Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    antibody

    1

    Each skill is a short checklist your agent loads at the moment it is about to make that mistake again.

    Free

    Security scanned
    antibody

    antibody

    Example session with this skill installed

    Our dev server on port 3000 died while another session was running its tests. It turns out the agent ran pkill node to free a port. Write a scar for this.

    • Read your context and instructions
    • Compiled the antibody

    Summary

    Added incident #4 to INCIDENTS.md and one rule line to CLAUDE.md. Facts come from the shell history and the test log; the exact downtime is not in any log, so it is marked unknown.

    INCIDENTS.md

    
    ## #4  2026-11-03  Port cleanup killed another session's dev server
    
    **What happened:** Port 3000 was busy, so the agent ran `pkill node` to free it. That stopped every node process on the machine, including the dev server and the test runner of a second session working in the same repo.
    
    **Damage:** The second session's test run failed partway (`ECONNREFUSED 127.0.0.1:3000`) and had to be restarted. Downtime of the dev server: unknown.
    
    **Rule:** Kill by PID only, and only a process you started. Find the owner with `lsof -nP -iTCP:3000 -sTCP:LISTEN`.
    
    **Enforced by:** rule only (see question below).
    

    CLAUDE.md

    - Do not kill processes by name (pkill, killall, taskkill /IM); find the PID and kill only what you started. Why: #4
    

    Next step

    A rule alone often fails to stop a repeat. Do you want a PreToolUse hook that blocks pkill, killall and taskkill /IM commands? If yes, I will write it and fill in "Enforced by".

    INCIDENTS.md

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

    About this skill

    antibody is a pack of four skills distilled from incidents where AI coding agents broke real things. Each skill is a short checklist your agent loads at the moment it is about to make that mistake again.

    What's included

    • evidence-before-done: before the agent says done, fixed, deployed or passing, the final message must include the command it ran and the result. "I wrote the function" is not evidence. Covers schedules and hooks that exist but were never registered.
    • verify-deploy-by-sha: a green CI run is not a deploy. The agent reads the commit SHA from the running server (health endpoint, image tag, platform API) and compares it with the SHA it pushed. Polls in the background with success, failure, timeout and stall conditions.
    • many-agents-one-machine: rules for several sessions or subagents sharing one machine or checkout. Kill by PID, never by process name. Stage and commit in one command so another session's commit does not pick up your files. One heavy job at a time.
    • write-a-scar: after damage, write a blameless entry in INCIDENTS.md (what happened, damage, rule, enforced by) and add a one-line rule to CLAUDE.md or AGENTS.md that points back to it.

    Where it stops

    • These are instructions, not enforcement. The open source antibody project also has a CLI and a PreToolUse hook that block commands repeating a known incident. Those install on your machine and are not part of this listing. Source, docs and install steps: https://github.com/dooly6276/antibody
    • No scripts, binaries or secrets in the package. The skills only tell the agent what to check and what to write.

    MIT licensed. Free.

    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.

    Instructions only: nothing in this pack blocks a command by itself. Deploy verification needs the project to expose the running commit SHA (an endpoint, header, image tag or stamped file). The free RAM check examples cover macOS and Linux; on Windows the agent picks its own command.

    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 14 days ago

    • Free with an Agensi account

    Software
    Git

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files
    Listed14 days ago

    What's inside

    Frequently Asked Questions

    No. These skills are instructions the agent reads. Blocking needs the antibody CLI and its PreToolUse hook, which you install on your own machine from https://github.com/dooly6276/antibody (MIT, free).

    Yes. many-agents-one-machine covers that case: kill only the PIDs you started, stage and commit in one command, use a worktree instead of switching branches in a shared checkout, and run heavy jobs one at a time.

    Any agent that reads SKILL.md. The rules refer to CLAUDE.md and AGENTS.md, and the command examples cover macOS, Linux and Windows.

    A way to read the SHA of the running build: a health or version endpoint, a container image tag, the platform's deployment API, or a stamped file or header. If none exists, the skill tells the agent to say so and propose adding one instead of trusting CI status.

    An entry in INCIDENTS.md at the repository root (or a file you name) and one rule line in CLAUDE.md or AGENTS.md that points back to the entry. It shows you both and does not commit unless you ask.

    Four SKILL.md files and a short README. No scripts, binaries or network calls in the package itself. It is free.

    New in Agents & Orchestration

    Free