research-exec-report

Files (1)

• SKILL.md

SKILL.md (first lines)

# Research Executive Report

## Trigger
`/research-exec-report`

## Workflow

### Step 1: Locate Project
Find `*/outline.yaml` in the current working directory. Read `topic` and `execution.output_dir`.

### Step 2: Read Research Data
- Read `{topic}/fields.yaml` to understand what each field measures and which dimensions
  are most important for decisions.
- Read all JSON files from `output_dir/`.
- Build a mental model of: who the top performers are, what dimensions vary most across
  items, and what trade-offs exist.
- Treat any value containing `[uncertain]` or any field listed in the `uncertain` array
  as unavailable — use it only if no better data exists, and never cite it confidently.

### Step 3: Ask One Focusing Question
Use `ask_followup_question` to ask:

> What decision is this research supporting?
> Examples: "choosing a model to self-host", "writing a team recommendation", "investment due diligence".
> Leave blank for a general executive overview.

Use the answer to frame recommendations. If left blank, write a general overview.

### Step 4: Synthesize and Write `executive_report.md`
Write the report **directly** — no Python script, no intermediate files, just prose.
Synthesize insights from the JSON data rather than repeating raw field values.

#### Report Structure

```
# {topic} — Executive Brief

*{date} | {N} items analyzed*

## Executive Summary

[2–3 paragraphs. What was researched, the current landscape in plain terms,
and the single clearest bottom-line recommendation. Name the winner up front.]

## Key Findings

1. **[Finding title]**: [One or two sentences explaining the finding, citing
   specific numbers or names from the data to back it up.]

2. **[Finding title]**: ...

[3–5 findings total. Each should convey a genuine insight — a trend, a surprise,
a clear winner in some dimension — not just a restatement of a field value.]

## Top Recommendations

### 1. [Item Name] — [One-line pitch, e.g. "Best overall" or "Best for coding tasks"]
[Paragraph: why this item is recommended. Cover its standout strengths, any
significant trade-offs, and the use case it fits best.]

### 2. [Item Name] — [One-line pitch]
[Paragraph: same structure.]

### 3. [Item Name] — [One-line pitch]
[Paragraph: same structure.]

[3–5 recommendations total. Be opinionated. Rank them.]

## At-a-Glance Comparison

| Item | [Key metric] | [Key metric] | [Key metric] | License | Notes |
|------|-------------|-------------|-------------|---------|-------|
| ... | ... | ... | ... | ... | ... |

[Include the top 6–10 items. Choose columns that are most relevant to the decision.
Omit columns where most values are unknown/uncertain.]

## Caveats & What to Watch

[One or two paragraphs. Note any data gaps, limitations of the research, or
factors that could change the recommendations (e.g., upcoming releases,
licensing restrictions, uncertain benchmark scores).]

## Next Steps

- [Concrete action 1]
- [Concrete action 2]
- [Concrete action 3]

[3–5 bullets. Each should be something the reader can actually do.]
```

#### Writing Principles

- **Prose over lists**: Use full sentences and paragraphs. Avoid field:value bullet
  dumps. The one exception is the comparison table and the Next Steps bullets.
- **Opinionated**: Make clear, ranked recommendations. Avoid hedging everything.
- **Silent omission**: Skip uncertain or null values without mentioning they were
  skipped. Only cite data you trust.
- **Decision-first**: Frame everything around the question "what should the reader do?"
- **Length**: Target 250–400 lines. Scannable, not exhaustive.

## Output
`{topic}/executive_report.md`

From lkiesow/zoo-deep-research, MIT.

More from lkiesow/zoo-deep-research