Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Loop Engineering: Codex Edition

    1

    Create bounded Codex goals with explicit scope, risk, evidence, review, and approval rules.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Loop Engineering: Codex Edition

    Loop Engineering: Codex Edition

    Example session with this skill installed

    Use $loop-engineering to update the shared request validator used by the API and background worker. Preserve current client error behavior, add focused tests, and stop before deployment. Inspect first, then start the goal only if the repository, tests, and an independent checker are available.

    • Read your context and instructions
    • Compiled the loop engineering

    Goal Preview

    • Objective: Update the shared request validator across the API and background worker without changing existing client-visible errors.
    • Done: Both paths use the agreed rule, focused tests pass, and a separate checker finds no blocking issue in the final diff.
    • Scope: Shared validator, two call sites, and directly related tests. Deployment and unrelated refactors are excluded.
    • Risk: T2 because the change affects shared behavior across two paths.
    • Capabilities: Confirm the repository, test runner, and a read-only checker that did not make the change.
    • Roles: Main agent investigates and implements; checker reviews the identified final snapshot; coordinator integrates evidence.
    • Evidence: E1 focused test results and exact diff; E2 call-site coverage and the review rubric.
    • Authority: This prompt authorizes implementation within the stated scope and excludes deployment. Ask before expanding scope or taking an unapproved consequential action.
    • Loop: Inspect baseline, implement, run focused checks, then review the final snapshot. Repair and repeat affected checks when new evidence warrants it.
    • Stop: If required tests or checker are unavailable, report the blocker and do not claim readiness. Stop before deployment or if the risk materially expands.
    • Start decision: Begin only if the required capabilities are present; otherwise return a not-ready report.

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

    About this skill

    Loop Engineering: Codex Edition turns an explicit goal request into a bounded plan before work begins. It helps Codex define the objective and done signals, scope, capabilities, risk tier, evidence, maker and checker roles, approval gates, and stop conditions.

    The skill keeps objective evidence, independent review, and human approval separate. T2 and T3 goals require a non-maker checker; T3 also requires explicit human authorization. It reuses existing approval only when it covers the exact action and scope, and ties retries to new evidence or changed conditions instead of arbitrary cycle limits. Delegation stays within the host's routing, permission, and lifecycle controls.

    This edition is designed for Codex goal mode and the host's create_goal capability. It does not start goals for ordinary requests, add tools, or override sandbox or permission controls. It includes SKILL.md, Codex UI metadata, two reference guides, and an Agensi license notice. It has no scripts or bundled MCP server. Other hosts need equivalent goal creation, routing, independent review, and approval capabilities.

    The public GitHub edition remains MIT licensed; this package does not revoke rights already granted for copies of that edition.

    This is instruction-only: it includes no scripts, MCP server, or tools. It cannot grant permissions or enforce human approval beyond what the host supports. Independent checking and goal creation require corresponding host capabilities. It does not assume a fixed number of execution cycles; the agent should state its retry budget for each goal.

    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

    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 1 month ago

    Permissions

    No special permissions declared or detected

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    For T2 and T3, a separate agent that did not make the change reviews the identified final snapshot against the done signals and blocking rubric. The checker cannot repair the candidate in the same pass. Objective evidence and human approval remain separate controls.

    This edition is designed for Codex goal mode. Other compatible hosts can read the SKILL.md format, but need equivalent goal creation, routing, independent review, and human approval capabilities; the package does not provide those tools.

    After purchasing, install the skill from Agensi into a compatible Codex environment, then invoke $loop-engineering or ask Codex to use it. It starts a goal only when your prompt explicitly asks to start or execute one.

    For T3 work, the skill marks human authorization as required and preserves the host's confirmation and sandbox controls. It cannot force an interactive approval prompt if the host does not provide one.

    The ZIP contains SKILL.md, agents/openai.yaml, two reference guides, and an Agensi license notice. It includes no scripts or MCP server. The public GitHub edition remains MIT licensed; prior grants for its copies are unchanged.

    New in Agents & Orchestration