More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    relay

    1

    Hand off to a fresh agent session at the end of a task instead of compacting mid-work. Fully automatic version for Claude Code: free plugin on GitHub.

    Free

    1 useSecurity scanned
    relay

    relay

    Example session with this skill installed

    I'm hitting the context limit and the agent is getting slow. Let's hand off the current auth refactor so I can start a fresh session.

    • Read your context and instructions
    • Compiled the relay

    I have serialized the current state of the authentication refactor. I saved the handoff as .claude/relay/handoffs/20261020-141200-auth-refactor.md with the pending migration steps and a .gitignore inside that folder, so the handoff never ends up in git. Your own .gitignore is untouched.

    Auth Refactor Handoff

    SectionDetails
    GoalMigrate JWT handling from local storage to HttpOnly cookies for the /api/v1/auth endpoints.
    Current StateMiddleware updated in src/auth/middleware.ts. Verified cookie extraction with Vitest.
    Next Steps1. Update login.ts controller to set headers. 2. Remove localStorage calls from AuthContext.tsx.
    Key Filessrc/auth/middleware.ts (new logic), src/controllers/login.ts (pending).
    ConstraintsAvoided js-cookie library to minimize dependencies; using native Set-Cookie headers.
    Open QuestionsShould we implement a refresh token rotation in this session or the next?

    Next steps

    1. Clear this conversation or start a new session.
    2. Type "resume the handoff" to pick up exactly where we left off.
    3. Working in several terminals at once? Each session gets its own handoff file, and a new session lists the waiting ones and asks which to continue.

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

    About this skill

    Using Claude Code? Get the plugin instead, it is better. This skill is the portable, manual version for any agent. The free relay plugin on GitHub (github.com/4mierS/relay) uses Claude Code hooks to do everything automatically. It measures the real context size, asks for a handoff when a task is finished, saves it without a permission prompt, and offers it after /clear in the same terminal.

    Maintain context and momentum during long development sessions without hitting the performance degradation of large context windows. Long sessions often lead to model compaction, causing the agent to lose track of subtle details, rejected approaches, and specific variable names. This skill formalizes a handoff process that serializes the current project state, pending tasks, and architectural decisions into a lightweight transfer file.

    What it does

    • State serialization creates a structured Markdown handoff containing goals, verified progress, and rejected paths.
    • Session bridging allows for a complete context clear while retaining critical project intelligence for the fresh agent.
    • Automated housekeeping manages handoff files in a hidden directory and ensures they are excluded from version control.
    • Resume logic detects existing handoff files to instantly restore project momentum in a new chat.
    • Parallel sessions every handoff is its own file, so several sessions in the same project never overwrite each other. If more than one is waiting, the agent asks which one to continue.
    • Branch check records the git branch and commit, and warns when you resume on a different branch.

    How it works

    1. Trigger the handoff when a task is partially complete or the context window is near capacity.
    2. Review the generated handoff in .claude/relay/handoffs/, which summarizes the technical state and next steps.
    3. Reset the agent session or start a new conversation in the same workspace.
    4. Resume by prompting the new agent to pick up the relay, which reads the handoff and moves it to .claude/relay/done/ so no other session picks it up.

    Frameworks & tools

    This skill works with any codebase managed by AI coding assistants like Claude Code or Cursor. It uses local file system operations and .gitignore configurations.

    Why this beats prompting it yourself

    Manual handoffs are prone to omitting the "why" behind rejected code patterns, leading the new agent to repeat previous mistakes. This skill enforces a schema that captures technical constraints and specific file paths that manual summaries often skip.

    Use cases

    • Context management when the agent starts hallucinating or forgetting earlier instructions due to session length.
    • Task suspension when you need to stop work for the day and want a "save point" for tomorrow.
    • Role switching when moving from a heavy refactoring phase to a feature implementation phase.

    Known limitations

    The handoff file is limited to 6000 characters to ensure it fits easily into the start of a fresh context window. It requires file system write access to the project root.

    Skill vs. plugin

    This skillrelay plugin (GitHub)
    Works inAny agent with SKILL.md supportClaude Code
    When the handoff happensWhen you ask for itAutomatically, based on the real context size, at the end of a task
    After clearing the contextYou type "resume the handoff"The new session offers it on its own
    SavingThe agent writes the fileThe hook saves it, no permission prompt
    PriceFreeFree, open source (MIT)

    Install the plugin in Claude Code:

    /plugin marketplace add 4mierS/relay
    /plugin install relay@relay
    

    Handoff summaries are capped at 6000 characters. Requires file system write permissions to the project root and a .claude directory.

    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

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 27 days ago

    • Free with an Agensi account

    Permissions

    Read Files
    Write Files

    File Scopes

    .claude/relay/**

    Read/Write Files: the skill only reads and writes its own handoff files in .claude/relay/ (handoffs/, done/ and a .gitignore). No shell commands, no network access.

    Listed27 days ago

    What's inside

    Frequently Asked Questions

    Relay solves 'context bloat' by serializing your project's technical state into a lightweight file, allowing you to start a fresh session without losing track of your goals, rejected code paths, or specific architectural decisions.

    Yes, Relay is designed to work with any terminal-based or IDE-integrated AI coding assistant, such as Claude Code or Cursor, provided the agent has permissions to read and write to your local file system.

    No. Handoffs live in a hidden `.claude/relay/` folder in your project root. The skill also creates a `.gitignore` containing `*` inside that folder, so handoff files never get committed. Your own `.gitignore` is not touched.

    While standard prompts often miss technical nuances, Relay enforces a specific schema that captures 'rejected paths' and specific file-state metadata, ensuring the new agent doesn't repeat past mistakes or hallucinate progress.

    To keep the fresh context window lean, Relay enforces a 6000-character limit on handoff files, focusing purely on high-density technical intelligence rather than redundant code snippets.

    Yes, for Claude Code. relay is also available as a free open-source Claude Code plugin (github.com/4mierS/relay). It uses hooks to ask for a handoff automatically once the context passes a token threshold, and after /clear the new session offers to continue from it. This skill is the portable version: you trigger the handoff yourself, and it works in any agent that supports SKILL.md.

    Every handoff is its own file in .claude/relay/handoffs/, named with the time and title, so parallel sessions never overwrite each other. When you resume and more than one is waiting, the agent lists them and asks which one to continue. The one you pick is moved to .claude/relay/done/ so no other session takes it.

    New in Agents & Orchestration