dev-skills
claudeSkills for code quality, Claude Code development, and context optimization
5 skills
A PM-led PDCA feedback loop for AI-assisted projects: /decide decision panel, standalone advise/verify/red-team panel primitives, /retrospect process loop, /sweep loop-until-clean audit with a mechanical stop verdict, /pdca-init scaffolding with a project-tailored advisor panel, the MSH ship-one-package command, a version-agnostic ADR system with a zero-dep corpus linter, a metrics-engine spec, and a muda-cutting CLAUDE.md template. Ships enforcement hooks (fan-out model guard, ADR post-edit lint, primitive spawn-log).
Plugin pdca-workflow from marketplace one21tools in one21labs/one21tools. Commands read from the repo's manifest — 14 days ago.
This plugin runs code on your machine — its hooks fire automatically on events like tool use. The exact commands are listed below; read them before installing.
adviseUse when weighing a contested judgment call — scope, threshold, design, sequencing — before acting on it. Advice only, no decision, no ADR.
decideUse when deciding a roadmap or product judgment call, or on any user feedback such as a bug report, feature ask, or behavioral observation. Runs the PM-led panel and records the decision as an ADR.
pdca-initUse when adopting the PDCA feedback workflow in a project. Scaffolds a themed CLAUDE.md and the ADR decision log, then generates a project-tailored advisor panel. Explicit-invoke only; never auto-fire.
red-teamUse when a decision, design, or change is about to be accepted and needs an adversary — especially safety-adjacent or assumption-heavy work.
retrospectUse when a session closes (standing, ADR 0081) or when shipped work felt wrong or worth learning from — never as per-PR ritual and never outside those triggers.
sweepUse when a skill, plugin, or codebase needs auditing until it is clean rather than audited once. Runs independent rounds until consecutive rounds find nothing new or a stated cap is hit, and reports which of the two happened.
verifyUse when a claim, a fix, or produced output needs independent confirmation before you rely on it or ship it.
/pdca-workflow:MSHMake Shit Happen — pick the highest-value open work and ship one cohesive package end to end, autonomously.
pmProduct Manager and accountable decider — owns the plan of record and strategy, weighs the advisor panel, and records each judgment call as an ADR (/decide).
red-teamAdversary — tries to break each accepted decision against the real product after the PM decides; the PM must respond before proceeding (/decide).
retrospectProcess-retrospective analyst — reads git history plus the session's friction and emits every REAL, systemic process improvement (zero is a valid result), each routed to its lowest home (/retrospect). Run on shipped work, not pre-decision.
tech-leadTech lead — turns an accepted PM decision (an ADR) into a buildable spec of file/function/test and directs implementation in the /decide flow.
verifierIndependent verification gate for /decide — reproduces load-bearing claims against the real code and produced output, checks the PM's [checkable] assumptions, and can block a decision on a verified correctness or safety finding.
${CLAUDE_PLUGIN_ROOT}/hooks/explicit-model-guard.sh
${CLAUDE_PLUGIN_ROOT}/hooks/spawn-log.sh
${CLAUDE_PLUGIN_ROOT}/hooks/spawn-log.sh
${CLAUDE_PLUGIN_ROOT}/hooks/adr-lint-post-edit.sh
Skills for code quality, Claude Code development, and context optimization
5 skills
Foundational engineering principles applied to software, documentation, and process, plus industrial control-narrative authoring
5 skills
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.