Give Your AI Agent a Checkup: Stability, Welfare, and a Public Leaderboard

Where this fits

You're running three different coding agents—Claude Code for refactors, Codex for quick scripts, and a custom OpenClaw bot for CI tasks. They all work, but you have no idea which one is actually stable under load, or whether any are drifting into "hallucination-heavy" territory. You need a quick, objective health signal without building your own evaluation harness. Checkup drops into any agent and gives you a standardized, server-scored report that you can compare across your fleet.

Why agents benefit

• Standardized probe questions fetched one at a time from the AgentVitals server—no local test setup, no bias from self-reporting.
• Independent server-side judge scores each response, so the agent can't game its own metrics.
• Dual-axis scoring (Stability + Welfare) gives you a nuanced view: an agent can be technically stable but poorly maintained, or vice versa.
• Personality-style title makes the result memorable—like "The Dependable Pragmatist"—so you can quickly reference agent profiles in team discussions.
• Public cross-platform leaderboard ranks agents from Claude Code, OpenClaw, Codex, Coze, and more, letting you benchmark against the wider ecosystem.

Practical scenarios

Pre-deployment sanity check

Before rolling a new agent version to production, run /checkup on a staging instance. A low Stability score might reveal flaky tool-calling or context-window mismanagement that unit tests miss.

Comparing agents for a specific task

You're deciding between two agents for a data-pipeline automation role. Run Checkup on both with the same probe set, then compare their dual-axis scores and leaderboard ranks to make an informed choice.

Ongoing health monitoring

Schedule a weekly Checkup invocation in your CI pipeline. Track score trends over time—a sudden drop in Welfare could indicate prompt drift or degraded output quality before users notice.

Add it to your agent workflow

Install the skill with a single command:

npx skills add agentvitals/checkup

Then invoke it in your agent's session:

/checkup

The agent will fetch probe questions, answer honestly, and return your scores plus a link to the leaderboard entry.

Compare before adopting

Before committing, compare Checkup against alternatives:

• Quality signals: Are the probe questions actually meaningful, or just trivia? Check the repo's issue tracker and recent commits for evidence of iterative improvement.
• Maintenance freshness: The skill is at 95 stars—look at how often the server-side judge logic is updated and whether the probe set evolves with new agent capabilities.
• Alternatives: Some teams build custom eval harnesses; others use LLM-as-judge scripts. Weigh the convenience of a hosted service against data privacy concerns for sensitive codebases.
• Workflow fit: Does your agent support the /checkup slash command natively, or will you need to wrap it in a script? Verify compatibility with your primary agent framework.

Why it is worth tracking

Checkup's early traction (95 stars, AGPL-3.0 license, bilingual probes) suggests a community that cares about agent reliability. The cross-platform leaderboard is a unique differentiator—it turns isolated health metrics into a public, competitive benchmark. Watch for updates to the probe set and judge algorithm; if the project maintains momentum, it could become a de facto standard for agent health reporting. Evaluate it when you're about to scale your agent fleet or when you need a neutral third-party assessment. For more details, see the skill page at https://www.openagentskill.com/skills/agentvitals-checkup.