agentsclimarketplace
Sign in

Deep research synthesis

Skill Chykalophia/cro-research-reporting/skills/deep-research-synthesis

This skill should be used for thorough, multi-source research that must be credible and well-sourced, and for turning that research into a synthesis. Trigger phrases: "deep research", "comprehensive research", "research report", "investigate X thoroughly", "find best practices with sources", "market/competitive research", "synthesize these findings", "fact-check this", "is this claim actually true?". Use it whenever a question needs rigorous, cited research rather than a quick answer.From its SKILL.md

Install
npx -y skills add Chykalophia/cro-research-reporting --skill deep-research-synthesis

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

4.0 KB, 774 tokens by cl100k_base, as published. Nobody here has run it

Deep Research & Synthesis

Produce research a skeptical expert would trust: every load-bearing claim sourced, reliability labeled, conflicts surfaced, and the synthesis honest about what's known vs. estimated.

Principles

  1. Scope before searching. Pin down the question, audience, deliverable, and depth. For research, start an initial recon search immediately rather than gating everything on a clarifying question.
  2. Ground in primary/first-party data first. If the subject owns data (analytics, a repo, internal docs), pull it before web opinion.
  3. Fan out, don't trickle. Decompose into independent streams; research in parallel (parallel subagents when available).
  4. Demand provenance. Every statistic gets a source name, URL, and year.
  5. Tier reliability. Primary/academic/first-party vs directional/vendor; use directional figures for direction only (references/source-tiering.md).
  6. Verify the load-bearing claims against primary sources; drop or label what doesn't survive.
  7. Reflect and stop deliberately. After each pass: what did I find, what's missing, is coverage enough? Stop when saturated (references/research-rigor.md).
  8. Surface conflicts; never fabricate.

Workflow

  1. Scope & clarify, then write a one-paragraph research brief (constraints stated, unknowns marked open).
  2. Recon the subject and any first-party data to sharpen the stream plan.
  3. Plan 3–5 independent streams mapped to eventual report sections.
  4. Fan out with precise, self-contained briefs (references/subagent-briefs.md); pull interactive/first-party data yourself.
  5. Collect with discipline (URLs + years; concrete examples; note source type).
  6. Tier, score, and verify (credibility scoring + primary-over-secondary in references/research-rigor.md); re-fetch primaries for load-bearing stats.
  7. Coverage check, then synthesize — lead with the conclusion, weave inline citations, show conflicts, end with a tiered source list. Hand to report-information-design for a polished deliverable.

Reference map

  • references/research-playbook.md — recon→plan→fan-out→verify→synthesize loop in detail.
  • references/research-rigor.md — query expansion, the reflection/gap loop, stop conditions, credibility scoring, self-contained briefs + verbatim-return contract (from gpt-researcher / STORM / open_deep_research).
  • references/source-tiering.md — the reliability rubric, confidence phrasing, handling conflicts.
  • references/subagent-briefs.md — the parallel-research brief template + worked examples.

Engagement context

At the start of a task, check for the engagement brief (default ./cro-engagement.md, created by the getting-started skill / /kickoff). If it exists, read it first and tailor scope, audience, and depth to it. If it's missing and the work is more than a one-off, offer to capture one via getting-started.

Related skills

report-information-design (turn the synthesis into a report), ecommerce-cro-audit (a frequent consumer of this skill's evidence). Agents: deep-researcher runs a single stream; cro-verifier adversarially checks the result.

Honesty guardrails

Label vendor/self-interested figures directional; distinguish "verified against primary" from "widely cited but untraced"; prefer recent sources for fast-moving topics; include a short source-reliability appendix so the reader can defend the work.

What ships with it: 4 files

11.6 KB alongside SKILL.md

references/

Gives 0 of the 12 instructions most web research skills give in 774 tokens

Counted across 292 of the 300 authors here whose files we hold, read 2026-09-06

  • Use web_search_exa for current information and broad discoveryin 22 of 292, across 8 files
  • Cite every claim with a sourcein 21 of 292, across 18 files
  • Configure the Exa MCP server with an API keyin 18 of 292, across 5 files
  • Use get_code_context_exa for code examples and API docsin 16 of 292, across 6 files
  • Verify exact tool names before depending on themin 13 of 292, across 4 files
  • Narrow results with site:, quoted phrase, and intitle: operatorsin 13 of 292, across 4 files
  • Adjust tokensNum lower for snippets, higher for full contextin 13 of 292, across 4 files
  • Break the topic into 3-5 research sub-questionsin 13 of 292
  • Confirm current Exa docs and exposed tool surface before usein 11 of 292, across 2 files
  • Get user confirmation after Phase 1in 10 of 292, across 9 files
  • Prefer primary sources when availablein 10 of 292
  • Verify extracted metadata against original sourcesin 9 of 292, across 5 files

Said here and by no other author read

  • Pin down question, audience, deliverable, and depth before searching
  • Start a recon search immediately
  • Pull first-party data before web opinion
  • Decompose research into independent parallel streams
  • Source every statistic with name, URL, and year
  • Tier sources by reliability

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.