Steer a Running Plan

    1

    Pause, kill, or rewrite a branch of a live agent plan without stopping the rest of the work. Review gates can move before launch. Drafts at a gate can be approved, edited, or rejected.

    $15

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Steer a Running Plan

    Example session with this skill installed

    Kill the web-research branch and change the instructions for the summary-node to focus only on the social-media data.

    • Read your context and instructions
    • Compiled the steer a running

    I have applied two interventions to the live plan. The web-research branch has been terminated, and the summary-node now has updated instructions.

    Plan Intervention Log

    ActionTargetStatusDetail
    kill_branchweb-researchExecutedNodes web-1, web-2, and verify-web marked as killed.
    edit_instructionssummary-nodeExecutedInstructions updated: "Focus only on the social-media data."

    Next steps

    • Monitor the social-media branch as it approaches the human_gate.
    • Verify the final summary output in output/final_report.md once the summary node completes.
    • Check plan.json to see the updated DAG state and intervention timestamps.

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

    About this skill

    Once a plan is in motion, most agents only understand stop or keep going. This skill gives you six precise controls on the same map: approve the plan, move a review gate before launch, pause a branch while others keep working, resume it, kill a branch (anything that only depended on it dies instead of hanging), and edit one node's instructions without rebuilding the graph. Human gates stay blocked until every item is approved, edited, or rejected, and the skill will not auto-approve. Every action is written to the same event log the agent uses, so the audit trail shows you stepped in. Where it stops: it does not invent a plan, and it won't steer a run that's already finished.

    Cannot move gates after plan approval. Does not support full plan redesigns. Requires explicit branch names from the plan graph. Commands only apply to one intervention at a time.

    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 19 days ago

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Listed19 days ago

    What's inside

    Frequently Asked Questions

    This skill allows you to intervene in live multi-agent workflows by pausing, killing, or editing specific branches without crashing the entire run. It solves the 'all-or-nothing' execution problem, letting you stop a wasteful research branch or fix a node's instructions mid-stream to save time and tokens.

    The skill is designed for any multi-agent system that utilizes a DAG-based orchestration layer and a plan.json file for state management. It works across various agent frameworks as long as they can interface with Python-based graph manipulation.

    You will receive the core steering logic, a specialized Python script for mechanical graph updates, and the implementation guidelines for integrating checkpoint enforcement into your existing subagents. It also includes the six primary intervention actions ready for deployment.

    No, for safety and stability, human review gates must be moved before the plan is officially approved and launched. Once the execution reaches a gate, you can resolve it (approve/edit/reject), but you cannot shift its position in the DAG once that specific branch is live.

    Setup involves integrating the provided Python script into your orchestration layer and ensuring your subagents are configured to check for status signals (pause/kill) between units of work. The skill is designed to plug into existing plan.json architectures with minimal refactoring.

    New in Agents & Orchestration