melancholic", "argumentative", "playful", "tense")
- intensity: low / medium / high — how emotionally or intellectually demanding
- themes: array of 1-4 thematic tags
- adjacent_relationship: how this chunk connects to neighbours (e.g., "resolution of previous chapter's cliffhanger", "new argument thread", "continuation of scene")
5. Update Session Status
python3 {BASE_DIR}/scripts/session_manager.py update {SESSION_ID} --set status=chunked
6. Initialise Reading State
python3 {BASE_DIR}/scripts/state_manager.py init {SESSION_ID}
7. Write Chunking Notes
Write a brief chunking_notes.md file to the session directory explaining:
- How many chunks you created and why
- What chunking approach you chose and the rationale
- Any notable structural features of the source (unusual formatting, embedded documents, etc.)
Save it to: {WORKSPACE}/memory/sequential_read/{SESSION_ID}/chunking_notes.md
Done
After completing all steps, the preread phase is complete. The reading phase can begin.