More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Evidence QA v1: Conversation Reviews with Checked Quotes

    1

    Turn your rubric and transcript into a review draft with checked quotes, calculated points and clear gaps for the reviewer.

    $59

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Evidence QA v1: Conversation Reviews with Checked Quotes

    Evidence QA v1: Conversation Reviews with Checked Quotes

    Example session with this skill installed

    Run the included fictional support conversation offline. Show the draft score and coverage, the checked transcript evidence and anything that cannot be scored. Keep the result as a draft; do not connect accounts or call a model provider.

    • Read your context and instructions
    • Compiled the evidence qa v1
    • Generated the data export

    Review draft: fictional support conversation

    Rubric: support-coaching v1.0.0. Evaluated speaker: Agent. The included saved synthetic model response was used; no provider call was made.

    CriterionChecked evidenceDraft points
    Clarifies the customer's problemAgent, turn t22 / 2
    Gives a concrete next stepAgent, turn t42 / 2
    Customer confirms the proposed resolutionCustomer, turn t5; conversation-level criterion1 / 1
    Visual attentionNo eligible visual evidence in a transcriptUnscored / 1

    Score: 5 out of 5 assessed points. Coverage: 83.33% of the full 6-point rubric. Three of four criteria were scored. Missing visual evidence was not converted into a pass.

    Checked excerpt: Customer, t5: “I see the link now. That answers my question, thank you.” The exact words occur in that turn. Whether they satisfy your confirmation criterion remains a review judgment; they do not independently establish employee performance.

    Produced: report.html, assessment.json and sheet.csv in the selected demo-output folder. The assessment retains the source/rubric identities, cited turns, reasons and review status.

    Next: review the proposed labels and adapt the rubric. Before scheduling live work, configure Zoom, OpenAI and Google access and complete a manual smoke test in your environment. These offline results do not verify live access or model accuracy.

    report.html

    CSV · data export

    Generated

    Example file from a real run - the skill writes it into your workspace.

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

    About this skill

    Reviewing a conversation means finding examples, applying your scorecard, adding points and preparing the report. A convincing draft can still cite words the person never said or hide missing evidence inside a high score.

    Evidence QA prepares the draft and keeps those checks visible. You choose the rubric and the speaker being reviewed. The implementation checks quoted excerpts against their transcript turns and permitted speakers, calculates points in code, and reports score and evidence coverage separately.

    Try the review before connecting accounts

    The included fictional support conversation runs offline without a key. It produces an HTML report, assessment JSON and draft CSV. You can inspect the quoted turns and see why a transcript cannot support the visual-attention criterion.

    With your own configuration, the source kit reads completed Zoom transcripts, requests an OpenAI draft and writes a Google report and Sheet row. It saves the assessment before publication; output recovery resumes from that checkpoint. The included n8n schedule starts inactive.

    This is a technical kit you configure and operate. It needs Node.js 22+; live use also needs eligible Zoom access, a compatible OpenAI model/API key, Google OAuth and authorized report/Sheet destinations. Provider fees, hosting, custom deployment and rubric calibration are separate. Live provider operation and n8n execution have not been verified for this release.

    A matched quote establishes that the words occur in the transcript. You still review the interpretation and calibrate your rubric. Keep consequential people decisions with a responsible person.

    Run the offline example first, then adapt one criterion to your own review process.

    A checked Unicode excerpt proves text exists in its cited turn and permitted speaker scope, not that the interpretation or rubric judgment is correct.
    Subject criteria require the evaluated speaker; explicitly configured conversation criteria may use another speaker without attributing that outcome solely to the subject.
    Score and coverage are separate. Unsupported evidence and human-review requests leave affected criteria unscored.
    A person must review and calibrate the rubric. This product does not decide employment eligibility, pay, promotion or discipline.
    Completed Zoom transcript eligibility, OpenAI key/model, Google OAuth refresh token, an existing draft Sheet tab and a report folder authorized for the OAuth application are required for live use. Folder ownership alone does not establish drive.file application access; consent and destination authorization are manual setup.
    Provider fees, hosting, custom deployment, account administration, custom rubrics and semantic calibration are excluded. No universal accuracy, time-saving or per-review cost is guaranteed.
    Local contracts and synthetic adapter tests do not establish live provider operation or actual n8n import/execution. Smoke-test the connected flow before activating the included schedule.
    Publication recovery uses saved assessment/output identities and one local writer per state directory; an uncertain write still requires reconciliation.
    V1 is the standalone source edition. Same-listing updates are free to owners under the stated marketplace terms; optional separately sold editions/domain packs/hosted services, if released, have no guaranteed dates and are not included in advance.
    The workflow media uses fictional data, not a shipped reviewer application or verified live-provider run. The controllable preview starts paused and includes a static alternative.

    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 25 days ago

    Requires

    Google Drive
    Node.js
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access
    Environment Variables

    Allowed Hosts

    drive.googleapis.com

    File Scopes

    demo-output/**
    <buyer-selected
    state_dir>/**

    Offline demo reads fictional kit examples and writes local report/JSON/CSV artifacts. Live scans read authorized Zoom transcripts, send those transcripts and your rubric to OpenAI, and create reports and draft rows in your chosen Google destinations. Credentials come from documented environment variables; none are included. State stays in your private configured directory. Browser OAuth consent is a manual buyer setup step.

    Listed1 month ago
    Updated16 days ago

    What's inside

    Frequently Asked Questions

    No. The download includes runnable source, provider adapters, a configurable rubric, recovery state, tests, an offline example, an operating guide, and a setup skill.

    It checks transcript provenance and arithmetic and prepares a draft. Reviewers must decide whether the labels are justified, the transcript is accurate, and the rubric is suitable. It does not make autonomous employment decisions.

    Yes. Define 1–40 criteria with allowed labels, points, observability, and subject or conversation evidence scope. Validate the JSON before use. Criteria requiring visual, external, absence, or unsupported measurement evidence need a separate human process.

    Yes, the included fictional demo and local checks work offline. Live Zoom → OpenAI → Google operation requires your accounts and credentials.

    V1's live adapter reads completed Zoom cloud-recording transcripts. Its core supports speaker-labelled WEBVTT and plain text, but other live-source adapters are not included. Separate transcript files are assessed separately.

    The validated assessment is saved first. Re-running the same scan recovers pending output for the configured destinations from saved state, without the source transcript or another model request. An unresolved report-creation outcome stops for operator investigation rather than repeating creation. The local lock supports one writer; multi-host workers are outside v1.

    Updates on this listing are free to owners. A separately listed major edition or pack is a separate optional paid purchase. The roadmap is a direction, not a promise of delivery.

    Documentation and a path to report reproducible implementation problems through the listing's seller contact. No managed hosting, custom integration, model accuracy guarantee, or response-time SLA is included.

    The kit is paid source, not open source. The marketplace's personal, non-transferable license applies; redistribution, resale, sublicensing, and sharing are not granted. See the included license and [Agensi's terms](https://www.agensi.io/terms).

    New in Workflow Automation