Docs/Examples

Pick an example

Each example is a lump running on a real repo, cut down to the pattern. Find the job closest to yours, copy the config, change the paths and the prompt, and check it with lumpcode lump-plan <lumpName> --contexts.

On this page
  1. What the good ones have in common
You want to Example
Migrate every component to a new styling system or framework A migration that shrinks as it goes
Make the same small change in hundreds of files A codemod, ten files per PR
Work through a list you generated first An inventory, then a migration
Bring every module to one standard shape, tests included One standard per module
Add tests where there are none Tests where they are missing
Fix what your code-quality server reports Static-analysis issues, grouped per file
Fix the bugs your team labels for the agent Labelled bugs, verified in a browser
Ship a roadmap of features from ticket folders A feature backlog in git
Run several backlogs with the same rules Lanes that share one factory
Turn a pile of notes into tickets Triage an inbox into lanes
Find duplicated code and refactor it Find, file, implement
Start one campaign after another is merged Wait on another lump
Never push code that fails your checks Retry until green
Skip an expensive step when it is not needed Run a step only when needed
Give the agent a running app to test against Start a dev server for the agent
Open a PR on GitHub, Azure DevOps, or GitLab Open a pull request after push
Pause a lump while its queue is full Pause when downstream is full
Run some lumps daily and others every minute Different schedules per lump

New to lumps? Start with the smoke test and Write a lump.

What the good ones have in common

  • The list shrinks by itself. Discovery skips work that is already done (a component with no <style> left, a file that already has a test), so re-running is always safe.
  • A check gates every branch. Lint, typecheck, or tests run after the agent, and failures go back to the agent with the output. See Retry until green.
  • Review stays small. numberOfContextsPerBranch groups tiny diffs; maximumNumberOfConcurrentBranches stops the worker from piling up PRs while you are away.
  • Prompts point at the repo. Long instructions live in a prompt file or a skill the repo already has (Follow @.agents/skills/…/SKILL.md), so the config stays short.

Edit this page