- Home
- Skills
- Testing & Debugging
- Defect Reproduction Check
Defect Reproduction Check
Assess a reported software defect using an explicit reproduction plan and traceable observations. Distinguish reproduced, not reproduced in specified conditions, and inconclusive.
Free
Works with the AI Agents you already use
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
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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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
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.