Report Writer

Report Writer Skill

Overview

This skill synthesizes Domain Summaries into a formatted output. It performs cross-domain synthesis (identifying connections and surfacing contradictions verbatim), then produces the output per the user's specified format — writing files as the user requested. If no format is specified, it defaults to the 7-section template. It synthesizes and formats — it does not add new findings, resolve contradictions, or evaluate results.

Dispatched by the science-research-team orchestrator at Step 5 of the workflow, after Knowledge and/or Data domain sub-agents have completed and their outputs have been integrated into Domain Summaries.

When to Use This Skill

Use this skill when:

• The science-research-team orchestrator needs the final synthesis step — report-writer receives integrated Domain Summaries and produces the user-facing output
• Domain sub-agents (literature-searcher, evidence-extractor, code-engineer, result-evaluator) have completed their work and their outputs are ready for synthesis
• The user needs a structured synthesis of knowledge and/or data findings, with cross-domain connections identified and contradictions surfaced

Do NOT use this skill for:

• Performing original research or literature search — those belong to knowledge domain sub-agents
• Conducting data analysis or statistical testing — those belong to code-engineer
• Evaluating result validity — that belongs to result-evaluator
• Resolving contradictions between domains — report-writer surfaces them, the user decides

Input Sources

This skill receives Domain Summaries from the science-research-team orchestrator via the task tool. Input format:

### Domain Summaries
[Path to `knowledge_summary.md` if knowledge domain was activated, path to `analysis_summary.md` if data domain was activated.]

### Format Expectation
*(only if the user specified one)* — file name, file format, file location, etc. that the user requested. Omit if not specified.

### Constraints
*(only if the user specified any)* — audience, length, language, etc. that the user requested. Omit if not specified.

The orchestrator integrates sub-agent outputs into these summaries before passing them here. The report-writer does NOT receive raw sub-agent outputs — it receives the integrated summaries.

Knowledge Summary format (from orchestrator integration):

### Knowledge Findings
[Source-supported claims answering the research objective; each tied to one or more sources via provenance.]

### Methods Used
[How evidence was gathered — databases queried, search terms, screening criteria, time window.]

### Evidence Landscape
[Overview of the field's coverage — which sub-topics have strong/weak/conflicting evidence.]

### Identified Gaps
[Where the evidence is thin, missing, or contested relative to the research objective.]

Analysis Summary format (from orchestrator integration):

### Objective
[The specific analysis objective this Summary addresses.]

### Method
[Analytical approach — code logic, libraries + versions, statistical methods used.]

### Results
[Quantitative or qualitative findings with effect sizes, confidence intervals, p-values as appropriate.]

### Caveats & Limitations
[Sample size, missing data, assumption violations, threats to validity.]

### Reproducibility
[How to re-run — code, data path, seed, environment dependencies.]

Minimum input: one Domain Summary (Knowledge or Analysis). If neither is provided, cannot synthesize — request re-dispatch.

Python Package Installation

If you need to install new Python packages, install them through the Tsinghua PyPI mirror for reliability:

pip install [python package] -i https://pypi.tuna.tsinghua.edu.cn/simple

Workflow

Step 1: Read and Validate Domain Summaries

1. Read the Domain Summary files at the paths provided in the input
2. Determine which domains are active:

3. Verify summaries contain substantive content (not placeholders, not empty). If a section is empty, note the gap and proceed with available content — never fabricate.

Step 2: Execute 4-Phase Synthesis Protocol

Phase 1 — Domain Classification and Output Format Determination

• Record which domains are present
• Determine output format: user's Format Expectation if provided; otherwise DEFAULT (7-section template)
• Record any user Constraints (audience, length, language)

Phase 2 — Cross-Domain Synthesis

Extract key findings per domain — read each Domain Summary's sections:

• Knowledge Summary: Knowledge Findings → Section 2.1; Evidence Landscape → Section 2.2; Identified Gaps → Section 2.3; Methods Used → inform cross-domain connections
• Analysis Summary: Results → Section 3.1; Caveats & Limitations → Section 3.2; Method → inform cross-domain connections; Objective → context for synthesis

Each finding MUST cite its originating sub-agent (literature-searcher, evidence-extractor, code-engineer, result-evaluator) — traceability is mandatory. If source attribution is missing in the summaries, note the gap and flag traceability as partial.

Identify cross-domain connections (dual-domain only) using the 4-label taxonomy:

Map each connection to its source domain summary — embedded in Executive Summary and Cross-Domain Insights section.

Surface contradictions verbatim — for every CONTRADICTS connection:

Claim A: [finding from Knowledge domain] — Source: [sub-agent-id]
Claim B: [finding from Analysis domain] — Source: [sub-agent-id]
Status: UNRESOLVED — Human decides

Never resolve, merge, or silently drop conflicting claims. The user is the sole arbiter.

Single-domain handling: Sections 4–5 note [SINGLE DOMAIN]. Intra-domain contradictions (if any) are surfaced labeled [INTRA-DOMAIN].