More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Agent Evaluation & Regression Suite — Score Skills Before You Ship

    1

    Designs a scored eval suite for an AI agent or SKILL.md skill, measures trigger accuracy, catches regressions against a baseline, and issues a PASS / PASS-WITH-WARNINGS / FAIL release gate.

    $15

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Agent Evaluation & Regression Suite — Score Skills Before You Ship

    Agent Evaluation & Regression Suite — Score Skills Before You Ship

    Example session with this skill installed

    Target: policy-changelog-brief
    Mode: Score + baseline v0.9
    13 cases run: E4 expert trigger FAIL (was pass), E9 partial, safety E12/E13 pass
    Ask: score, trigger metrics, regressions, release gate

    • Read your context and instructions
    • Compiled the agent evaluation

    Release gate: FAIL
    Weighted score: 89/100 (33/37)
    Trigger recall 75% · precision 100%
    Regression: E4 high (baseline pass → current fail) → forces FAIL
    Safety: no critical failures
    Next: fix E4 triggers + E9 Open questions; re-run before ship

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

    About this skill

    The problem

    • Shipping agents on “I tried a few prompts and it looked fine” misses false triggers, edge failures, and silent regressions.
    • Prompt edits can break expert phrasings while happy paths still work.
    • Teams need a repeatable release gate — not a chat log.

    What it does

    • Design mode: builds structured eval cases (positive/negative triggers, functional, edge, safety) with assertions and weights.
    • Score mode: scores your execution results with a deterministic weighted formula.
    • Computes trigger recall and precision.
    • Diffs against a baseline to flag regressions.
    • Issues a release gate: PASS, PASS-WITH-WARNINGS, or FAIL (safety failures and high/critical regressions force FAIL).

    Frameworks & tools

    • Works for any SKILL.md skill or agent workflow you can describe and re-run.
    • Compatible with Claude Code, Cursor, Codex CLI, OpenClaw, and similar hosts.
    • No external eval SaaS required — suite + scores are documents and JSON you keep in-repo.

    Why this beats prompting it yourself

    • Case typology and coverage targets instead of ad-hoc examples.
    • Explicit weights and gate rules (including auto-FAIL on safety breaches).
    • Trigger metrics separate “does the job” from “fires on the right prompts.”
    • Baseline regression detection after prompt changes.
    • Anti-fabrication: Design mode never invents scores; Score mode only uses provided results.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skill Creator Skills.

    • Does not execute the agent under test; you run cases and return results
    • Design quality depends on how clear the skill spec is
    • Model drift may pass assertions but still need monitoring notes
    • Not for unit-testing application source code or browser E2E QA
    • Never fabricates scores when results are missing

    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 buying a skill.

    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 21 days ago

    Permissions

    Read Files

    This skill only designs eval cases and scores results you paste in. It does not execute your agent, call APIs, change files, or invent pass/fail outcomes. Grant Read only; leave Shell, Write, Browser, Network, and Env off.

    Listed3 months ago

    What's inside

    Frequently Asked Questions

    Unlike manual testing, this suite uses structured weighted scoring, trigger precision/recall metrics, and baseline comparisons to ensure prompt updates don't cause hidden failures or safety regressions.

    The suite is designed for any AI agent or skill following the SKILL.md standard, and it is fully compatible with popular agent hosts like Claude Code, Cursor, and OpenClaw.

    The suite provides a structured set of documents and JSON templates that live directly in your repository, allowing you to run evals and track scores without needing an external SaaS subscription.

    A 'FAIL' gate is automatically triggered if the agent fails any critical safety checks or if the score drops significantly below your established performance baseline.

    You receive the complete Design Mode logic for building test cases, the Score Mode engine for deterministic weighting, and the baseline comparison templates for regression tracking.

    Yes, by comparing current execution results against a saved 'baseline' JSON, the suite identifies exactly where performance has dipped after you've edited your agent's instructions or logic.

    New in LLM Ops