A raw idea
working feature
Skills that give an AI coding agent the habits of a careful senior engineer. One command installs all 47 into the agent you already use.
npx skills add AI-Degen-69/issue-to-pr-skills
Onboarding Guide · Table of Contents 6 chapters ↓
AI agents default to the shortest path.
These skills close the shortcut.
Proof before review
Station #4 refuses to review unproven code. Station #3 runs the failing test first and never pushes — shipping is a different station’s job.
Never invent
Missing reviewer? Skipped and recorded. Missing trigger ack? Reported as missing. No simulation, no assumption.
Spec axis separately
“Follows every standard but does the wrong thing” is a different finding from “breaks standards”. They never merge.
Quota-conscious
One focused review round per PR. On rate-limit, reuse prior evidence + delta check before fresh review.
Give your agent a senior engineer’s instincts.
One command installs all 47 skills into 70+ agents. Or grab one station.
npx skills add AI-Degen-69/issue-to-pr-skills
--list · One skill: --skill iv-review-build-and-pr/plugin marketplace add AI-Degen-69/issue-to-pr-skills
/plugin install issue-to-pr-skills@ai-degen-marketplace
cp -r skills/… → .agents/skills/
/i-pick-issue → pick issue → follow stations.npm run validate: frontmatter + links. CI runs on every PR.Project structure: where things live
| Layer | Paths | Purpose |
|---|---|---|
| Shared core | skills/ (47) | Portable SKILL.md workflows |
| Reviewers | agents/ (17) | Specialist personas |
| References | references/ | Checklists loaded on demand |
| Docs | docs/ | Getting started, pipeline, agents |
| Adapters | .claude-plugin/, plugin.json | Claude Code & Codex marketplace |
| Site | site/ | This website (GitHub Pages) |
How it feels in practice: one issue, start to finish
- 1. You paste an idea
/create-issue add dark mode. The skill researches the repo, uses real paths with line numbers, and publishes aready-for-agentissue. - 2. You pick + plan
/i-pick-issueshows backlog grouped by domain./ii-plan-issue 123sizes it, locksCONSTRAINTS.md, writestasks/plan.mdslices. - 3. It builds, reviews, merges
/iii-build-plan autoTDDs each slice./iv-review…proves then reviews, opens PR, triggers @coderabbitai./v-babysit…merges./present-prbuilds your showcase on request.
One command per phase. Skills load automatically.
Stations are narrow. Each one has an entry condition, a verification gate, and a plain-English report that ends with the single next step. No station simulates missing work.
tasks/plan.md.System & Ad-hoc Skills
These 4 skills support the pipeline without being a numbered station in the sequential #1→#6 chain: a pre-flight state gate before Station #1, an intake branch off Station #1, a fast lane that diverts trivial work straight past the chain, and an ad-hoc visual showcase runnable any time.
Ship past a panel, not a rubber stamp
17 narrow experts. The AI picks only the ones that match your change. In plain words: a panel beats a single stamp. The full roster, with simple explanations, lives on its own page.
47 skills spanning the whole delivery
On this page, a taste of the recipes. The full searchable catalog, with plain-language explainers, lives on its own page, built for non-coders too.