Premature compacting when using ds4-agent

Running on MacStudio Ultra M1 128GB
sudo sysctl iogpu.wired_limit_mb=112640
cd Git/ds4/ds4-main/

Git/ds4/ds4-main/ds4-agent \
-m Git/ds4/gguf/DeepSeek-V4-Flash-IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ8-chat-v2-imatrix-0731.gguf \
--mtp-model Git/ds4/gguf/DeepSeek-V4-Flash-0731-DS4-Quality128-DSpark-support.gguf \
--metal \
-c 32768

iogpu.wired_limit_mb: 112640 -> 112640
ds4: DSpark support model detected: Git/ds4/gguf/DeepSeek-V4-Flash-0731-DS4-Quality128-DSpark-support.gguf (stages=3 block=5 markov_rank=256 tensors=81 missing=0 invalid=0 metadata_errors=0); use --dspark to enable experimental runtime decode
ds4: Metal device Apple M1 Ultra, 128.00 GiB RAM
ds4: Metal 4 tensor API disabled for pre-M5/pre-A19 devices
ds4: drift-patch flags hc_stable=on norm_unify=on kv_raw_f32=off rope_exp2_log2=off math_safe=off tensor_matmul=off
ds4: requesting Metal residency (may take tens of seconds)... done
ds4: warming Metal model views... done
ds4: Metal model views created in 3.118 ms, residency requested in 667.125 ms, warmup 4.001 ms (mapped 82697.67 MiB from offset 5.09 MiB)
ds4: Metal mapped mmaped model as 2 overlapping shared buffers
ds4: metal backend initialized for graph diagnostics
ds4: memory: KV 0.61 GiB (raw 0.36 + compressed 0.25) + buffers 0.25 GiB + resident model 80.76 GiB = 81.62 GiB planned
ds4: memory detail: ctx=32768 prefill_cap=4096 raw_kv_rows=4352 compressed_kv_rows=8194 backend=metal
DwarfStar 🐋 Agent, context 32.8k tokens

* can you read check_indexes.py

Let me read the file check_indexes.py.

🛠️ Reading check_indexes.py 1:500...

COMPACTING tool result would exceed context: summarizing durable task state
Internal ds4-agent context compaction request. This is not a user request.
Write a durable task-state summary of the conversation so far. Preserve only facts that matter for continuing the work:
- user goals, constraints, and preferences
- files inspected or edited
- commands run and important results
- decisions, rejected approaches, known bugs, and pending next steps
- reloadable bulky data with exact paths/ranges/commands when available

Do not invent facts. Do not include generic narration. Do not include raw file contents unless they were essential to a conclusion.
After the summary, stop. Do not continue the user task, do not call tools, and do not output thinking tags or DSML markup.
Output only the compact summary.

Compaction reason: tool result would exceed context
COMPACTING rebuilding context: old=1804 summary+tail=1960 tail=67

ds4-agent: context full after compaction

The status is showing ctx 1.9k/32.8k | idle

Even that I specified context 32.8k tokens, it seems that the context is capped at 1960