Dual-Mode Synergy for BioinfoMCP

Positioning

Bioinfo-V1 (Execution Engine): deterministic execution, protocol orchestration, file retrieval, data cleaning, report generation.
Bioinfo-T1 (Thought/Consultant Engine): retrospective analysis over logs/failures/success parameters, insight extraction, strategy recommendation.

Shared Memory Bridge

Persistent storage root:

BioinfoMCP/shared_knowledge/
  experiment_reports/
  insights/
  pipeline_configs/

Flow:

V1 executes task and writes a structured ExperimentReport.
T1 periodically runs reflection (review_reports) and emits Insight.
T1 can output a PipelineConfiguration from consultation.
V1 loads latest config by task_scope before the next execution.

Minimal Runtime Commands

python -m agent_system.main propose-config \
  --task_scope rnaseq_quant \
  --strategy_name "safe-default-rnaseq-v1" \
  --tools '["fastqc","trim-galore","star","stringtie","multiqc"]' \
  --parameters '{"threads":8,"quality_cutoff":20}' \
  --rationale "Derived from prior successful runs."

python -m agent_system.main execute \
  --task "RNA-seq quantification for sample set A" \
  --task_scope rnaseq_quant \
  --input_manifest '{"r1":"data/SRR3056858_R1.trimmed.fastq","r2":"data/SRR3056858_R2.trimmed.fastq"}'

python -m agent_system.main reflect

python -m agent_system.main consult \
  --task_scope rnaseq_quant \
  --user_goal "Need a robust low-failure first-pass strategy"

Kimi Backend (Optional)

The included adapter (agent_system/llm/kimi_client.py) uses OpenAI-compatible API mode.

Set environment variables:

set KIMI_API_KEY=your_key
set KIMI_BASE_URL=https://api.moonshot.cn/v1
set KIMI_MODEL=kimi-k2-0711-preview

Then call KimiCodeClient.complete(system_prompt, user_prompt) inside T1 planning/reflection routines when you want model-assisted synthesis.