craft-skills

Plugin craft-skills in JNU-SWCU/oss-hub. Commands read from the repo's manifest — today.

Skills (45)

• api

  Defines and evolves public HTTP API contracts while preserving published incumbent behavior. Use when asked to design the public REST contract for a resource, document an endpoint contract, choose API pagination or error shapes, standardize a greenfield REST API, or API 계약을 설계할 때. Not for service st

• architect

  Sketches types, signatures, caller usage, and module boundaries before implementation and stays in the loop while code fills in the chosen shape. Use for \"/architect\", \"architect this\", \"design this\", or non-trivial work where jumping to code would lock in the wrong structure. Grounds through

• arena

  Runs N candidates at the same task, picks a base, and grafts the strongest parts into one verified artifact. It applies to '/arena', 'arena this', 'throw it in the arena', and design choices where one attempt could lock in the wrong shape. It fans out parallel subagents across model families and has

• automate-me

  Drafts or updates one personal <handle>-mode skill from the user's working conventions, mined from the active workspace's recent transcripts and asked directly, and runs unslop on its prose. It applies to 'automate me', 'update my mode skill', 'capture my preferences', and 'work in my style'. It upd

• benchmark-checklist

  Vets performance measurements for the limiter, tuning, physical limits, errors, repeatability, relevance, and proof that timed work happened. It applies to 'check this benchmark', 'report this speedup', measured regressions, and choosing a library or configuration from numbers. It reports spread and

• blast-radius

  Finds what a change could break beyond the diff and proves the safety fact by running real code. It applies to 'blast radius of X', 'what could this break', and small diffs whose callers or wire formats hide risk. It returns evidenced risks, cleared cases, and a proof or an explicit unproven fact. N

• bro

  Restates the assistant's last message in plain human language without jargon, keeping its meaning while making it shorter and easier to understand. Use for \"/bro\", \"say that simply\", \"drop the jargon\", \"what are you actually saying\", or \"explain your last message like a human\". Applies onl

• browser

  Owns personal composition for live authenticated browser work through Aside as the sole managed route. Use when a request says "Use Aside to inspect my logged-in dashboard", "inspect my signed-in dashboard", "open this in my browser", "click this button", "fill out this form", "continue this Aside s

• cicd

  Designs CI/CD changes that preserve the repository's delivery topology and make releases observable and reversible. Use when asked to set up the PR pipeline and deployment for this repo, configure CI/CD, add a deployment pipeline, define required CI checks, design release rollback, or 배포 파이프라인을 설계할

• correct

  Finds repeated mistake classes and changes the repository so agents cannot repeat them. Use for \"/correct\", \"agents keep leaving debug prints\", \"make this correction stick\", or recurring review fixes with two or more observed incidents. Chooses architecture, types, lint/CI, tests, then docs an

• craft-help

  Guides users through craft-skills setup, craft-mode, and picking the skill, playbook, or principle for a task, then hands back a prompt they can send and the file the answer came from. Use for "craft-help", "which skill should I use", "how do I use craft-mode", "which playbook fits this", "how do I

• craft-mode

  Routes multi-step engineering work through a chosen playbook, situational workflows, and domain-owned principles. Use for craft-mode, a rigorous feature or bug fix, a migration, a measured performance problem, or work that needs a clear finish condition and evidence. Keeps replies concise, applies u

• db

  Diagnoses database workloads and anti-patterns and makes evidence-backed schema and operating decisions after persistence selection. Use when a slow query needs EXPLAIN review, bulk DML or an online schema change needs a safe plan, keys, FKs, JSON, indexes, partitioning, pooling, replication, cache,

• debug

  Diagnoses a failing program under a hypothesis-driven loop: reproduce the failure before theorizing, log observed facts separately from inferences, hold competing hypotheses until the cheapest probe discriminates between them, and confirm the mechanism with instrumentation before any fix lands. Use

• design

  Owns canonical DESIGN.md artifacts and evidence-first UX/UI judgment for coherent product design roots. Use when requests ask to define interface direction; choose type, color, spacing, or motion; audit a user journey; establish design tokens or state specifications; redesign information hierarchy;

• distil

  Distils transferable rules and conventions from an external source — a well-crafted repo, an engineering article, an AGENTS.md, or a third-party skill — and lands them in the owning package with provenance recorded. Use when the user says "파쿠리", "distil the rules from this repo", "absorb this skill"

• document

  Scaffolds and authors repository documentation through a six-type ontology and canonical artifacts. Use when asked to scaffold repository docs, "record this decision", "where does this spec go", "draft the project CHANGELOG", or "comment-the-why". Not for READMEs (use readme), conducting research (u

• figure-it-out

  Designs an auditable playbook when no narrower workflow fits a large migration, multi-part change, or unattended run. It applies to '/figure-it-out', 'figure it out', and work a human reviews after stepping away. It scales rigor, tests hypotheses against real artifacts, and records decisions through

• git

  Guides version-control craft: a ground-truth and incumbent-style detection gate before the first git mutation, the atomic-commit `git add -p` split protocol, commit/branch/PR conventions matched to the repo''s own history, and non-interactive-safe history surgery (fixup, reword, split, scripted bise

• gpu

  Applies GPU environment and resource discipline — probe the hardware before choosing any install, budget the host before launching any job — to CUDA/PyTorch setup, attention-backend builds, and GPU job launches. Use when setting up CUDA or PyTorch on a new GPU machine, asked to "install flash attent

• how

  Explains how a system or process works by tracing entry points, data flow, boundaries, and ownership in real source evidence. Use for \"how does X work\", \"walk me through this code\", \"where should this live\", \"which package owns this\", or onboarding to a subsystem before changing it. Scales e

• init

  Maps a repository into a maintained hierarchical AGENTS.md knowledge base. Use when asked to "init this repo" for AGENTS, deep-init a codebase, generate or update AGENTS.md, map repository conventions, audit existing AGENTS coverage, or report stale managed AGENTS regions. Not for package-manager or

• interrogate

  Has adversaries try to break a change or design and keeps only the findings a run proves. It applies to 'interrogate', 'adversarial review', 'multi-model review', 'challenge this', 'stress test this code', 'find blind spots', and 'tear this apart'. It reports verified findings without applying fixes

• issue-triage

  Decides what to work on: intake triage of one incoming report (pasted message, linked issue, or chat thread) with dedupe and a create-or-not tracker decision, and refinement of existing backlog issues with rule-citing verdicts and routing. Use to triage this issue report, triage the backlog, ask is

• ml

  Applies ML/DL research engineering discipline — reproducible project layout, leakage-safe dataset construction, and a training-discipline ladder — to classical ML, deep learning, fine-tuning, and vision work. Use when scaffolding a new ML project, asked to "build a dataset" or "데이터셋 구축", running or

• orca

  Organizes Orca around low-clutter project identities and preserves running work while diagnosing which local connection layer still needs evidence. Use to "organize Orca", "reduce sidebar clutter", consolidate the "same repo across hosts", clean old branches in Orca, reuse or account for task termin

• principle-architecture

  Guides cross-domain structural decisions before implementation. Use when choosing core data structures, integrating a requirement into an existing design, comparing architectures without precedent, planning a rewrite, retiring internal APIs, or isolating concurrent writers to a file or branch. Servi

• principle-backend

  Routes backend service architecture and persistence selection, including engine, provider, ORM, production major, roles, destructive-target proof, persistence implementation, schema-ledger retirement, and folder conventions. Use when building a service layer, choosing a persistence stack, adding a r

• principle-frontend

  Routes frontend engineering through rendering, ownership, reuse, state, CSS, and performance decisions. Use when building or reorganizing a React/Vue/Svelte UI ("프론트엔드 구조 잡아줘"); choosing a React + Vite or Next.js shell, folder/public-API, or server/client boundary; improving component reuse or state

• principle-programming

  Guides correctness-first, type-strict Python and TypeScript implementation. Use when asked to write a `.py` or `.ts` file, scaffold a Python/TypeScript project, add strict types, assess an implementation diff for correctness or type holes, or fix a reproducible defect. Not for smell-only assessment

• principle-testing

  Designs, improves, and audits test suites around behavior and risk, independent oracles, counterfactual evidence, deterministic diagnosis, and cost. Use for generated-test review, unit/component/integration/e2e or smoke placement, test-suite health, surviving-mutant triage on changed code, flaky-tes

• readme

  Authors or overhauls a repository README as a showcase front page with a hero, tagline, language switch, accurate badges, navigation, previews, features, quick start, and collapsed depth, checks every claim against the repository, and fills the community files it links. Use when asked to "write the

• recall

  Reconstructs recent working context from the active workspace's chat history, scoped shared records, and live state. It applies to 'recall my work on X', 'catch me up', 'what have I been working on', and 'where did I leave off'. It returns a cited current-state brief and reads another project's tran

• refactor

  Restructures code without changing what it does — extracting functions, renaming, removing duplication, flattening nested conditionals, and other mechanical moves backed by a detection command and threshold. Use when the user says \"refactor this\", \"clean up this code\", \"리팩토링 해줘\", or \"this fun

• reflect

  Reviews durable learnings from the current session through judgment, tooling, and divergent lenses, then routes each accepted learning to an edit on an existing skill. Use for \"reflect\", \"/reflect\", \"what did this session teach us\", or \"capture the workflow lessons from this conversation\". R

• research

  Researches an open question by reframing it first: rewrites the ask as one problem statement, defines its terms, and splits it into sub-questions tied to established topics, then sweeps trustworthy primary sources, verifies claims by risk, and replies with cited findings, options side by side, and h

• security

  Finds and fixes vulnerabilities in code the user owns across web, API, and LLM surfaces, and owns confidentiality for authorized remote credential handoffs. Use for a security review, "is this safe to ship," "check for vulnerabilities," "보안 점검," secrets hygiene, dependency risk, PR security regressi

• show-me-your-work

  Keeps a reviewable TSV decision trail for long-running, autonomous, or multi-phase work, with evidence pointers and outcomes. It applies to '/show-me-your-work', 'keep a decision log', and work a human reviews after stepping away. It audits each run's rows and keeps the file local unless a reviewer

• swarm

  Coordinates bounded workers across independent slices or declared races, drains their results, and returns one evidence-backed report. It applies to '/swarm', 'swarm this', parallel coverage, gauntlets, and exploration. It respawns a worker once when its result misses the brief and records a second

• tdd

  Executes a focused red-to-green bug fix when the user asks for \"TDD\", \"write a failing test first\", \"add a regression test\", or the defect has an obvious cheap local test target. Writes and runs the regression check for the intended failure before editing production code, then makes the minimu

• teach

  Teaches a body of work in plain language at the person's pace by composing how and why into one account of what it is, how it works, and why it has that shape. Use for \"teach me this\", \"help me really understand X\", or \"explain this change or subsystem to me\" when understanding, not editing, i

• unslop

  Removes AI writing patterns while preserving meaning and the intended tone. It applies to all prose the agent writes, including replies, docs, READMEs, PR descriptions, and commit messages, and to 'remove AI tells', 'rewrite this plainly', or 'unslop this'. It keeps stable numbered rule ids for call

• why

  Investigates why code or a decision has its current shape through available source history, tickets, documents, chat, telemetry, error records, and analytics. Use for \"why does X work this way\", \"why did we pick Y\", design rationale, postmortems, regressions, or \"where did this threshold come f

• write-prd

  Authors product requirements documents by filling a provided or packaged PRD template from product context, with One Pager, scope, metrics, rollout, open issues, and history kept coherent. Use when asked to "write a PRD", "fill this PRD template", "turn this product idea into a PRD", "complete the p

• write-report

  Scaffolds and authors a project''s one-off canonical technical report against a single YAML frame (technical-report.yaml) whose depth is the enforced table of contents — section, then heading, then required must-content. Use when asked to "scaffold a technical report" or "기술 보고서", when writing or re

Similar plugins