A second push, on autopilot. blog.bumper turns the structured end-of-task reports your
coding agent already writes into published blog posts — no second context-switch, no copy-paste,
no "I'll write it up later" that never happens.
The friction that kills most dev blogs is the second push: you finish the work, you ship it, and
then you're supposed to stop, switch gears, and write about it. blog.bumper removes that step. Your
agent (Claude Code, or anything that can post a message) drops a structured report into a chat
channel. bumper reads it, turns it into a validated MDX post, commits it to your blog repo, and
pushes — and your host (Vercel, etc.) builds it live. The report you were already writing becomes
the post.
It's deliberately loosely coupled: bumper triggers off a chat message, never off your build.
A bumper failure can't break your work, and a skipped report just means there's nothing new to
post. There's no AI summarization in the loop — the report is already structured, so the post is
deterministic and yours. Today the chat source is Discord and the host is Vercel, but the
system is built around roles (a chat source, an agent, a content repo, a host) so other adapters —
Telegram, others — slot in without reworking the core.
Read in this order. Each builds on the last.
| Doc | What it covers | When to read |
|---|---|---|
| docs/INSTALL.md | Step-by-step setup: prerequisites, the Discord bot, the wizard, the config, and your first dry-run + bump. | When you're ready to install. |
| docs/OPERATION.md | Day-to-day use: the commands, the approval gate, removing posts (unbump), and the project registry. |
When you're running bumper. |
| docs/HOW_IT_WORKS.md | The pipeline stage-by-stage — what happens inside a bump, with the flow diagram. | When you want to understand or modify behavior. |
| docs/ARCHITECTURE.md | The system in context: how bumper sits between your chat source, agent, GitHub, and host. The diagrams, trust boundaries, and the roles model that lets you swap Discord for something else. |
When you're wiring it into your own stack, or extending it. |
| docs/CONFIG.md | Every .bumper.toml field, with defaults and purpose. |
When you're configuring bumper for your repos. |
| docs/CHANGELOG_CONTRACT.md | The exact report format your agent must post. The load-bearing interface. | When you're setting up your agent to post reports. |
| docs/DISCORD_SETUP.md | Creating the Discord bot, token, permissions, and channel IDs. | When you're setting up the Discord side. |
| docs/TROUBLESHOOTING.md | Symptom-first diagnosis when something doesn't behave. | When a bump doesn't do what you expected. |
bumper sits at the boundary of a few external services. The tricky parts of setup are usually on
their side, not bumper's. Links to the authoritative, current docs for each:
- A Discord bot + token —
bumperreads your report channel and writes traces to a debug channel. Create an application and bot, and copy its token, in the Discord Developer Portal. You'll need the Bot permissions Read Message History (on your report channel) and Send Messages (on your debug channel). To copy channel/message IDs, enable Developer Mode in Discord (User Settings → Advanced). - A blog repo your host builds from — any Git repo your static host watches.
bumperwrites MDX into it. - A host with a deploy-on-push setup — e.g. Vercel, which builds and deploys when the watched branch updates.
- Node 22+ and npm to run
bumperitself.
If something breaks at one of these boundaries during setup, it's almost always a permission scope, a wrong channel ID, or a host build setting — the troubleshooting guide covers the common ones.
blog.bumper is open source and free. If you use it, consider keeping the small bitmosh.dev
attribution in your site footer — and here's the actual reason, beyond credit:
That tag can hyperlink to a post on bitmosh.dev showcasing your use of the system. The point
isn't empty recognition — it's that we genuinely want to see how people are running and modifying
this. Keeping the tag keeps the loop open: it's how interesting builds get found and featured, and
how the project learns what people actually do with it. You're free to strip it entirely — it's
your site — but if you leave it, you're part of the visible community using it, not just a silent
clone.
blog.bumper is small because it stands on excellent open-source work: commander (CLI) · inquirer (prompts) · simple-git (git) · zod (validation) · gray-matter & yaml (frontmatter) · smol-toml (config) · built and tested with TypeScript, vitest, and tsx.
- Working today: Discord → agent report → MDX → GitHub → Vercel, with a review-branch flow and injection-safe rendering.
- Planned: additional chat-source adapters (Telegram and others), a setup-wizard CLI, and a template system for different post types (release notes, now-playing/Spotify posts, link posts) so the agent can categorize a report and pick the right template.
Contributions, adapters, and "here's how I'm using it" notes are all welcome.
MIT licensed. Built and maintained alongside bitmosh.dev.