scientific-analysis-architect

What it does

Multi-phase workflow for planning scientific research analyses producing markdown documents with pseudocode. Biology-agnostic design ensures agents request context via user prompts, never inject biological interpretation.

How it works

• You are the orchestrator who coordinates specialists; you do not perform specialist work yourself.
• You delegate all specialist work using the appropriate tool (Task tool or Skill tool) and manage session setup, state, and validation.
• Phase flow:
  • Phase 0: Initialization creates a session directory, validates output, initializes session state, and performs archival compliance checks.
  • Phase 1: Birds-Eye Planning asks the user for dataset and goals, then generates research-structure.md with 3-7 chapters.
  • Phase 2: Subsection Planning assigns expert panels (statistician, mathematician, programmer) per chapter, aggregates recommendations, and outputs chapter{N}-notebook-plans.md.
  • Phase 3: Structure Review analyzes structure and plans, generating structure-review-report.md and awaiting user approval.
  • Phase 4: Plan Review performs per-chapter reviews of pseudocode and data flow, producing notebook-review-report.md and awaiting user approval.
  • Phase 5: Document Generation synthesizes a master strategy overview and creates per-chapter analysis documents with required sections and fenced code blocks, storing outputs in both the main output directory and session backup.
  • Phase 6: Statistical Fact-Checking interviews concerns and applies corrections, regenerating the overview if needed, producing statistical-review-report.md and corrections manifest.
  • Phase 7: Audience phase finalizes documents; outputs include researcher plan and architecture handoffs.
• Tools: Default to Task tool for specialist delegation; use Skill tool for loading domain knowledge when needed.
• Output artifacts include analysis-strategy-overview.md, chapter{N}-notebook-plans.md, and per-chapter analysis markdowns with structured sections and fenced code blocks.

When to use it

• Planning multi-chapter scientific data analyses (RNA-seq, proteomics, imaging)
• Need expert consultation (statistician, mathematician, programmer perspectives)
• Want markdown analysis documents with pseudocode for implementation
• Research project requires 3-7 chapters of analysis

What it can touch

• Creates and reads session-state.json in the session directory
• Reads research-structure.md and chapter{N}-notebook-plans.md as inputs
• Generates analysis-strategy-overview.md and per-chapter analysis markdowns with fenced python blocks
• Writes outputs to {output_dir} and {session_dir} backups

Caveats

• The workflow strictly delegates specialist work; the orchestrator does not implement statistical methods or analysis code directly.
• It performs validation and approvals only as defined in the plan; actual domain-specific content is supplied by specialists.
• Outputs rely on user approvals at gating stages (Phase 3 and Phase 4).