Agent skill

Evidence Based Reviews

by rampstackco in rampstackco/claude-skills

Produce honest product reviews and buying guides without fabricated first-hand experience, using disclosed evidence tiers: verified specs, owner-experience synthesis at scale, expert triangulation…

MITAuto-check passed

Install Evidence Based Reviews

skills CLI
$ npx skills add rampstackco/claude-skills --skill evidence-based-reviews -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install rampstackco/claude-skills evidence-based-reviews --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/rampstackco/claude-skills.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/evidence-based-reviews .claude/skills/evidence-based-reviews && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
evidence-based-reviews
GitHub stars
940
Token cost
~3.1k tokens
SKILL.md length
1,653 words
Files
4 (incl. references)
Skills in repo
103
Repo updated
First seen
Licence
MIT

At a glance

Produce honest product reviews and buying guides without fabricated first-hand experience, using disclosed evidence tiers: verified specs, owner-experience synthesis at scale, expert triangulation…

  • Works in 7 steps: Inventory the evidence honestly. What… → Set the criteria first. Ordered, before… → Gather per tier. Verify specs at the… → …
  • The user wants to write product reviews
  • SKILL.md covers When to use, When NOT to use, Required inputs and The framework: one rule, four…, plus 10 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Evidence Based Reviews is an agent skill from rampstackco/claude-skills. Produce honest product reviews and buying guides without fabricated first-hand experience, using disclosed evidence tiers: verified specs, owner-experience synthesis at scale, expert triangulation, and hands-on only when true. Use this skill whenever the user wants to write product reviews or buying guides without hands-on access, set up a review methodology, add evidence disclosure to review content, align reviews with Google's reviews system or FTC affiliate disclosure expectations, or decide when Review and…

Its SKILL.md is about 3.1k tokens, which your agent loads only when the skill is triggered. The skill folder holds 4 other files, including reference files (for example `README.md`, `references/evidence-tiers.md` and `references/methodology-block-template.md`).

The repository describes itself as: Stack-agnostic Claude Skills covering the full website lifecycle: brand, design, content, SEO, dev, ops, growth, and research. Build, ship, audit, optimize. The licence is MIT.

When your agent uses it

  • The user wants to write product reviews
  • Buying guides without hands-on access
  • Set up a review methodology
  • Add evidence disclosure to review content

Example prompts

  • “/evidence-based-reviews”

Workflow steps

7 steps, taken from the first numbered list in SKILL.md.

  1. Inventory the evidence honestly. What tiers are actually available for this category? What hands-on exists, if any?
  2. Set the criteria first. Ordered, before any product is examined, so the criteria cannot quietly bend toward a favored pick.
  3. Gather per tier. Verify specs at the source. Pull owner corpora at scale and note sample sizes. Collect named expert sources, including…
  4. Synthesize. Findings per criterion, drawbacks per pick, splits surfaced.
  5. Write the methodology block from what was actually done, not from what sounds good.
  6. Match the markup to the claims. ItemList and Article by default; Review only where tier 4 is true.
  7. Maintain. Date updates, state what changed, upgrade pieces when hands-on arrives, and mark the upgrade.

What it can do on your machine

Read from SKILL.md and the folder at commit 482c9bf. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Evidence Based Reviews loads about 3.1k tokens when it runs, and up to ~4.8k if it reads all its reference files. Until then it costs about 233 tokens; SKILL.md has 1,653 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~233
When it runs · the whole SKILL.md, loaded when a task matches
~3.1k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~4.8k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from rampstackco/claude-skills at commit 482c9bf, republished under its MIT licence (© rampstackco). 1,653 words, ~3,103 tokens.

Download SKILL.mdSave it as .claude/skills/evidence-based-reviews/SKILL.md (or your agent's skills folder). This skill also uses 3 other files; get the full folder from GitHub.
name
evidence-based-reviews
description
Produce honest product reviews and buying guides without fabricated first-hand experience, using disclosed evidence tiers: verified specs, owner-experience synthesis at scale, expert triangulation, and hands-on only when true. Use this skill whenever the user wants to write product reviews or buying guides without hands-on access, set up a review methodology, add evidence disclosure to review content, align reviews with Google's reviews system or FTC affiliate disclosure expectations, or decide when Review and Product schema are honest to use. Triggers on product review, buying guide, best-of list, review methodology, evidence basis, hands-on testing, we tested, affiliate review site, review disclosure, methodology block, original research, reviews system. Also triggers when review content claims testing that did not happen, or when an affiliate site needs a trust mechanism that survives scrutiny.
category
content
catalog_summary
Evidence tiers, methodology disclosure, honest review claims
display_order
13

Evidence-Based Reviews

Produce product reviews and buying guides whose claims match their evidence. The method is evidence synthesis with a disclosed basis, and one rule anchors everything: never claim hands-on time that did not happen.

Most review sites fail this rule quietly. "We tested" appears above content assembled from spec sheets, and readers have learned to smell it. The honest alternative is stronger, not weaker: synthesis of owner experience at scale, verified specifications, and triangulated expert sources is demonstrable original analysis, and disclosing it builds the trust that fabricated testing destroys.


When to use

  • Writing product reviews or buying guides when hands-on access is partial or absent
  • Setting up the review methodology for a new review or affiliate site
  • Adding evidence disclosure to existing review content
  • Auditing review content for claims its evidence cannot support
  • Deciding whether Review or Product schema is honest for a given piece
  • Aligning review content with Google's reviews system and FTC disclosure expectations

When NOT to use

  • General content writing and editing (use content-and-copy; this skill governs the evidence and claims layer of review content specifically)
  • Defining how the brand sounds (use brand-voice)
  • Optimizing the review page itself for search (use seo-onpage)
  • Per-piece editorial briefs (use content-brief-authoring)
  • Instructional how-to content that teaches a procedure or concept to a first-time reader, such as a tutorial or an explainer: a teaching piece of this kind is owned by its writing, its structure, and its search craft, with the evidence discipline riding alongside as a constraint on its claims, so this skill is one input among several rather than the single choice for it

If genuine hands-on testing at scale exists, with instrumentation and protocols, this skill still applies: the tiers do not change, the methodology block simply states tier 4 and the testing protocol becomes the disclosed basis.


Required inputs

  • The product category and the pieces planned (reviews, buying guides, or both)
  • Access to evidence sources: manufacturer pages, retailer review corpora, forums or owner communities, expert publications
  • The brand's disclosure language, if one exists (this skill supplies a template if not)
  • An honest inventory of what hands-on experience genuinely exists, if any

The framework: one rule, four evidence tiers

The anchor rule: never claim hands-on time that did not happen. Every other part of the method exists to make honest content strong enough that the lie is unnecessary.

Tier 1: verified manufacturer specs

Specifications cited to the maker's own page, not to aggregator copies that drift. The verification step is the work: cross-check the spec against the manufacturer's current listing, note the date, and flag discrepancies between sources instead of picking one silently. A spec table built this way is a fact layer; a spec table pasted from another review site is a rumor layer.

When it suffices: factual comparisons, fit and compatibility answers, price-tier groupings. It never supports a quality verdict on its own.

Tier 2: owner-experience synthesis at scale

Retailer review corpora, forum threads, warranty and return patterns, read across sources and summarized honestly. Honest synthesis means:

  • Recurring complaints and recurring praise, not the loudest single anecdote
  • Sample-size candor: "across roughly 1,400 owner reviews" reads differently from "owners say," and the reader deserves the number
  • No cherry-picking: if owners split, the split is the finding
  • Source naming: which corpora, which communities, over what period

When it suffices: durability and reliability verdicts, real-world quirks specs never show, satisfaction patterns by use case. This tier is the workhorse of honest no-hands-on reviewing, and it is genuine original analysis when done at scale.

Tier 3: expert-source triangulation

Named expert sources compared against each other. Triangulation means surfacing agreement and disagreement, not averaging verdicts into mush. When two credible testers reach opposite conclusions, the honest move is to say so and explain the conditions that might account for it. Anonymous "experts agree" is not a tier; it is decoration.

When it suffices: performance claims that require instrumentation the site lacks, technique-dependent judgments, category context.

Tier 4: hands-on, only when true

Stated only when it happened, flagged as such, with the extent quantified: what was done, how much, under what conditions. When hands-on experience arrives later for a piece published on tiers 1 to 3, the piece is upgraded and the update is marked with what changed. Never backfilled to look like it was always hands-on; the upgrade trail is itself a trust signal.


The methodology block

Every review and buying guide carries a short disclosure block naming its evidence basis, placed with the criteria, written for readers rather than lawyers. The fillable template and a worked example live in references/methodology-block-template.md.

The block names: the criteria in the order they were weighted, the tiers actually used (specifically, with sources), what was done hands-on or the words "none claimed," and the update line. A block that claims a tier the piece did not use is the same lie the anchor rule bans, in smaller type.


Structure discipline the method implies

  • Criteria stated and ordered before picks. A verdict the reader cannot trace to a stated criterion is an opinion wearing a methodology costume.
  • A named drawback per recommended pick. Every product has one; a review that finds none has not looked. Consistent placement makes the honesty visible.
  • Update transparency. Dated updates with what changed. Stale best-of content silently rotting is one of the most common failures in the category, and a dated what-changed line beats the field.

Google reviews-system alignment

The guidance asks for demonstrated first-hand experience OR demonstrable original research and analysis. The second branch is this skill's lane, and "original analysis" means concrete work product:

  • The synthesis itself: owner-experience patterns no single source contains
  • Comparative tables built from verified data, structured around the stated criteria
  • Decision frameworks: who each pick fits and who it does not
  • What paraphrased spec sheets are not: rearranging a manufacturer's bullet points is neither experience nor analysis, and ranks accordingly

A site that does tier 2 and 3 work honestly is doing original analysis by the guidance's own definition. The methodology block is how the work shows.


Show full SKILL.md (650 more words)Show less

FTC alignment

Affiliate relationships are disclosed in plain language, placed where the recommendation is, not in a footer. The disclosure travels with the monetized content: near the picks, in body-size text, before or beside the first affiliate link a reader can act on. "Plain language" means a reader who has never heard the word affiliate understands that the site earns a commission and that the price they pay does not change. Euphemisms ("partner links," "support the site") fail the plain-language test.


Schema judgment

Markup is a claim. Apply the same honesty to structured data as to prose:

  • Review and Product markup only when the piece's claims support what the markup asserts. Review markup asserts an evaluative review of an item the reviewer assessed; a tier 1-to-3 buying guide asserting hands-on style review markup is making a machine-readable version of the banned claim.
  • Buying guides without hands-on use guide-shaped markup: ItemList for the picks, Article for the piece.
  • When tier 4 is real and stated, Review markup becomes honest; add it then, per piece, not as a template default.

Workflow

  1. Inventory the evidence honestly. What tiers are actually available for this category? What hands-on exists, if any?
  2. Set the criteria first. Ordered, before any product is examined, so the criteria cannot quietly bend toward a favored pick.
  3. Gather per tier. Verify specs at the source. Pull owner corpora at scale and note sample sizes. Collect named expert sources, including the disagreements.
  4. Synthesize. Findings per criterion, drawbacks per pick, splits surfaced.
  5. Write the methodology block from what was actually done, not from what sounds good.
  6. Match the markup to the claims. ItemList and Article by default; Review only where tier 4 is true.
  7. Maintain. Date updates, state what changed, upgrade pieces when hands-on arrives, and mark the upgrade.

Failure patterns

  • "We tested" inflation. The anchor-rule violation. It is also unnecessary: honest synthesis outperforms fake testing once readers compare the work.
  • Aggregator specs. Spec tables copied from other reviews, drift included. Verify at the maker's page or do not publish the number.
  • Anecdote dressed as synthesis. Three forum posts is not owner experience at scale. State the sample or downgrade the claim.
  • Averaged experts. Splitting the difference between conflicting expert verdicts erases the most useful information: that credible testers disagree, and why.
  • The drawback-free pick. A recommendation with no named drawback reads as advertising because it is structured like advertising.
  • Backfilled hands-on. Quietly rewriting an old synthesis piece as if it had been tested all along. The upgrade-and-mark pattern exists so this never has to happen.
  • Footer disclosure. An affiliate disclosure the reader has to hunt for fails both the FTC's placement expectations and the trust purpose it exists to serve.
  • Schema overreach. Review markup on synthesis content. The piece's prose is honest and its markup lies.

Output format

For a methodology setup: a short methodology standard the site adopts (the tiers, the block template filled with the site's sources, the schema policy), suitable for docs or an editorial guide.

For a piece-level engagement: the methodology block for that piece, the criteria list, the evidence file (sources gathered per tier with sample sizes), and the claims audit if the piece existed before this skill did.


If required data is unavailable

This skill's output depends on data, measurements, or tool results it cannot generate on its own. When a required input, tool, or data source is unavailable or unverifiable, the sanctioned output is the deliverable with the gap stated: what was needed, what was actually obtained or verified, and which parts of the output are affected. Fabricating, estimating, or interpolating a required number to complete the deliverable is never sanctioned. A stated gap is a complete answer.


Reference files

© rampstackco, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 3 other files (references) in skills/evidence-based-reviews of rampstackco/claude-skills.

  • SKILL.md
  • README.md
  • references/evidence-tiers.md
  • references/methodology-block-template.md

Open the folder on GitHubat commit 482c9bf

Compare with similar skills

Evidence Based Reviews next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Evidence Based Reviews compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Evidence Based Reviews this skillrampstackco/claude-skills940—~3.1kAutomated safety check: PassMIT
Fabric Lakehousegithub/awesome-copilot40k1 repos~1.5kAutomated safety check: PassMIT
Fabricynulihao/AgentSkillOS6172 repos~3.4kAutomated safety check: PassNone
Fabriclamm-mit/scienceclaw244—~433Automated safety check: PassApache-2.0
Rent Vs Buymohitagw15856/pm-claude-skills1.4k—~1.2kAutomated safety check: PassMIT
Fabric CLIdata-goblin/power-bi-agentic-development1k—~10kAutomated safety check: PassGPL-3.0

Similar skills

  • Fabric Lakehouse

    github/awesome-copilot

    Official

    A skill your agent uses to get context about Fabric Lakehouse and its features for software systems and AI-powered functions.

    40k GitHub starsUsed in 1 repo~1.5k tokens
    DatabasesAuto-check passed
  • Fabric

    ynulihao/AgentSkillOS

    Intelligent pattern selection for Fabric CLI. An agent skill from ynulihao/AgentSkillOS.

    617 GitHub starsUsed in 2 repos~3.4k tokens
    SecurityAuto-check passed
  • Fabric

    lamm-mit/scienceclaw

    Pattern-based analysis using Fabric's 242+ specialized prompts for summarizing papers and extracting insights

    244 GitHub stars~433 tokensUpdated 1 mo ago
    Auto-check passed
  • Rent Vs Buy

    mohitagw15856/pm-claude-skills

    Model rent-vs-buy honestly — year-by-year net position for both paths including the assumption everyone drops (the renter invests the difference), with a breakeven horizon instead of a verdict.

    1.4k GitHub stars~1.2k tokensUpdated yesterday
    Auto-check passed
  • Fabric CLI

    data-goblin/power-bi-agentic-development

    Expert guidance for the Fabric CLI (fab) and the Fabric and Power BI REST APIs: workspaces, items, lakehouses, notebooks, pipelines, semantic models, reports, capacities, OneLake, deployment and…

    1k GitHub stars~10k tokensUpdated yesterday
    Data & AnalyticsAuto-check passed
  • Official

    A skill your agent uses when integrating NVIDIA NeMo Fabric into a consumer application, service, evaluation harness, or platform through the typed Python SDK — translating the consumer's own…

    3.5k GitHub stars~5.5k tokensUpdated yesterday
    AI & LLM EngineeringAuto-check passed

More from rampstackco/claude-skills

All 103 skills in this repo
  • After Action Report

    rampstackco/claude-skills

    Run a structured after-action review (postmortem, retrospective) on a launch, incident, or completed project to capture timeline, root cause analysis, contributing factors, and actionable lessons.

    940 GitHub starsUsed in 1 repo~2.5k tokens
    Auto-check passed
  • Analytics Strategy

    rampstackco/claude-skills

    Design measurement frameworks including event taxonomy, KPI hierarchy, dashboard architecture, attribution models, and analytics implementation strategy.

    940 GitHub starsUsed in 1 repo~2.4k tokens
    Auto-check passed
  • Brand Style Guide

    rampstackco/claude-skills

    Build or audit a comprehensive brand style guide that documents the full brand system including story, logo system, color, typography, imagery, voice, applications, and dos/don'ts.

    940 GitHub stars~2.1k tokensUpdated yesterday
    Auto-check passed
  • Brand Voice

    rampstackco/claude-skills

    Develop or document a complete brand voice and tone system covering voice attributes, tone shifts by context, vocabulary preferences, grammar rules, and copy examples.

    940 GitHub stars~2.2k tokensUpdated yesterday
    Auto-check passed
  • Content And Copy

    rampstackco/claude-skills

    Write or edit website copy, blog content, and editorial pieces with attention to voice, structure, and goal.

    940 GitHub stars~2.1k tokensUpdated yesterday
    Auto-check passed
  • Content Strategy

    rampstackco/claude-skills

    Develop a content strategy covering editorial positioning, content pillars, formats, calendar, governance, and topical authority planning.

    940 GitHub stars~2.6k tokensUpdated yesterday
    Auto-check passed

Questions about Evidence Based Reviews

What does Evidence Based Reviews do?

Produce honest product reviews and buying guides without fabricated first-hand experience, using disclosed evidence tiers: verified specs, owner-experience synthesis at scale, expert triangulation…. Evidence Based Reviews is an agent skill from rampstackco/claude-skills. Produce honest product reviews and buying guides without fabricated first-hand experience, using disclosed evidence tiers: verified specs, owner-experience synthesis at scale, expert triangulation, and hands-on only when true.

When should I use Evidence Based Reviews?

Evidence Based Reviews fits situations like: the user wants to write product reviews; buying guides without hands-on access; set up a review methodology; add evidence disclosure to review content.

How do I install Evidence Based Reviews in Claude Code?

Run `npx skills add rampstackco/claude-skills --skill evidence-based-reviews -a claude-code`. Or copy the skill folder (skills/evidence-based-reviews in rampstackco/claude-skills) into .claude/skills/evidence-based-reviews in your project. Claude Code loads it when a task matches its description.

How do I install Evidence Based Reviews in Codex?

Run `npx skills add rampstackco/claude-skills --skill evidence-based-reviews -a codex`. Or copy the skill folder (skills/evidence-based-reviews in rampstackco/claude-skills) into .agents/skills/evidence-based-reviews in your project. Codex loads it when a task matches its description.

Can I use Evidence Based Reviews in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add rampstackco/claude-skills --skill evidence-based-reviews -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/evidence-based-reviews, .gemini/skills/evidence-based-reviews, .github/skills/evidence-based-reviews and .opencode/skills/evidence-based-reviews in your project.

What does Evidence Based Reviews need to run?

SKILL.md names no scripts, command-line tools or credentials: Evidence Based Reviews is instructions for the agent only.

Does Evidence Based Reviews access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Evidence Based Reviews safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Evidence Based Reviews use?

Evidence Based Reviews is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Evidence Based Reviews use?

About 3.1k tokens (SKILL.md is roughly 12k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 1.7k tokens, read only when the agent opens those files.

What are the alternatives to Evidence Based Reviews?

Skills that share tags, products or a category with Evidence Based Reviews: Fabric Lakehouse (github/awesome-copilot, 40k stars), Fabric (ynulihao/AgentSkillOS, 617 stars), Fabric (lamm-mit/scienceclaw, 244 stars) and Rent Vs Buy (mohitagw15856/pm-claude-skills, 1.4k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Evidence Based Reviews?

rampstackco (a GitHub organization) maintains it in rampstackco/claude-skills, which has 940 GitHub stars. The repository holds 103 skills in this directory. The repository was last updated on October 7, 2026.

Source: rampstackco/claude-skills on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.