>_ yoshiko-flow
portable, beads-backed agent skills + the yf CLI
yoshiko-flow is a family of portable, cross-harness AI-agent skills plus a single
compiled CLI, yf, that installs, upgrades, verifies, and preflights those skills and the
toolchain they depend on. Beads-backed planning and research, drift checking, and markdown
tooling — one binary, embedded skills, installed into Claude Code or any agent harness.
$ curl --proto '=https' --tlsv1.2 -LsSf https://yoshikoflow.sh/install.sh | sh
# plan →
Structured, beads-tracked planning: scope → investigate → plan → execute in a git worktree, with upstream issue reconciliation.
/yf-plan build the auth service
# research →
Multi-phase, citation-backed deep research that decomposes a topic into a resumable DAG of focused subtasks.
/yf-research compare vector databases
# beads-backed →
Every stateful skill tracks its work in beads (bd), not scratch TODOs — durable, resumable, and pushable upstream.
bd ready --json
# skill lifecycle →
Install → preflight → invoke → coordinate/execute. One shared kernel gates every skill before it runs.
yf preflight yf-plan --json
# one binary, embedded skills →
The whole skill tree is embedded in the yf binary at build time — installing skills needs no network access or repo clone.
yf skills install
# self-updating →
A vendor install updates itself in place — binary and every installed skill in one step. GitHub Releases stays canonical.
yf self update