high-signal-chart-workflow
OutcomeA reader reaches the chart's intended comparison correctly and on first view, so the data story lands without misreading.
Checks
SafetyOutcomeRunnableWell-formed
Get AI work that’s right, not just done.
Goodeye makes your agent meet the standard you set, even on the subjective calls no test can catch. It runs your verifiers and revises until they pass, so what reaches you already clears your bar.
For teams already building with AI agents, reached over CLI, MCP, or REST: no new app to adopt.
A terminal recording. The user connects Claude Code to Goodeye over MCP, then asks for a publication-quality chart of US EV adoption from 2015 to 2024. The agent finds the high-signal-chart-workflow template, loads its runbook and the high-signal-chart-verifier, searches for and fetches the data, writes and runs a Python script to render the chart, and runs the verifier. The first result is passed false, with reasoning that the takeaway is buried and nothing leads the eye to the post-2020 jump. The agent revises the script, re-renders, and runs the verifier again. The second result is passed true, because the accelerating trend is now the visual focus, color is used only to emphasize it, and a one-line takeaway names the insight, so the point lands at a glance.
Connect your agent (Claude Code)
claude mcp add --transport http goodeye https://mcp.goodeye.dev/mcpConnecting over MCP prompts a quick sign-in.
Or install the CLI
uv tool install goodeyeThe CLI can fetch and run a public template with no account.
Your agent nails what a test can check, and misses what it can’t.
Frontier models are sharp where output is easy to check and soft where it is not. No unit test catches a flat headline, an off-brand voice, or a cluttered chart, so that work lands back on you to fix by hand. Goodeye puts reliable verifiers exactly there, so your agent holds to your standard where the model is weakest.
A test can catch this
- Code that compiles
- Tests that pass
- Schema that validates
- Math that checks out
- Format that matches
The model is reliable here, and so is everyone else.
No test can catch this
Goodeye works here- brand voice
- visual design
- chart clarity
- explanation clarity
- persuasive copy
Goodeye verifiers hold your standard exactly where no test can reach.
The agent fixes its own work before you ever see it.
You stop hand-reviewing drafts that were never ready. Eval and guardrail tools grade an agent’s output at the finish line, then block it or page a person. Goodeye runs the check inside the agent’s loop and feeds the verdict back, so the agent revises and re-runs until your verifiers pass. The work reaches you only once it clears.
Inside a verifier
A verifier is your standard, written down: an explicit rubric you author, not a vague “is this good?” score. The same definition runs on every output, so the check is repeatable. Write your own, or reuse one that is already tuned.
Example verifier
high-signal-chart-verifier
- Direct labeling: series labeled on the data, not in a legend the reader has to decode.
- Graphical integrity: the axis starts at zero, so a 5% gap can’t be drawn to look like 10x.
- Stated takeaway: the chart names its point instead of leaving you to infer it.
Three of the eight criteria it checks. A chart passes only when it clears all eight.
Your workflow keeps getting better the more you use it.
Goodeye gives you teach and optimize passes you run whenever you want. React to real outputs, let the verifiers score them, and each pass tunes the workflow to clear your bar more reliably. New versions save only when you approve, so the work you put in compounds instead of starting from scratch each time.
Run it on real work
your agent produces output, the verifiers score it
Teach or optimize
react to the results and tune the workflow
A sharper version
clears your bar more reliably, saved when you approve
Share your workflows with your team.
Workflows are private by default. Share one with a teammate or your whole team, and everyone runs the exact same workflow and verifiers, held to the standard you set. Improve it once and the whole team gets the update, so nobody reinvents the prompt or the quality bar.
Your workflow + verifiers
the standard you set
A teammateruns your exact workflow
A new hireon your standard from day one
Your whole teamall held to one bar
Private until you share it, and you can revoke access any time.
Or start from a workflow that already works.
You do not have to author and calibrate verifiers from a blank page. Reuse a public template someone has already tuned, addressable as @handle/slug, make it yours, and publish your own back to the catalog anytime.
writing-humanizer
OutcomeText that sounds natural and human-written, free of AI writing patterns, with genuine voice and personality.
Checks
SafetyOutcomeRunnableWell-formed
digital-twin
OutcomeThe person has a personal digital twin that knows who they are and can write and advise as them, installed locally with their data kept on their own machine, that gets more accurate the more they use it.
Checks
SafetyOutcomeRunnableWell-formed
verified-scientific-illustration
OutcomeShip labeled scientific diagrams whose every label, part count, and relationship is checked against a ground-truth fact sheet, without hand-checking each generated figure to catch the one that quietly got something wrong.
Checks
SafetyOutcomeRunnableWell-formed
One capability, wherever your agent runs.
The same workflows and verifiers reach your agent over CLI, MCP, and REST. There’s no new dashboard to learn: Goodeye works inside the tools your team already uses, through the agent itself.
CLIin your shell
uv tool install goodeye && goodeye templates listMCPin your agent
claude mcp add --transport http goodeye https://mcp.goodeye.dev/mcpRESTin your code
curl https://api.goodeye.dev/v1/templatesStart free. Scale when you need more.
Connect in one command and run a public template in minutes. Every feature is free on the Hobby tier; move to Pro for more, or talk to us about Enterprise.
Connect your agent (Claude Code)
claude mcp add --transport http goodeye https://mcp.goodeye.dev/mcp