by gintasz
AI agent harness tool allowing it to wipe its own session context and re-run the first message
# Add to your Claude Code skills
git clone https://github.com/gintasz/neuralyzerneuralyzer is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by gintasz. AI agent harness tool allowing it to wipe its own session context and re-run the first message. It has 58 GitHub stars.
neuralyzer's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/gintasz/neuralyzer" and add it to your Claude Code skills directory (see the Installation section above).
neuralyzer is primarily written in TypeScript. It is open-source under gintasz on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh neuralyzer against similar tools.
No comments yet. Be the first to share your thoughts!
Unlocks once the catalog security scan passes (runs nightly).
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
Watch the first minute of this video as an introduction. This extension adds 1 tool for AI agent harness to call, named neuralyzer (no arguments). When the tool is called, all user and assistant messages in the session context are wiped, and a copy of the first message is sent again. Example:
USER: Hi, how are you?
ASSISTANT: Good. How can I help?
USER: Call neuralyzer tool
🕶️✨ Neuralyzer has flashed.
USER: Hi, how are you? [sent automatically]
ASSISTANT: Ready to help!
USER: Was neuralyzer tool used in this conversation?
ASSISTANT: No, never used.
Easier and more ergonomic loop engineering. A traditional Ralph loop is basically running this command in your shell: while :; do cat PROMPT.md | pi -p ; done, but then you have to save the prompt to a file, handle loop exit conditions, or adapt your workflow to whatever a third-party tool or extension demands. The loop controller lives outside the agent. This tool gives it back to the agent. You can just send the agent a message with control flow like this:
Check if @john has submitted a GitHub PR in this repo fixing authentication bug.
If yes -> add GitHub comment to that PR saying "Thank you".
If no -> wait 5 min and call neuralyzer.
/loop keeps adding to your session's context window, causing context rot and increased session continuation cost due to more tokens being in it, whereas neuralyzer gives the agent a fresh start and makes a loop setup super easy.
Pick your harness:
pi install npm:@gintasz/pi-neuralyzer
opencode plugin @gintasz/opencode-neuralyzer
No. As of June 20, 2026, Claude Code exposes no extension surface (hooks, skills, MCP, agents) that can wipe, fork, rewind, or rewrite session context mid-session. Fork (--fork-session) and /rewind exist, but they are startup-flag / interactive-human only — an agent cannot neuralyze itself.
Don't see your harness? Adding one is the most welcome kind of PR — see CONTRIBUTING.