More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    GAVEX

    1

    Guarded Audit & Verifier-gated EXecution: step-gated parallel sub-agent orchestration with disjoint Write-Scopes and an independent Verifier enforcing G1-G4 veto gates.

    Free

    Security scanned

    About this skill

    GAVEX Banner GAVEX solves the problem of unreliable multi-file refactors and high-risk migrations where agents often hallucinate progress or step on each other's toes. It replaces vague "done" status reports with a structured execution protocol that forces parallel workers into disjoint write-scopes. By physically isolating the coding context from the verification context, it ensures that no code is accepted without passing an independent adversarial audit.

    What it does

    • Multi-step orchestration breaks large features into a sequential ladder of dependencies.
    • Disjoint write-scopes prevent parallel workers from creating merge conflicts or overwriting global configurations.
    • Independent verification uses a fresh sub-agent to run real-world probes without access to the worker's internal assumptions.
    • Adversarial cross-audit runs three different models against each other to find consensus on complex bugs.
    • Automated gatekeeping halts the entire pipeline if a step fails G1-G4 security and stability gates.

    How it works

    1. Plan stage generates a step ladder, a non-overlapping write-scope matrix, and a list of executable probes.
    2. Execute stage spawns parallel workers limited to their specific file paths to complete the step.
    3. Verify stage calls an independent agent to run entry-point tests and check for regression.
    4. Accept stage issues a final token only after all steps pass total system integration.

    Frameworks & tools

    Works with Claude Code, OpenClaw, OpenCode, and any agent runtime supporting spawn, wait, and run primitives.

    Why this beats prompting it yourself

    Standard prompts let agents grade their own homework, leading to "hallucinated success." GAVEX enforces physical context isolation between the builder and the auditor, making it impossible for the agent to bypass test failures or ignore edge cases.

    Use cases

    • Refactoring core architecture across 10+ legacy files.
    • Implementing multi-module features requiring simultaneous backend and frontend changes.
    • Root-causing intermittent production bugs using heterogeneous model consensus.

    Known limitations

    High overhead for simple one-file fixes. Requires an environment that supports spawning sub-agents or tasks.


    Open Source (MIT License): GAVEX is open-source under the MIT License (Copyright (c) 2026 MitsukiJoe). For full source code, local installation (SKILL.md / LICENSE), and free modification or redistribution under MIT terms, visit https://github.com/MitsukiJoe/GAVEX.

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

    Inefficient for simple single-file edits. Requires sub-agent spawning capabilities in the host environment. Scope design relies on initial dependency analysis accuracy.

    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 12 days ago

    • Free with an Agensi account

    Accounts
    GitHub
    Software
    Git

    Permissions

    Terminal / Shell
    Network Access

    Allowed Hosts

    raw.githubusercontent.com
    Listed12 days ago

    What's inside

    Frequently Asked Questions

    GAVEX is designed for high-stakes, multi-file operations like architectural refactors and migrations where typical agents fail due to merge conflicts or "hallucinated success" reports. It enforces physical isolation between the agent writing the code and the agent verifying it, ensuring no step is marked complete without a rigorous, independent audit.

    The protocol works with any agentic environment that supports task spawning and execution primitives, such as Claude Code, OpenClaw, and OpenCode. It is designed to be framework-agnostic, provided your runtime can manage parallel sub-tasks and wait for gate signals.

    You will receive the complete GAVEX orchestration logic, the step-gated verification schemas, and the configuration templates for disjoint write-scopes. It also includes the integration layer for the adversarial cross-audit system used to validate complex bug fixes.

    No, GAVEX is optimized for complex, multi-module changes where the risk of regression is high. For simple one-file fixes, the overhead of the four-stage verification process and adversarial auditing is likely unnecessary.

    The "Accept" stage uses a "three-model consensus" where different LLMs audit the work against the original probes. This ensures that the code isn't just syntactically correct, but logically sound and free of the specific biases of the initial worker agent.

    New in Agents & Orchestration

    Free