vbw

Plugin vbw from marketplace vbw-marketplace in swt-labs/vibe-better-with-claude-code-vbw. Commands read from the repo's manifest — today.

Skills (34)

• approve

  Approve the VBW contract (requirements, checks, plans) and the project commands it may run. Only the user can run this.

• compress

  Compress a long instruction file (CLAUDE.md, rules, docs for Claude) into terse caveman style to save tokens, keeping every fact, code block and path.

• config

  Show or change VBW settings - the model profile (quality, balanced, budget), per-role models, how much VBW does on its own, and the status line.

• convert

  Bring a VBW 1 project (.vbw-planning/) into VBW 2 through a short Q&A; the old folder stays unless you choose to delete it.

• debug

  Investigate a bug from three angles at once (reproduce, trace the code, recent changes) and get its root cause with evidence and a proposed fix.

• discuss

  Think something through with VBW before building (a decision, an idea, a concern): plain questions, trade-offs, a recommendation, and the outcome recorded.

• doctor

  Check that everything VBW needs is in place (tools, Claude Code version, workflows, status line, the project) and fix what is wrong.

• fix

  Make a quick fix or small change without planning ceremony; VBW's guards stay on and every proof is re-run afterwards so nothing proven breaks.

• help

  What VBW does and every VBW command, with when to use each.

• init

  Set up VBW in this git repository (.vbw/ with the spec and the plan of record) and the VBW status line, then start agreeing on what to build.

• interview

  Ask the user, once at the start of a project, how much software they have built, how VBW should explain things, how involved they want to be, and what they are building and for whom. Used by the vibe router when the kernel says the interview is due.

• list-todos

  List the ideas parked for later, with what you can do with each.

• loop

  Let VBW work through named backlog items on its own, within a spend ceiling. Only the user can run this.

• map

  Map an existing codebase (stack and commands, structure, conventions, tests, risks) so VBW plans fit the code; saves .vbw/map.md.

• panel

  Check whether the VBW panel and its sound work on this Claude Code.

• pause

  Stop for now safely - VBW checks nothing is left half-done and notes where you are, so /vbw:resume picks up exactly there.

• profile

  Switch how VBW works in one go - which AI models it uses, how much it does on its own (Careful, Standard, Fast, or your own mix), and how it talks to you (your level, how much it explains, how much you decide).

• qa

  Check the finished work now - run every proof and explain in plain words what passes, what fails and what only you can judge.

• report

  Prepare a bug report or idea about VBW as a GitHub issue; you see the full text and decide before anything is sent.

• research

  Research a question on the web, in documentation and in this project, and get a sourced answer with a recommendation (up to four Scouts in parallel).

• resume

  Pick up where you left off - where the project stands, what was decided, and the next step.

• rtk

  Set up, check or remove RTK, a third-party tool that compresses command output to save tokens; uses RTK's own official installer.

• skills

  Look for the best tools for this project (safety scanners, code-quality tools, test frameworks, community skills), show a short sourced list and install only what the user approves.

• status

  Where this VBW project stands - progress, the roadmap and the next step. Reads only.

• suggest

  Offer the user at most three suggestions at three moments: when requirements are proposed (spec proposal), right after triage adds requirements to the active milestone, and when a plan is presented for approval. Used by the vibe router at those moments only.

• teach

  Teach Claude and VBW's agents this project's conventions (naming, structure, testing, style); saved as Claude Code rules in .claude/rules/.

• todo

  Keep an idea for later in the VBW backlog, or list the backlog.

• triage

  Sort every new idea the user brings as now, next or later, with a reason and a size, before any requirement, planning or file change. Used by the vibe router for each new request.

• uninstall

  Remove VBW from Claude Code - the status line setting first, then the plugin; project files stay.

• update

  Update VBW to the latest version.

• verify

  Check, one at a time, the things only you can judge (look, feel, wording) for the work that is built, and record your verdict.

• vibe

  Take this VBW project to its next step (spec, plan, approve, build, prove, fix, accept, ship). Use it for any work on a VBW project.

• whats-new

  What changed in VBW, newest first.

• whats-next

  Have the Architect recommend what to do next (a top pick and up to two runners-up, each with a size) and store it. Used by the vibe router after a ship and when the user asks what's next.

Subagents (7)

• architect

  VBW Architect. Requirements to roadmap - the decisions the user must make, then phases with goal-backward success criteria, then what's next; planning only, never code.

• debugger

  VBW Debugger. Scientific method - reproduce, hypothesize, gather evidence, diagnose; then a minimal root-cause fix with a regression test, verified and documented.

• dev

  VBW Dev. Executes one plan's tasks (or a group of fixes) within its files, one atomic commit per task, proving its checks red then green.

• docs

  VBW Docs. Writes and updates documentation (README, changelog, guides, API and inline docs) for a plan's tasks; documentation files only.

• lead

  VBW Lead. Researches, decomposes each phase into small plans with tasks and tests that fail today, self-reviews, and writes them with vbw apply.

• qa

  VBW QA. Goal-backward verification of a built phase against its goal, criteria and plans; any deviation from the plan is a failure; changes nothing.

• scout

  VBW Scout. Researches one assigned angle - a codebase, the web, documentation - and reports verified findings with sources; changes nothing.

Hooks (8)

• SessionStart

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/session-start.sh"

• Stop

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/stop.sh" 2>/dev/null || true

• StopFailure

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/stop.sh" 2>/dev/null || true

• PreToolUse

  ulimit -t 5 2>/dev/null; s="${CLAUDE_PROJECT_DIR:-$PWD}"; p="$s"; while [ "$p" != / ] && [ -n "$p" ] && [ ! -e "$p/.git" ]; do case "$p" in */*) p="${p%/*}"; p="${p:-/}" ;; *) p=/ ;; esac; done; [ -e

• PreToolUse

  jq -c -f "${CLAUDE_PLUGIN_ROOT}/hooks/approve-ask.jq" 2>/dev/null | { IFS= read -r in && bash "${CLAUDE_PLUGIN_ROOT}/hooks/approve-ask.sh" "$in"; } 2>/dev/null || true

• PreToolUse

  ulimit -t 5 2>/dev/null; s="${CLAUDE_PROJECT_DIR:-$PWD}"; p="$s"; while [ "$p" != / ] && [ -n "$p" ] && [ ! -e "$p/.git" ]; do case "$p" in */*) p="${p%/*}"; p="${p:-/}" ;; *) p=/ ;; esac; done; [ -e

• PreToolUse

  ulimit -t 5 2>/dev/null; jq -c 'select((.agent_type // "") | test("^vbw:(architect|lead|dev|qa|scout|debugger|docs)$"))' 2>/dev/null | { IFS= read -r in || exit 0; s="${CLAUDE_PROJECT_DIR:-$PWD}"; p="

• PostToolUse

  out=$(jq -c -f "${CLAUDE_PLUGIN_ROOT}/hooks/approve-answer.jq" 2>/dev/null); [ -z "$out" ] || bash "${CLAUDE_PLUGIN_ROOT}/hooks/approve-answer.sh" "$out" 2>/dev/null || true

Similar plugins