Defect Reproduction Check

    1

    Assess a reported software defect using an explicit reproduction plan and traceable observations. Distinguish reproduced, not reproduced in specified conditions, and inconclusive.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Defect Reproduction Check

    Example session with this skill installed

    Assess the attached synthetic SORT-DEMO case using its supplied run evidence. Check the required conditions, completed steps, observed result, and evidence hash. Return the complete reproduction assessment without claiming that the assessor executed the target.

    • Read your context and instructions
    • Compiled the defect reproduction check

    Fresh local synthetic demonstration from 2026-11-06 on Python 3.12.15. The demo utility actually executed an identity-copy function with [3, 2, 1], recording [3, 2, 1]. It separately executed built-in sorted, recording [1, 2, 3]. The assessor classified those supplied observations; it did not execute either target. The third scenario was deliberately blocked and did not call a target.

    Complete assessed reproduction output

    Defect reproduction assessment

    Outcome: reproduced
    Case: "SORT-DEMO"
    Claim: "Synthetic sorting target returns descending input unchanged."

    Conditions: {"build": "synthetic-demo", "input": [3, 2, 1]}
    Expected: [1, 2, 3]
    Defect signature: [3, 2, 1]
    Runs assessed: 1
    Reproducing runs: ["synthetic-reproduced"]
    Expected-behavior runs: []

    • Run "synthetic-reproduced": defect_signature; eligible=true; synthetic=true
      Conditions: {"build": "synthetic-demo", "input": [3, 2, 1]}
      Evidence: {"bytes": 403, "integrity": "matched", "path": "reproduced.json", "sha256": "a6162f336a7d0b124ea992e2983cda5074dc9809ff01c363d4816c0e947422bf"}

    Preserve the failing run and investigate causes separately; reproduction does not identify a cause.

    Limitation: This assessor did not execute the target. Outcomes summarize supplied observations and locally checked evidence integrity, not authenticated execution provenance, a fix, root cause, or universal absence of a defect.

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

    About this skill

    What it does

    Turn a defect report and its existing run evidence into a scoped reproduction assessment. Define the expected behavior, defect signature, build, environment, inputs, and steps. The included local Python helper validates structured inputs, checks the hashes of explicitly supplied evidence files, and reports qualifying observations, excluded runs, gaps, and the next useful check. It produces JSON or Markdown and distinguishes reproduced, not reproduced in specified conditions, and inconclusive. The helper does not run the target or authenticate evidence; an agent or human must inspect the evidence before treating a classification as a factual reproduction conclusion.

    The included examples are wholly synthetic. Local verification passed 22 tests, including 144 gating combinations, and exercised two synthetic target functions plus one deliberately unexecuted blocked scenario. The assessor itself only reads supplied observations and explicitly scoped evidence files; it does not execute the reported target.

    Requirements

    Requires Python 3.9+ standard library and permission to read supplied case/evidence files, write a temporary script, and execute it locally. No network, account, API key, third-party package, install step or paid dependency is required.

    No genuine external application, production service, repository, UI, CI system or customer incident was exercised. On November 6, 2026, creator-account Agensi MCP delivery was verified. The delivered scripts matched the release ZIP and ran unchanged locally on macOS with Python 3.12.15. This was not Agensi-hosted execution, a non-creator buyer test or production validation. The delivery test exercised all three classifications, checked supplied evidence hashes, generated Markdown and rejected malformed JSON. The assessor did not run an external target. No public GitHub repository or commit is claimed.

    Source and license

    Author: Evgeny Tkachenko. Copyright (c) 2026 Evgeny Tkachenko. MIT license included and existing contributor notices retained. Package version 1.0.1 is an attribution-only revision; assessor/runtime behavior remains unchanged.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Testing & Debugging.

    The assessor does not execute a target, fetch URLs, inspect a repository, or execute recorded commands.
    Evidence hashes check local file integrity only. They do not authenticate evidence or prove that the evidence supports an assertion.
    An agent or human must inspect the underlying evidence and the declared scope before treating a classification as a factual conclusion.
    Not reproduced applies only to the submitted conditions and runs. The output does not establish root cause, prove a fix, or prove universal absence of a defect.
    All bundled examples are synthetic. No genuine external application, production system, repository, UI, CI system, or customer incident was exercised.
    Verification covered creator-account MCP delivery and local execution only, not Agensi-hosted execution, non-creator access, production use or broader compatibility.
    Python 3.9+ is the stated requirement; this local verification used Python 3.12.15 only.

    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

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 10 days ago

    • Free with an Agensi account

    Software
    Python 3.9+

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Reads only the caller-supplied JSON case and explicitly referenced evidence files under the chosen evidence root. The host writes and runs a temporary unchanged Python assessor and may save requested output. The assessor does not run the target, execute recorded commands, fetch URLs, inspect unrelated repositories or change application/CI state. Evidence hashes check integrity, not authenticity or semantic support.

    Listed10 days ago

    What's inside

    Frequently Asked Questions

    The included assessor does not execute the target or any recorded command. It classifies supplied run observations against declared conditions, completed steps and local evidence integrity. Any new target execution needs its own authorized workflow.

    Only that the supplied eligible runs showed expected behavior under the stated conditions. It does not prove universal absence, identify root cause, establish a fix or authenticate the underlying evidence.

    Provide the structured case, expected behavior and defect signature, explicit conditions and steps, per-run observations, and referenced local evidence files with hashes. Python 3.9+ and scoped local file/script access are required. Examples are synthetic. Creator-account MCP delivery and unchanged local execution were verified on November 6, 2026; no external application, Agensi-hosted runtime or independent buyer was tested.

    New in Testing & Debugging

    Free