Reasoning-first development for Claude Code. Git is the source of truth; status is derived, never stored. YOLO turns "I want X" into a planned, executed, verified, and landed change, driven by conversation and git rather than tracked status files.
Turn one hard "which should I do?" into a judged tournament — and get back a defensible winner. Terminal summary from a real run — verdict and scores are from the saved report, re-rendered at a readable pace. bakeoff is a Claude Code skill.
@corichuiFAQ
yolo is a Claude Code plugin with 11 hand-picked skills for development work, indexed on Flowy. Install it with the command on its page. It includes yolo-debug, yolo-decide, yolo-feature. Its skills do not fire on their own yet. Request auto-invocation to have Flowy route them as you prompt. Free and open source.
> /plugin marketplace add CoriChui/yolo> /plugin install yolo@yolo
Reasoning-first development for Claude Code. Git is the source of truth; status is derived, never stored.
YOLO turns "I want X" into a planned, executed, verified, and landed change, driven by conversation and git rather than tracked status files.
workspace/features/<slug>/brief.md)
plus a branch (feature/<slug>). Progress is recorded as commit trailers
(YOLO-Task, YOLO-Verified, YOLO-Feature) and a yolo/done/<slug> tag at landing.state.yaml.yolo-* skill that Claude routes to from plain intent. (The lone
.claude/commands/yolo-validate.md is maintainer tooling, explicit-invoke only, not part
of the workflow.)The precise, authoritative rules — naming, trailers, status derivation, gating, config
fallbacks — live in .claude/yolo/conventions.md, the single
source of truth every skill cites.
| Skill | Use it when… |
|---|---|
yolo-feature | you want to build/add/implement a feature (the main entry point) |
yolo-research | you want to understand the code before changing it |
yolo-plan | you want a goal broken into testable, committable tasks |
yolo-verify | you want work checked against its success_criteria |
yolo-finish | you want to land verified work (PR + CI check, or fast-local) |
yolo-status | you want a derived view of where every feature stands |
yolo-roadmap | you want a large goal decomposed into several feature briefs |
yolo-decide | you want a recorded, multi-perspective decision (X vs Y) |
yolo-debug | you have a bug and want a systematic reproduce → isolate → fix → verify pass |
yolo-intake | you want a feature to draw on the project's docs for context (yolo reads the docs/ folder; connectors/docs-engine fill it — yolo never fetches) |
yolo-init | you want to set up (or repair) YOLO in a repo |
yolo-feature composes the lifecycle: capture intent → brief → research → plan →
plan gate → execute → verify → ship gate. Any step is also invokable on its own.
As a Claude Code plugin (auto-updates):
/plugin marketplace add CoriChui/yolo
/plugin install yolo@yolo
Or copy it into a project (vendored, tracked in your repo):
# from your project root
git clone https://github.com/CoriChui/yolo.git /tmp/yolo && cp -r /tmp/yolo/.claude .
Then, either way:
yolo-init): it scaffolds workspace/config.yaml and
workspace/{features,decisions,debug}/ (plus docs/), and installs the routing block into your
CLAUDE.md. It writes only project state — as a plugin, the framework stays installed and updates
through the marketplace.you: add CSV export to the reports page
yolo: (drafts brief: workspace/features/csv-export/brief.md → status "planned")
git switch -c feature/csv-export → status "in-progress"
research → plan.md
Here's what I found + the plan. Approve? ← plan gate (before any code)
you: yes
yolo: execute (test-first, one commit per task)
verify: all success_criteria met → YOLO-Verified: true
Open PR and merge on green? ← ship gate (irreversible)
you: ship it
yolo: merges, tags yolo/done/csv-export → status "done"
See .claude/yolo/getting-started.md for the full
walkthrough and a glossary.
workspace/config.yaml (scaffolded by yolo-init) holds genuine config — never tracked
state. The groups:
project — name, base_branch.agents — model tier per billed step (research/plan/execute/verify/finish,
plus roadmap/decide).finish — landing policy: mode (pr|local), ci (auto|require|off), and
auto_merge_on_green (headless/unattended runs only — off by default).risk — risk-classifier inputs: sensitive_paths, max_diff_lines, max_diff_files.Confirmation is not a config knob: YOLO always confirms at the two gates (plan + ship);
say "just ship it" or "walk me through each step" per-feature in prose. See
conventions.md → The two gates.
If workspace/config.yaml is absent, YOLO falls back to safe defaults (see
conventions.md → Config-absent fallback).
.claude/
.claude-plugin/
marketplace.json
plugin.json
commands/
yolo-validate.md
settings.json
skills/
yolo-debug/
SKILL.md
yolo-decide/
SKILL.md
yolo-feature/
SKILL.md
yolo-finish/
SKILL.md
yolo-init/
SKILL.md
yolo-intake/
SKILL.md
yolo-plan/
SKILL.md
yolo-research/
SKILL.md
yolo-roadmap/
SKILL.md
yolo-status/
SKILL.md
yolo-verify/
SKILL.md
yolo/
conventions.md
getting-started.md
templates/
brief.md
claude-routing-block.md
config.yaml
.gitignore
CLAUDE.md
docs/
validation/
ab-judge/
rubric.md
adversarial/
findings-2026-06-27.md
behavioral/
scenarios.md
git-acceptance/
run.sh
README.md
LICENSE
README.md
workspace/
.gitkeep
decisions/
.gitkeep
auto-vs-explicit-routing.md
features/
.gitkeep
yolo-debug/
brief.md
plan.md
research.md
verification.md© 2026 Flowy · Free and open source
Built for Claude Code · Not affiliated with Anthropic