Agentcrumbs — AI skill for Claude Code

agentcrumbs

AI agents can read your code but they can't see what happened at runtime. agentcrumbs fixes that. Agents drop structured traces inline as they write code. When something breaks, the agent queries those traces and sees exactly what ran, with what data, in what order.

Crumbs are development-only. They get stripped before merge and cost nothing when disabled.

Service A ──┐                                ┌── $ agentcrumbs tail
Service B ──┤── fetch() ──> Collector :8374 ──┤── $ agentcrumbs query --since 5m
Service C ──┘  (fire & forget)               └── ~/.agentcrumbs//crumbs.jsonl

Getting started

npm install agentcrumbs

Then tell your agent: **"Run `npx @tanstack/intent@latest install` to set up agentcrumbs skills, then run the agentcrumbs/init skill."**

The agent will wire the skills into your agent config (CLAUDE.md, .cursorrules, etc.), then scan your repo to build a namespace catalog so all agents use consistent names.

Agent skills

agentcrumbs ships with [@tanstack/intent](https://tanstack.com/blog/from-docs-to-agents) skills inside the npm package. When your agent runs `npx @tanstack/intent install`, it sets up skill-to-task mappings in your agent config so it knows when to load agentcrumbs patterns.

Skill	What it teaches
agentcrumbs	Core workflow, API, markers, CLI reference, common mistakes
agentcrumbs/init	Scans repo, discovers namespaces, writes config

Skills travel with the package version. The agent always has docs matching the installed code.

How it works

The agent writes crumbs as part of the code it's implementing:

import { trail } from "agentcrumbs"; // @crumbs
const crumb = trail("auth-service"); // @crumbs

export async function handleLogin(token: string) {
  crumb("login attempt", { tokenPrefix: token.slice(0, 8) }); // @crumbs

  const user = await validateToken(token);

  crumb("login success", { userId: user.id }); // @crumbs
  return us