Set up an orchestration pattern where an expensive model judges and plans while cheaper agents do the heavy research and coding.
Automatically pause a long-running agent session before it exhausts its usage budget mid-task.
Turn a text plan into an interactive visual plan with diagrams and file maps for human review.
Turn a pull request diff into an annotated visual recap so reviewers can see the shape of a change quickly.
| builderio/skills | spupuz/vibenvr | nodejs/github-bot | |
|---|---|---|---|
| Stars | 292 | 287 | 299 |
| Language | JavaScript | JavaScript | JavaScript |
| Last pushed | — | — | 2026-06-29 |
| Maintenance | — | — | Active |
| Setup difficulty | — | moderate | hard |
| Complexity | — | 3/5 | 4/5 |
| Audience | developer | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Skills for coding agents is a small collection of composable "skills" you can install for AI coding assistants like Claude Code or Codex. Rather than a large process framework you must adopt wholesale, it is a set of separate pieces you can pick from and adapt to your own project, aimed at teams who want their coding agent to stay sharp on tasks that need judgment, such as orchestration, review, planning, validation, and clear status updates. The collection includes five named skills. Efficient Fable and its more general sibling Efficient Frontier set up an orchestration pattern where an expensive, high-capability model acts as the architect, synthesizer, and final judge, while cheaper, lighter agents handle token-heavy work like research, coding, testing, and reducing long log output. The goal is to avoid spending expensive model time on tasks a cheaper agent could do just as well. Stay Within Limits checks a user's current 5-hour and weekly usage before starting substantial work, and pauses new work once usage crosses 95 percent until there is enough room left in the window to continue. This targets long-running agent sessions that might otherwise blow through a usage budget mid-task instead of stopping cleanly. Visual Plan converts an ordinary text plan into an interactive visual plan with diagrams, file maps, annotated code, and open questions, so a human reviewer can approve a plan before any code changes start. Visual Recap does something similar for completed work, turning a branch, commit, or pull request diff into an interactive recap with annotated diffs, diagrams, and focused review notes, so reviewers can understand a change's shape before reading every line. Quick Recap adds a simple convention: every completed agent response ends with a colored status marker, green for done, yellow for done but needing a specific follow-up step, or red for blocked, removing ambiguity about whether more action is needed. Installation runs through an npx command that offers a picker for which skills to install, which agent platform to target, and whether to add managed instruction blocks or a companion GitHub Action for visual pull request recaps.
Skills for coding agents is an installable set of five composable skills, covering model-cost orchestration, usage-limit pausing, and visual plan and diff recaps for AI coding assistants.
Mainly JavaScript. The stack also includes JavaScript, Claude Code, Codex.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.