Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Triangulate Research: Cited Answers That Flag Disagreement

    1

    Evidence-first research: splits a question into sub-questions, ties every claim to its source URL, scores source credibility, and flags where independent sources contradict each other instead of quietly picking a side.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Triangulate Research: Cited Answers That Flag Disagreement

    Triangulate Research: Cited Answers That Flag Disagreement

    Example session with this skill installed

    Research whether coffee is good for cardiovascular health, break it into sub-questions, and extract claims from 3 sources into case.json.

    • Read your context and instructions
    • Compiled the triangulate research

    I have completed the research plan and evidence extraction for the cardiovascular effects of coffee. I investigated two sub-questions regarding CVD risk and blood pressure, gathered three distinct sources (one peer-reviewed journal, one health news site, and one blog), and synchronized the findings through triangulate.py.

    Research Report: Coffee and Cardiovascular Health

    Subject KeyClaimSourceScore
    coffee_cvd_risk2-3 cups/day associated with 15% lower CVD risk.[PMC10293]0.98
    coffee_cvd_riskRegular consumption linked to reduced heart failure.[HealthLine]0.65
    coffee_bp_impactCaffeine causes temporary spikes in blood pressure.[MayoClinic]0.90
    coffee_bp_impactCoffee has no long-term effect on hypertension.[HealthLine]0.65

    Contradictions Detected

    Textual Discrepancy (coffee_bp_impact): Source MayoClinic (High Credibility) notes acute spikes, while HealthLine (Mid Credibility) focuses on long-term neutrality. These are not mutually exclusive but represent different timescales of physiological impact.

    Findings by Sub-Question

    1. Is coffee linked to lower CVD risk?
      • High-credibility evidence from PMC (Score 0.98) indicates a 15% risk reduction.
      • Corroborated by HealthLine (Score 0.65), though with less specific statistical backing.
    2. Does coffee impact blood pressure?
      • Mixed findings. High-credibility sources focus on short-term hypertensive spikes, while mid-tier sources claim no long-term damage.

    Confidence & Gaps

    • High Confidence: CVD risk reduction (corroborated across two independent domains).
    • Gaps: No claims found regarding decaffeinated coffee vs. caffeinated coffee specific to CVD.

    Next steps

    • Investigate the specific impact of brewing methods (filtered vs. unfiltered) on cholesterol.
    • Expand research to include meta-analyses from the last 12 months to improve recency scores.
    • Run a JSON export of the report for integration into the health dashboard.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Ask your agent a research question and you usually get one confident paragraph built from whatever the top few search results said, even when those results disagree or one of them is a random blog.

    Triangulate Research gives your agent an evidence-first workflow instead.

    What it does

    • Plans first. Splits the question into 2–5 sub-questions and searches each one separately, so broad questions get broad coverage.
    • Every claim is sourced. Findings are recorded as discrete claims, each tied to the exact URL it came from. Nothing comes from model memory.
    • Every source is scored. Each source gets a 0–1 credibility score: .gov/.edu/peer-reviewed highest, user-generated platforms lowest, stale pages penalized. Claims repeated by several independent domains get a corroboration bonus.
    • Contradictions are surfaced, not resolved. When independent sources disagree on the same fact, the report says so at the top and shows both sides with their credibility. It never quietly picks a winner.
    • Structural gate. A built-in check fails the run if any claim lacks a real source, an ID dangles, or a contradiction is malformed, so broken reports never reach you.

    How it works

    Your agent does the judgment work: planning, searching, reading and extracting claims. A small deterministic Python script (standard library only, no API keys, no network calls) does the scoring, contradiction detection, validation and the Markdown report, so the same evidence always gives the same result.

    What's inside

    • SKILL.md: the full research workflow
    • scripts/triangulate.py: credibility scoring, contradiction detection, validation gate, report
    • references/case-format.md: data format and subject-key guide
    • examples/coffee-case.json: a worked example

    Good for

    Fact-checking, literature and market scans, due diligence, "is X actually true?" questions, anything where being confidently wrong is expensive.

    Requires search tool access for real-time URLs. Published dates must be explicit; the agent will not guess. Textual contradiction detection is a heuristic flag requiring manual review.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 19 days ago

    Requires

    GitHub
    python3
    web search tool
    web fetch tool (recommended)

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access

    Allowed Hosts

    github.com

    File Scopes

    case.json
    report.md

    Network access is only the agent's own web search/fetch for gathering sources. The bundled script makes no network calls and reads no environment variables. It reads case.json and writes report.md in the working directory.

    Listed19 days ago

    What's inside

    Frequently Asked Questions

    Unlike standard LLMs that try to reach a single consensus, this skill uses a code-based validator to force the agent to identify and flag numeric or textual disagreements, preventing it from 'averaging out' conflicting data points.

    The skill is built for any AI agent capable of executing Python 3.9 scripts and managing JSON files, making it compatible with most custom GPTs, AutoGPT setups, and local agent frameworks.

    You receive the core `triangulate.py` validation script, the `case.json` schema documentation for evidence mapping, and the workflow instructions to guide your agent through the sub-question planning phase.

    While the Python script scores credibility and detects contradictions automatically, the skill requires an external search tool or manual input to provide the initial source URLs for processing.

    The skill uses a deterministic 'corroboration bonus' system where claims backed by multiple high-authority domains receive a higher confidence score, which is then reflected in the final Markdown report.

    New in Research & Analysis