409 current commands, shortcuts and tips across 11 cheat sheets
“I’m gonna be King of the Pirates!” — Monkey D. Luffy
BMad skills are invoked as slash commands in Claude Code (e.g. /bmad-prd) or by naming them in chat. The [CODE] prefix is the agent menu code shown by bmad-help. Run each workflow in a fresh context window (/clear). Version 6.12.4 (core and bmm 6.12.4 · bmb v2.2.2 · cis v0.3.2 · tea v1.19.0 · wds v0.4.3) — sections tagged {v6.12.4} apply to it; sections tagged {v6} are the previous generation, hidden until you tick "Show v6 (legacy) tips". Most v6 skill names still work as forwarding shims (removal rides the v7 cut).
npx bmad-method installuv — every workflow skill starts with a uv run render step and halts if uv is missingbmad-bmb-setupbmad-loop-setupbmad-customize_bmad/custom/bmad-helpbmad-party-modebmad-sprint-planning statusbmad-sprint-statusbmad-agent-analystbmad-agent-pmbmad-agent-architectbmad-agent-ux-designerbmad-agent-devbmad-teabmad-cis-agent-brainstorming-coachbmad-cis-agent-creative-problem-solverbmad-cis-agent-design-thinking-coachbmad-cis-agent-innovation-strategistbmad-cis-agent-presentation-masterbmad-cis-agent-storytellerwds-agent-saga-analystwds-agent-freya-uxbmad-brainstormingbmad-deep-recon*-research skillsbmad-product-briefbmad-prfaqbmad-forge-ideabmad-spec / bmad-buildbmad-prdbmad-create-prd / edit-prd / validate-prdbmad-uxbmad-specbmad-project-contextbmad-generate-project-context and bmad-document-projectbmad-architecturebmad-create-epics-and-storiesbmad-sprint-planningsprint-status.yaml. Intents: readiness, status, validate, fix. Replaces bmad-check-implementation-readinessbmad-buildbmad-quick-dev and, in practice, create-story + dev-storybmad-code-reviewbmad-walkthroughbmad-checkpoint-previewbmad-qa-generate-e2e-testsbmad-retrospectivebmad-story-automatorbmad-story-automator-reviewbmad-build-autobmad-loop-resolve <story-key>bmad-loop-sweepbmad-correct-coursebmad-reviewadversarial, edge-case-hunter, verification-gap (code), structure, prose (docs). Ask for lenses by namebmad-advanced-elicitationbmad-teach-me-testingbmad-testarch-test-designbmad-testarch-frameworkbmad-testarch-cibmad-testarch-atddbmad-testarch-automatebmad-testarch-test-reviewbmad-testarch-nfrbmad-testarch-tracebmad-cis-innovation-strategybmad-cis-problem-solvingbmad-cis-design-thinkingbmad-cis-storytellingbmad-agent-builderbmad-workflow-builderbmad-module-builderbmad-eval-runnerwds-0-project-setupwds-0-alignment-signoffwds-1-project-briefwds-2-trigger-mappingwds-3-scenarioswds-4-ux-designwds-5-agentic-developmentwds-6-asset-generationwds-7-design-systemwds-8-product-evolutionbmad-helpbmad-party-modebmad-sprint-statusbmad-agent-analystbmad-agent-pmbmad-agent-architectbmad-agent-ux-designerbmad-agent-devbmad-agent-tech-writerbmad-teabmad-brainstormingbmad-market-researchbmad-domain-researchbmad-technical-researchbmad-product-briefbmad-prfaqbmad-forge-ideabmad-prdbmad-uxbmad-specbmad-architecturebmad-create-epics-and-storiesbmad-check-implementation-readinessbmad-sprint-planningbmad-create-storyvalidate → [VS] Validate Story)bmad-dev-storybmad-code-reviewbmad-checkpoint-previewbmad-qa-generate-e2e-testsbmad-retrospectivebmad-story-automatorbmad-quick-devbmad-correct-coursebmad-investigatebmad-document-projectbmad-generate-project-contextbmad-index-docsbmad-shard-docbmad-review-adversarial-generalbmad-review-edge-case-hunterbmad-editorial-review-prosebmad-editorial-review-structurebmad-advanced-elicitationBuilt-in Claude Code slash commands (typed inside a session) and CLI commands/flags (run from the shell). Grouped by category.
/usage/cost/usage-credits/context/cost/context/compact <focus>/compact keep the API decisions)/clear/model haiku / /model sonnet/effort low/mcp/statuslineclaude --max-budget-usd <amount>claude -p "query" --output-format jsontotal_cost_usd and token usage — log it in CInpx ccusage@latest/config/model/effort/agents/hooks/statusline/terminal-setup/vim/keybindings/memory/compact/clear/resume/cd/add-dir/exit (alias /quit)/rewind/export/status/artifacts@path/to/file!command/skill-name args/bmad-help what next?)/permissions/login/logout/code-review (alias /review)/security-review/verify/code-review ultra/code-review ultra <PR#>)/simplify/mcp/ide/install-github-app/plugin/loop/goal/fast/schedule/fewer-permission-prompts/init/help/doctor/release-notesclaudeclaude "query"claude -p "query"cat file | claude -p "query"claude -p "query" --output-format jsonstream-json)claude -c / --continueclaude -r <session> / --resumeclaude -r <session> --fork-sessionclaude --model <model>claude --effort <level>claude --permission-mode <mode>claude --add-dir <path>claude --bg "prompt" / --backgroundclaude --max-budget-usd <amount>claude updateclaude doctorclaude mcp addclaude mcp listclaude mcp remove <name>claude plugindevflow statusdevflow statusdevflow status ITEMdevflow status 8-3devflow checklist ITEM --to ÉTATdevflow checklist 8-3 --to IN_PROGRESSdevflow advance ITEM --to ÉTAT --checklist-ok --approve "QUI (RAISON)"devflow advance 8-3 --to IN_PROGRESS --checklist-ok --approve "JB (dev terminé)"devflow advance ITEM --to ÉTAT --mockup-version Ndevflow advance 8-1 --to READY --mockup-version 2devflow sync-plan ITEMdevflow sync-plan 8-3devflow sync-match ITEMdevflow sync-plan 8-3 | devflow sync-match 8-3devflow sync-transition "<STATUS>"jira-api get-transitions LUFFY-42 | devflow sync-transition "In Progress"devflow sync-record ITEM --key <CLÉ> --status "<STATUT>"devflow sync-record 8-3 --key LUFFY-42 --status "In Progress"Typed in Copilot Chat (VS Code) unless noted otherwise.
/explain/fix/tests/doc/new/newNotebook/setupTests/fixTestFailure/startDebugging/search/clear/help@workspace@vscode@terminal@github#file#selection#codebase#changes#problems#terminalLastCommand#terminalSelection#fetch.github/copilot-instructions.md.github/instructions/*.instructions.mdapplyTo glob in frontmatter).github/prompts/*.prompt.md/prompt-name in chatAGENTS.mdnpm install -g @github/copilotcopilotcopilot -p "query"copilot --continuecopilot --resume/resume (CLI)/usage (CLI)/session (CLI)/context (CLI)/compact (CLI)/model (CLI)/clearmicrosoft/hve-core (Hypervelocity Engineering) — the RPI loop: Research → Plan → Implement → Review. In Claude Code the phases are skills (/rpi-research …, installed with hve.ps1). In GitHub Copilot (VS Code extension *HVE Core*) they are prompts bound to agents (/task-research …). Every phase writes its artifact under .copilot-tracking/ — that folder, not the chat, is the memory.
./hve.ps1 list./hve.ps1 add rpi -Target '<project>'rpi-* skill into the project's .claude/ (quote paths with spaces)./hve.ps1 add all -Target '<project>'./hve.ps1 add rpi -Ref v1.0.0 -Target '<project>'./hve.ps1 update -Target '<project>'./hve.ps1 status./hve.ps1 remove rpi -Target '<project>'.claude/hve-manifest.json, your own skills are untouched)/rpi task=.../rpi continue.copilot-tracking/ artifacts/rpi followUp=.../rpi-research topic=... posture=focused.copilot-tracking/research/<date>/<slug>-research.md (posture balanced | focused | expansive)/rpi-plan research=<file>.copilot-tracking/plans/<date>/<slug>-plan.md, standard critique included/rpi-plan critique=deep / critique=skip/rpi-plan-critique plan=<file>PC-xxx findings, Pass / Revise / Blocked)/rpi-implement plan=<file> phase=P02Pxx phase / Pxx-Txx task), checks boxes, keeps .copilot-tracking/changes//rpi-reviewRV-xxx findings + routing → .copilot-tracking/reviews/logs//rpi-review plan=<file> changes=<file>/rpi-review depth=deepstandard)/rpi-challenger subject=... focus=.../rpi-walkthrough target=... detail=brief.copilot-tracking artifact (brief | normal | deep)/rpi task=.../rpi continue=11,2 or all also work)/rpi suggest/task-research topic=....copilot-tracking/research//task-plan research=<file>/task-implement plan=<file> phaseStop=truephaseStop / stepStop pause for your review/task-review scope=todayplan= changes= research=, scope this week / since last review/checkpoint mode=save description=.../checkpoint mode=continue description=<search>/checkpoint mode=incremental/pull-request branch=origin/main.copilot-tracking/pr/pr.md/git-commit / /git-commit-message/compact that keeps every .copilot-tracking path and the current phaseresearch/<date>/<slug>-research.mdplans/<date>/<slug>-plan.mdPxx phases and Pxx-Txx taskschanges/<date>/<slug>-changes.mdreviews/logs/<date>/<slug>-review.md-review-2.md for the next review of the same task)challenges/ / walkthroughs/code)12code .code <file>code -r <path>code -ncode -g <file>:<line>code -d <file1> <file2>code <name>.code-workspacecode --list-extensionscode --install-extension <id>anthropic.claude-code)code --uninstall-extension <id>code --disable-extensionscode --profile "<name>"Developer: Reload WindowPreferences: Open User Settings (JSON)Preferences: Open Keyboard Shortcuts (JSON)Workspaces: Add Folder to Workspace...Files: Compare Active File With...Markdown: Open Preview to the SideTerminal: Select Default ProfileFormat Document With...Change Language Mode>@#:?/terminal-setup in the terminal)VS Code, Windows default keybindings.
Windows default keybindings. Ctrl+K Ctrl+S opens the full list.
Keeping AI usage visible and cheap with Claude Code and GitHub Copilot. The matching commands are in the Claude and GithubCopilot command sheets, under Tokens & Cost.
/cost (Claude Code) or /usage (Copilot CLI) and paste the numbers into the PR description, the BMad spec or story file (*Implementation Notes* in 6.12.4, *Dev Agent Record* in v6), the commit body or the report. This shows the real cost of each feature and makes estimates better over timeAI usage: Claude Code · Opus 5.5 · effort high · 1.2M in / 85k out tokens · $4.20 · 2 sessions. Use the same format everywhere so it's easy to grep and add up/clear between tasks so the cost you read belongs to exactly one deliverable. BMad already recommends a fresh context per workflowclaude -p … --output-format json returns total_cost_usd and usage. Store them next to the build artifacts/usage counter can undercount after compaction. Reconcile monthly with GitHub billing, and with your Anthropic console or /usage for Claude/clear between unrelated tasks/compact <focus> before the context fills/context@file / #file/effort low when no deep reasoning is neededCLAUDE.md / copilot-instructions.md shortclaude --max-budget-usd 5 stops a runaway agent loop before it gets expensiveThe commands that give the most for the least: BMad build cycle (6.12.4; the previous v6 story cycle is kept under {v6}, hidden by default), the BMad loop for unattended runs, HVE rpi-review, and how to pick up an existing session in Claude Code or GitHub Copilot. Full syntax is in the Bmad, Hve, Claude and GithubCopilot command sheets.
/bmad-help to set the context/bmad-help what is the next story of epic 2 and what context does it need?uv firstuv run render step and HALTS if uv is missing. npx bmad-method install needs it too/bmad-project-context (setup / adopt / refresh / record / audit) maintains a verified block in AGENTS.md. It replaces generate-project-context; use record when an agent repeats a mistake so it becomes a pitfall/bmad-sprint-planning before the first buildsprint-status.yaml. Run it again with intent status any time: /bmad-sprint-planning show sprint status/bmad-build in a fresh context/clear, then /bmad-build with a story, an issue link, a spec file or just the intent. One flow: clarify → plan → implement → review → present. The spec file is the hand-off, not the chat/bmad-build: it lists active specs (draft, ready-for-dev, in-progress, in-review) and lets you resume one instead of starting over/bmad-spec to break a spec into stories/bmad-code-review, or /bmad-review lenses=edge-case-hunter,verification-gap on a diff or a document/bmad-walkthrough explains what the change is for, what to inspect closely and how to test it before you merge/bmad-story-automator chains create → dev → QA → review → retro across an epic; bmad-loop runs unattended and pauses on escalations for /bmad-loop-resolve <story-key>bmad-quick-dev, bmad-create-prd, bmad-sprint-status, bmad-checkpoint-preview, bmad-market-research… forward to the new skills (shims, removed with v7 only). Prefer the new names/cost before closing, then the AI usage: line in the spec's *Implementation Notes* (bmad-build) or the story's *Dev Agent Record* (v6), one line per session, a total at the endbmad-loop6.12.410/bmad-loop-setup installs the tool from Git (uv tool install "bmad-loop[tui] @ git+https://github.com/bmad-code-org/bmad-loop.git"), then bmad-loop init --project "<project root>" --cli claude registers the hooks and writes .bmad-loop/policy.tomlbmad-loop validate --project "<project root>" preflights the install and the per-role adapter config. Fix what it reports before you start a runclaude once in the project and accept the workspace-trust and hooks-approval prompts. Spawned sessions cannot answer them, and the run stalls thereinit --cli claude --cli codex sets up more than one profile. Valid names: claude, codex, gemini, copilot, antigravitybmad-loop tui opens the Textual dashboard (installed with the [tui] extra) to follow the run, the story and the sessions/bmad-loop-resolve <story-key>. The human makes the call; the spec is then updated and the story re-drivenbmad-loop resume <run>. Do not edit the spec for itbmad-loop sweep verifies each open entry in the deferred-work ledger against the code and sorts it into bundles, already resolved, blocked, skip or human decision. It is read-only/cost on the sessions you care about and log one AI usage: line per session in the spec, as with /bmad-build/bmad-help to set the context/bmad-help what is the next story of epic 2 and what context does it need?project-context.md once/bmad-generate-project-context writes the project's AI rules. Story creation and dev load it, so every story inherits the same conventions without you repeating them/bmad-create-story in a fresh context/clear, then /bmad-create-story 2.3. The story file gets everything the dev needs (ACs, tasks, architecture notes, files to touch) — it is the hand-off, not the chatvalidate action ([VS]) to catch missing context while it is still cheap to fix/bmad-dev-story <story file> in another fresh context/clear between create and dev. Dev works task by task and checks them off in the file, so rerunning on the same story resumes at the first unchecked task/bmad-code-review with fresh eyes. Issues → back to /bmad-dev-story, approved → next /bmad-create-story/bmad-story-automator chains create → dev → QA → review → retro across an epic/cost before closing, then the AI usage: line in the *Dev Agent Record* (one line per session, a total at the end).copilot-tracking/. /clear between phases: the next phase reads the file, not the chat, so you pay for less context/rpi-reviewRV-xxx findings with a route: defect → /rpi-implement, missing decision → /rpi-plan, missing evidence → /rpi-research, extra work → follow-up/rpi-review explicit paths/rpi-review plan=<plan.md> changes=<changes.md> — it stops if several unrelated tasks matchdepth=deep only on purpose/rpi-review depth=deep for critical changes onlystarted) continues from its record. A finished review is reused unless you ask for a new one (-review-2.md)/rpi-plan runs a standard critique by default; /rpi-plan-critique plan=<file> for an independent second look. Fixing a plan is cheaper than fixing code/rpi-challenger subject=... asks one skeptical question per turn, no answers given — good before an architecture choice/rpi-walkthrough target=<files> detail=brief explains segment by segment with file links/task-research, /task-plan, /task-implement phaseStop=true, /task-review scope=today, or /rpi task=... for the autonomous RPI Agent/bmad-code-review. RPI-tracked task (plan + changes in .copilot-tracking/) → /rpi-review. Each review expects its own artifactsclaude -cclaude -rclaude -r <id or name> jumps straight to one/resumeclaude -r <id> --fork-sessioncd into the project first, or -c finds nothing/clear session + the artifact (spec / story file, plan, /rpi continue) is often cheaper than resuming a long conversationcopilot --continue reopens the last session, copilot --resume lets you pick one, /resume from inside the CLI/checkpoint (HVE)/checkpoint mode=save description=story-2-3 before leaving, then /checkpoint mode=continue description=story-2-3 in a new chat to restore the context/rpi continue=1 (HVE).copilot-tracking path and the current phase — use it instead of a plain /compactsprint-status.yaml and the spec / story file: in a new chat, /bmad-help or /bmad-sprint-planning show sprint status tells you where you left offsprint-status.yaml and the story file: in a new chat, /bmad-help or /bmad-sprint-status tells you where you left off