More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    AI Agent Testing and Evaluation Bundle

    4

    Six skills to compare agent runs, turn failures into regression tests, build eval datasets, gate provenance, inspect untrusted content, and write audit reports whose proof a stranger can re-check.

    $19

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    24 uses5.0 (1 review)
    AI Agent Testing and Evaluation Bundle

    AI Agent Testing and Evaluation Bundle

    Example session with this skill installed

    Prepare a release candidate for our 'CustomerSupport-v1' dataset. We have 3 JSONL sources from GitHub, but source C has a non-commercial CC license. Check for leakage and PII.

    • Read your context and instructions
    • Compiled the ai agent testing

    VERDICT: HOLD
    Reason: Source C license (CC-NC) conflicts with commercial distribution model.
    Action: 45 records moved to EXCLUSIONS.jsonl.
    Integrity: Deduplication found 12 clusters; split leakage prevented by grouping source families.
    Artifacts generated: PROVENANCE.jsonl, SOURCE_REGISTRY.json.

    PROVENANCE.jsonl, SOURCE_REGISTRY.json

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

    About this skill

    A six-skill bundle for debugging and evaluating AI agents: run diff, regression-test builder, eval dataset builder, provenance/release gate, untrusted-content analysis, and retained-proof audit reports. For operators who need evidence, not another prompt pack. What you get - Compare two agent runs and show what changed - Turn a failed trace into a regression test fixture - Build a deterministic eval dataset with recomputable results - Gate dataset provenance, rights, and leakage before release - Inspect untrusted content for prompt-injection before an agent reads it - Write an audit report whose proof a stranger can re-check, with SHA-256 manifest and negative control These are workflow skills rather than application-specific scripts, so they can be adapted across different agent stacks, models, coding environments, and evaluation pipelines. Built for people working on AI agents, LLM evaluation, agent QA, regression testing, dataset quality, prompt-injection analysis, and AI reliability. 6 skills. One workflow: capture evidence, compare behavior, reproduce failures, build evaluations, and verify what you ship.

    Does not provide legal advice. Requires deterministic transformation logs and manual entry of rights evidence for review. Cannot automatically resolve complex legal conflicts.

    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

    5.0
    1 review
    5
    1
    4
    0
    3
    0
    2
    0
    1
    0

    This skill has been used 24 times.

    Trust & safety

    Security scan

    Security scan passed 19 days ago

    Permissions

    Terminal / Shell
    Write Files
    Read Files

    File access is limited to artifacts explicitly supplied or selected by the user for analysis, comparison, evaluation, or packaging. The skills do not require network access, browser access, API credentials, or environment variables. Live external side effects are not performed implicitly; workflows requiring expanded access must be separately authorized.

    Listed2 months ago
    Updated19 days ago

    What's inside

    Frequently Asked Questions

    This skill establishes a deterministic audit trail for AI training data by tracking source hashes, documenting every transformation step, and identifying duplicate records that could cause split leakage between training and testing sets.

    You will receive the full skill package including the provenance agent logic, pre-defined metadata schemas for source pinning, and a template for generating compliance-ready attribution files.

    While designed to integrate seamlessly with data engineering agents (such as AutoGPT or LangChain-based workers), this skill can be added to any agent workflow that supports JSON-based data input and output hooks.

    Setup involves defining your 'Gate Policy'—a set of rules for PII, integrity, and rights evidence—which the skill then uses to automatically evaluate and categorize data records as PASS, HOLD, or FAIL.

    The skill handles the technical verification of hashes, lineage, and source documentation, but it does not provide legal advice or guarantee that your data meets specific regional copyright laws.

    The Dataset Provenance Gate is built for scale and can process high-volume datasets, though execution speed will vary based on the computational resources allocated to your host AI agent.

    New in Testing & Debugging