Research Planning

Produce

plan.json

. It serves as a scoping tool, report form constraint, and execution map simultaneously, guiding the generation of

sub_reports/*.md

and the final

report.md

; it is not a preset of research conclusions.

Behavioral Boundaries

Do:

Identify research goals, boundaries, audiences, purposes, time/region, and assumptions from
```
request.md
```
.
Conduct report format discovery first when necessary, then design evidence-based report forms.
Design lightweight report forms: final draft chapters, structural basis, mandatory tables/graphs/comparison items, and writing constraints.
Break down key questions into executable research dimensions.
Ensure each dimension can independently form a sub-report.
Clarify what each dimension needs to answer, what methods to use, and what materials to produce.
Specify search strategies and sufficiency criteria for each dimension.
Arrange a reasonable execution sequence and real dependencies.

Don't:

Do not write research conclusions.
Do not directly copy final draft chapters as research dimensions.
Do not create separate
```
briefing.md
```
,
```
blueprint.json
```
, or
```
synthesis.md
```
.
Do not conduct literature reviews or expand on facts.
Do not create unnecessary parallel/dependency complexity.

Input

```
{report_dir}/request.md
```

Planning Process

Lock Targets: Extract the judgment that the final report needs to support from
```
request.md
```
.
Define Scope: Clearly write down the objects, time, region, audience/purpose, exclusions, and execution assumptions.
Determine if Format Discovery is Needed: If the user has clearly provided the report structure, or the report type is very standard with obvious constraints, proceed directly to the next step; otherwise, conduct a lightweight format discovery first.
Discover Report Format: Read
```
sn-report-format-discovery
```
when necessary, identify the report type, credible structure sources, mandatory chapters, mandatory elements, and style constraints, then compress them into
```
plan.json.report_shape
```
.
Set Report Form: Provide suggestions for final draft chapters, structural basis, mandatory elements, suitable tables/graphs, and writing constraints.
Extract Questions: Categorize key questions into basic facts, mechanisms/causes, comparisons, impacts, risks, trends, or recommendations.
Design Dimensions: Merge or split dimensions according to the standards of "independently researchable, answerable, and contributable to the final draft".

Define Tasks: For each dimension, clearly write

key_questions

method

search_strategy

expected_output

, and

depends_on

Define Search Sufficiency: Explain which source types need to be covered, at least how many rounds of retrieval, and which key questions must be cross-verified.
Sort Execution: First conduct dimensions related to definition, facts, background, and caliber, then conduct dimensions related to comparison, explanation, impact, and judgment.
Write Completion Criteria: Clarify the conditions under which one can move from dimension-specific research to final draft writing.

Report Format Discovery

Format discovery is not required every time during the planning phase, but should be proactively conducted in the following situations:

The user only says "do a research report" without providing chapter structure.
The report type strongly depends on domain specifications, such as policies, laws, regulations, academic/medical reviews, due diligence, and technical selection.
The target audience or usage scenario significantly affects the structure, such as external reporting, investment decision support, internal selection, and regulatory communication.
You are unsure whether the final draft should prioritize presenting conclusions, timelines, comparison matrices, risk lists, or method descriptions.

You can directly infer

genre

domain

audience

region

from

request.md

; if this information is insufficient to robustly determine the structure, then read

sn-report-format-discovery

When discovering the format, only need to answer 4 things:

Which type of report does this report most closely resemble?
What chapters must this type of report usually have?
What non-chapter elements must appear in the final draft, such as comparison tables, timelines, risk matrices, and method descriptions?
What is the basis for these structural judgments; if no highly credible basis is found, what is the fallback logic?

The results of format discovery should be compressed into

plan.json.report_shape

to constrain subsequent dimension design; do not write an independent format file unless explicitly required by the caller.

Dimension Design Guidelines

A good dimension should meet the following criteria simultaneously:

Can independently produce
```
sub_reports/{id}.md
```
.
Has 2-5 specific
```
key_questions
```
.
Has clear boundaries with other dimensions, and overlaps are explainable.
Has a clear contribution to
```
report.md
```
.
Has clear search entry points and stop conditions.
Does not preset conclusions, only defines what needs to be verified.

Control ordinary research to 3-5 dimensions; complex research can have up to 8 dimensions. When there are too many dimensions, prioritize merging similar questions.

Output

Write to

{report_dir}/plan.json

json

{
  "research_goal": "One-sentence goal",
  "scope": {
    "objects": ["Research objects"],
    "time_range": "Time range",
    "region": "Geographical range",
    "audience_or_use": "Target audience or purpose",
    "exclusions": ["Explicitly excluded items"],
    "assumptions": ["Execution assumptions"]
  },
  "report_shape": {
    "format_basis": [
      {
        "type": "standard_guideline | official_template | real_exemplar | domain_convention | fallback",
        "name": "Name of structural basis",
        "credibility_reason": "Why it is credible",
        "what_extracted": "What structural constraints were extracted from it"
      }
    ],
    "sections": ["Suggested chapters for final draft"],
    "mandatory_elements": ["Mandatory tables, graphs, lists, or comparison items to produce"],
    "style": "Writing caliber and tone constraints"
  },
  "completion_criteria": ["Completion criteria"],
  "change_notes": [],
  "dimensions": [
    {
      "id": "d1",
      "name": "Dimension name",
      "description": "Research scope and boundaries",
      "key_questions": ["Questions that must be answered"],
      "method": "Retrieval, comparison, case analysis, data sorting, timeline restoration, etc.",
      "search_strategy": {
        "source_types": ["Source types to cover"],
        "seed_queries": ["Initial retrieval terms or directions"],
        "freshness_requirement": "Whether latest information is needed and its time range",
        "sufficiency": ["Conditions for judging sufficient search"]
      },
      "expected_output": "Materials that the sub-report of this dimension should provide for final draft writing",
      "depends_on": [],
      "status": "pending"
    }
  ],
  "execution_order": ["d1", "d2"]
}

Field Requirements

```
research_goal
```
: A one-sentence explanation of what judgment the entire research aims to support.
```
scope
```
: Must cover objects, time, region, audience/purpose, exclusions, and current assumptions; explicitly state unknown items.
```
report_shape.format_basis
```
: Explain where the report structure basis comes from. If
```
sn-report-format-discovery
```
was called, retain at least one basis; if not called, also state whether it is "user-specified structure" or "general domain convention".
```
report_shape.sections
```
: Suggestions for the final draft structure, not research dimensions.
```
report_shape.mandatory_elements
```
: List comparison tables, timelines, risk lists, indicator matrices, flowcharts, etc., that the final draft must support.
```
report_shape.style
```
: Explain the report tone, expression of certainty, and handling of uncertainty.
```
completion_criteria
```
: List the conditions that must be met before proceeding to final draft writing.
```
change_notes
```
: Initially an empty array; if dimensions or execution sequences are adjusted during the research process, append the reasons here instead of writing a separate log file.
```
id
```
: Use stable short IDs, such as
```
d1
```
,
```
d2
```
.
```
description
```
: Explain the research scope and what is not included.
```
key_questions
```
: Written as answerable questions, not titles.
```
method
```
: Explain how this dimension should be researched.
```
search_strategy.source_types
```
: List source types to cover, such as official documents, regulatory announcements, academic papers, industry reports, company disclosures, product documents, news reports, community discussions, etc.
```
search_strategy.seed_queries
```
: Provide 3-6 initial retrieval directions, not just a single keyword.
```
search_strategy.freshness_requirement
```
: Explain whether the latest information must be checked; must require latest verification when involving markets, policies, products, companies, prices, laws, and people.
```
search_strategy.sufficiency
```
: Clearly write stop conditions, such as "each key question is supported by at least one material", "core facts come from two independent sources", "stop when there is no new information increment", etc.
```
expected_output
```
: Explain whether the sub-report contributes facts, comparisons, frameworks, judgment basis, risk lists, or scenario analysis.
```
depends_on
```
: Only write real dependencies; leave as an empty array if there are no dependencies.
```
execution_order
```
: Must cover all dimension IDs, and dependencies appear before the dependent items.

Completion Criteria Examples

All
```
key_questions
```
are covered by at least one dimension.
```
report_shape
```
has a clear structural basis, not arbitrarily listed chapters.
All
```
report_shape.mandatory_elements
```
have corresponding dimensions to provide materials.
Each dimension can form an independent sub-report.
Each dimension defines a search strategy and stop conditions.
Key uncertainties are handled by a dedicated dimension, or clearly marked in relevant dimensions.

Common Failures

Treating report chapters as research dimensions.
Each dimension only has a title, no answerable questions.
Significant overlap between dimensions.
No search entry points defined for dimensions, leading to shallow searches during execution.
Execution sequence ignores dependency relationships.
Report structure has no basis, only chapters written based on experience.
Ignoring mandatory elements of the final draft.
Writing conclusions in advance in the plan.

sn-research-planning

NPX Install

Tags

SKILL.md Content (Chinese)