How does the plugin handle code changes?

It uses staleness detection; if a file is modified after a note is created, the note is marked as [STALE] to alert the next agent to re-verify the context.

What is a breadcrumb in Claude Code?

A breadcrumb is a persistent note attached to a specific file or line that communicates context, warnings, or reasoning to other AI agents or developers in future sessions.

Does it modify my actual source code?

No, Breadcrumb stores metadata externally via its own tracking system, allowing you to view notes through the CLI without cluttering your source code with extra comments.

Is Breadcrumb useful for solo developers?

Absolutely. It serves as a persistent memory for your future self, helping you remember why specific, non-obvious coding decisions were made when you return to a project.

Can I use Breadcrumb for bug reporting?

Yes, it includes specific fields for evidence-input, expected results, and actual results, making it ideal for documenting edge cases and complex bug fixes.

Breadcrumb Communication

Name: Breadcrumb Communication
Author: tylergibbs1

bytylergibbs1

•

协作工具

Facilitates agent-to-agent communication by leaving persistent, context-aware notes and warnings directly on source files.

Breadcrumb is a specialized skill designed for persistent communication between AI agents and human developers across multiple sessions. By attaching "breadcrumbs" to specific files or lines of code, it ensures that non-obvious logic, subtle bug fixes, and intentional complexities are not accidentally refactored or broken by future agents. It features staleness detection to alert users when a file has changed since a note was left, alongside evidence-based reporting to provide clear context for security-sensitive or high-risk code paths.

主要功能

01Recursive directory scanning to check for existing warnings before starting work

02Automated staleness detection using file hashing to track outdated notes

031 GitHub stars

04Evidence support including input, expected, and actual values for bug reproduction

05Line-specific anchoring for high-precision context within source files

06Persistent agent-to-agent messaging across different Claude Code sessions

使用场景

01Preventing future agents from 'optimizing' intentional logic that fixes subtle, hard-to-track bugs

02Documenting why a seemingly redundant piece of code is necessary for security or edge-case handling

03Handing off complex tasks to other AI agents with specific warnings about non-obvious implementation details

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add tylergibbs1/breadcrumb breadcrumb

For use in Claude.ai and ChatGPT

Download Skill