Minimal, opinionated skills for AI agents.
Work in progress. I'm continuing to iterate on the skills and workflows.
bStack aims to give agents more of the work between discovering a problem and merging its fix. Don't spend an afternoon solving something the agent could figure out in minutes. Give it the problem first. If it gets stuck, put on your thinking hat. Until then, spare yourself the hat hair. Let it investigate, implement, test, and follow the PR through review. This follows the approach in Theo's walkthrough.
The agent opens a draft PR, listens for your comments, checks its fixes, and resolves addressed threads. Once you mark the PR ready, it explains new feedback and waits for your decision. It uses the host's PR listeners when available and merges only with your authorization.
| Skill | What it does |
|---|---|
| unslop | Cut filler and template language from prose. |
| bro | Restate the last response in plain language. |
| coding-best-practices | Deliver verified changes with small diffs and readable TypeScript. |
| no-comments | Remove redundant comments and flag unclear code. |
| yeet | Open a draft PR and work through your review comments. |
| time-report | Estimate time spent on tickets from agent conversation history. |
bunx --bun skills add 0xr3ngar/bstackOr use npx skills add 0xr3ngar/bstack. Choose the skills and agent when prompted. Add --global to install across projects, or --skill unslop to pick one skill.
Philosophy · Contributing · Releases · MIT license · Third-party notices