Workflow plugins

Artifact-driven development loop for any kind of project, run by hand or by the factory orchestrator.

8 skills · 4 agents · hooks

Sonnet이 일하고, Haiku가 찾고, Opus가 결재하는 Claude Pro용 개인 에이전트 키트

9 skills · 3 agents · hooks

Recommendation Decisioning Records — the /rdr-* skill suite + engine. Run /rdr:rdr-init once to bind a project, then /rdr:rdr-seed to start the first record.

12 skills

Shared MCP brain, parallel module research, overnight autoresearch, and team coordination

7 skills · 4 agents · hooks

Complete Spec-Driven Development flow: constitution, spec (EARS), clarification, plan, tasks, implementation, validation, and change.

3 skills · 9 commands

Claude stays mastermind and reviewer while opencode does the grunt implementation work, in an isolated git worktree, for a fraction of the cost. Claude reviews, verifies, and decides accept/reject/iterate — nothing merges without an explicit accept.

11 commands · 1 MCP server

Closed-loop project governance using nested PDCA (inner per-phase + outer project-level). 8 skills, 2 agents, 2 hooks. Delegates to superpowers, frontend-design, and ECC plugins.

9 skills · 2 agents · hooks

Full spec implementation cycles with filesystem and GitHub backends. Orchestrates spec, plan, track, implement, review, test, verify, and retro.

7 commands · 4 agents · hooks

The work loop: open a tracked piece of work, research it, plan it, argue the plan, turn it into an executable contract, ship it. Each skill hands the next exactly what it needs, and every step leaves a trail the next session can read.

12 skills

Scope a project by interview, plan each gate on plan mode with an adversarial reviewer, slot findings in without ad-hoc edits, break every gate with a fresh red-lane agent, and hand off only what the product owner ran and accepted. Hooks are silent outside projects that opt in.

7 skills · 3 agents · hooks

Eleven guided skills that take a non-technical team from an empty laptop to a judge-ready single Docker image: bootstrap, feature building, preview, bugfix, DB changes, nginx image build, proxy push, code zip for manual upload, submission check, a one-shot hackathon-deploy (build + check + zip + push + go live), and a plain-language explainer.

11 skills

Engineering workflow skills and review agents: refine a plan to convergence through a reviewer panel, staff its execution with right-sized subagent clusters, weigh design options, port reference algorithms rigorously, audit duplication, and write up a session.

6 skills · 5 agents

A fair, defensive guardian for a solo builder's usage meter, ported from CONCIERGE's bash token-scheduler to one tested Rust binary (tf). Estimates fan-out cost with a converging estimate→actual EWMA, requires an explicit +Xk budget, throttles waves, and gates each wave against the LIVE rate-limit window (fail-closed) — halting before a paid lockout and resuming cheaply from a per-job ledger that records only what's LEFT. Off-peak runs reserve a morning allowance. Durable, session-safe scheduled jobs (project + per-machine registry) re-arm after a crash and install as OS-cron. A Stop hook writes the session.json .tokens actual-spend signal that makes convergence work. Determinism is proven by cargo frozen-vectors + a byte-for-byte differential against the bash oracle. /schedule routes a job; the token-scheduler skill is the discipline that calls tf.

5 skills · 4 commands · 1 MCP server · hooks

BDD-enforced feature lifecycle: problem → Gherkin scenarios → design → plan → close, with mutation testing gate at close; /scaffold-bdd-docs copies the bare templates into a project

4 skills · 3 commands · 4 agents

An optional software factory over the sdlc change loop: a spec-writer, a builder, code, security and test reviewers, and an approver run a change's spec, plan, build and review, behind the same human gates. Install it only where you want agents to do that; sdlc works the same without it.

3 skills · 8 agents

Routes a task to the right execution shape — inline, a single subagent, or a Workflow fan-out — and picks the model tier and reasoning effort to match, for speed, result quality and cost together, including what the orchestrator's own context costs. Ships eight pre-tiered agent definitions, each pinning a model, an effort, a turn budget and a tool surface, because a subagent that names none inherits the session's.

2 skills · 8 agents · hooks

Software factory pipeline: groom, architect, implementer and validator as skills plus the autonomous agents that run them.

6 skills · 4 agents · hooks

9 cost-tiered subagents plus /aesir:playbook and /aesir:setup

2 skills · 9 agents

Evidence-gated agent workflow and cost-aware multi-agent orchestration. Routes work across cheap/current/strong model tiers, enforces exact file ownership, and requires independent review before integration.

6 skills · 2 agents · 1 MCP server · hooks

Frontier-model reasoning disciplines as skills for coding: decomposition, clarification, deep understanding, parallel subagents, verification, multipath exploration, and tool-grounded reasoning.

9 skills · hooks

Governs the scaffolding around a change: issue as spec of record, feature branch off the latest pre-release train, conventional commits, change request, and a bounded, infra-aware CI escort to green. Works across GitHub and self-hosted GitLab.

9 skills · 1 command

Runs a project through a decision register (ADR), an open-questions register (OQ), a phase register whose acceptance criteria are agreed before the work, and a verification register. Enforces the discipline behind them: refuse rather than guess, name the costs of every decision, and prove a gate can fail before trusting it.

10 skills

Spec-driven development workflow: subagents for review/planning/verification, a deterministic progress engine, templates, a SessionStart hook, and a PreToolUse drift guard. Stop vibe-coding.

1 skill · 5 commands · 3 agents · hooks

The Ralph sprint loop — plan, build, polish, review sprint stories autonomously.

7 skills · 3 agents

A spec-driven workflow: /milestone keeps local-only project direction and optionally syncs epic progress read-only from Beads; /specify authors stories/epics, /refine revises a story, /board shows the backlog, /solve implements one in an isolated worktree, /validate merges after human review, and /orchestrate runs ordered story/epic targets behind one GitHub PR or GitLab MR with explicit post-merge --finalize for complete epics. Requires bd for the story workflow and milestone sync.

7 skills · 2 agents · hooks

Project housekeeping for agentic engineering at five levels, Minimal to Immaculate, within a session, repo, or machine scope: quest and lore sync, safe git landing and pruning with containment proofs, workspace, runtime and Claude Code harness cleanup, and a verified disposition for every in-scope item, through a fingerprinted plan/apply engine with safety classes and undo.

8 skills · 1 command · hooks

Spec → Roadmap → Phase-Tasks → Task — a phased planning + execution workflow with explicit user-approval gates and JIRA-style task dashboards. Includes /bymax-workflow:brainstorm, /bymax-workflow:plan, /bymax-workflow:verify, /bymax-workflow:checkpoint, the universal /bymax-workflow:standards reference (with the §0 reuse-first simplicity ladder), and /bymax-workflow:autopilot — the loop-engineering executor that autonomously drives an approved roadmap to done, one merge-gated PR per phase.

2 skills · 8 commands

Én arbeidsflyt for en journalist, fra radarsignal til saksforslag. Selve arbeidet kalles SAKSLØPET, og inngangen for den som ikke har bestemt seg er `/kasus:start`: den legger fram grunnlaget valget tas på og spør hva journalisten vil gjøre, uten å bearbeide, kvittere, skrive eller researche noe. Den legger fram MATERIALET slik det ligger, i fire kall i samme melding: SISTE DØGN av radarsignalene (maks hundre), av Kasus' EGNE story-briefs (maks førti) og av redaksjonens egne artikler (maks seksti), pluss statusen «nytt siden sist» (`nytt`, det ene som kjenner kvitteringen) — fordi spørsmålet journalisten faktisk har om morgenen er «hva har vi å jobbe med i dag?», og de tre kildene peker mot ULIKE neste steg: en egen sak mot en oppfølger, et signal eller en brief mot en dybdeartikkel. Døgnet er et GULV og ikke et tak: vinduet er `max(24 timer, tiden siden forrige kvittering)`, med brukerens `--hours N` som overstyring for alle fire — og regelen står ETT sted, i `references/vindu.md`, fordi inngangen og saksløpet bruker samme vindu og at begge leser den derfra er en test. Gulvet finnes for å lukke en skjøt som ikke er til å se: `nytt` måler mot kvitteringen mens listene måler i timer, så et fast døgn ville betydd at fem døgn med «nytt siden sist» ble telt av `nytt` mens signalene manglet i lista de skulle MERKES i — og et signal som ikke kan merkes blir usynlig. Konsekvensen er at vinduet varierer med hvor lenge det er siden sist (daglig bruk gir et døgn, en ukes pause gir uka), så perioden SKAL stå i svaret med det faktiske tallet — en periode som ikke sies, leses som «alt» — og at takene binder oftere når vinduet vokser: er `taketNådd` sann, heves `--limit` framfor at vinduet krympes, for det er nettopp de gamle usette signalene gulvet skulle fange. Kvitteringen kjøres derfor FØRST, alene, fordi den er det eneste som kjenner tidspunktet; listene hentes i runde to. Er vinduet et døgn, er et tomt felt vanlig og betyr et fravær i PERIODEN, aldri i Kasus: er to av tre t

3 skills · 3 commands · 4 agents

Workflow skills for brainstorming a change, mapping a feature, cutting it into tickets, driving it to a PR, verifying what landed, and writing technical articles.

10 skills

Hook-less vendored subset of obra/superpowers (MIT, Jesse Vincent) at 8ca22db: brainstorming, writing-plans, subagent-driven-development, systematic-debugging, and the skills they reference. No SessionStart context injection.

10 skills