| name | chunked-review |
| description | Implement a plan in logical, ordered sequential chunks. Each chunk is proposed with current/after blocks for approval before applying. |
| triggers | ["chunked build","build in chunks","implement in chunks","chunk by chunk"] |
Chunked Build
Implement a plan incrementally — one logical chunk at a time, with user approval before each change.
Workflow
1. Present the plan
Break the work into ordered chunks. Each chunk is one logical change. Include tests as chunks but group them at the end.
### Plan
Chunk 1 (`worker.py`) — scope table context to assigned tables only
Chunk 2 (`worker.py`, `transformer.py`) — extract shared briefing into ExtractionContext
Chunk 3 (`prompts.py`) — remove duplicate workflow section from worker prompt
Chunk 4 (`tests/test_worker.py`, `tests/test_transformer.py`) — tests for chunks 1-3
Rules:
- List affected file(s) in backticks
- One-line description, not verbose
- Order by dependency — if chunk 3 depends on chunk 1, chunk 1 comes first
- Group test chunks at the end, referencing which implementation chunks they cover
Wait for plan approval before starting.
2. Propose each chunk
For each chunk, show #current / #after blocks per affected file:
### Chunk 1/N: <description>
**Files:** `worker.py`
**#current**
\```python
<exact code as it exists today>
\```
**#after**
\```python
<proposed replacement>
\```
Apply, skip, or modify?
Rules:
- Show enough surrounding context (5-15 lines)
- If a chunk touches multiple files, show each file's
#current / #after separately
- One logical change per chunk — don't bundle unrelated work
3. Wait for response
STOP and wait. Do NOT proceed until the user responds:
- Apply / "yes" / "lgtm" → apply the change, propose the next chunk
- Skip / "no" → skip it, propose the next chunk
- Modify → user gives feedback, re-propose with revisions, wait again
- Stop → stop the build
4. After all chunks
Run tests using the /run-tests skill. Show summary:
Applied: #1, #2, #4
Skipped: #3 (user: "not needed")
Modified: #2 (adjusted per feedback)
Tests: ✓ all passing
Important
- If a chunk depends on an earlier one, note it: "Depends on: #1"
- If the user says "apply all remaining" — apply everything left without further review