Help us improve
Share bugs, ideas, or general feedback.
From sonzai-sdk
Use when given a meeting transcript or scoping doc and asked to ship a working Sonzai SDK integration end-to-end with zero human prompts. Reads transcript, derives tech stack autonomously (greenfield) or audits existing repo (brownfield), assembles a masterplan, dispatches a builder subagent that always-searches current package versions before writing any manifest, then generates Dockerfile + docker-compose, boots the app locally with `docker compose up -d --wait`, runs functional QA against the running app, and auto-fixes via a bounded fixer loop. Bounded, fully autonomous, no operator prompts.
npx claudepluginhub sonz-ai/sonzai-claude-skill --plugin sonzai-sdkHow this skill is triggered — by the user, by Claude, or both
Slash command
/sonzai-sdk:full-autoThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
**Closed-loop autonomous transcript → running-app pipeline.** Transcript in → running local docker-compose stack out. Zero prompts to the operator. Bounded fix loop.
answer-derivation.mdbrownfield-audit.mdbuilder-dispatch.mdfeedback-iteration.mdfinal-report.md.templatelocal-deploy.mdmasterplan-assembly.mdnotify-setup.mdpipeline.mdproject-type-detection.mdqa-loop.mdsubagent-prompts/auditor.mdsubagent-prompts/builder.mdsubagent-prompts/fixer.mdsubagent-prompts/notifier.mdsubagent-prompts/reply-poller.mdsubagent-prompts/version-checker.mdtech-stack-derivation.mdtemplates/docker-compose-postgres-redis.templatetemplates/docker-compose-postgres.templateProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Provides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Share bugs, ideas, or general feedback.
Closed-loop autonomous transcript → running-app pipeline. Transcript in → running local docker-compose stack out. Zero prompts to the operator. Bounded fix loop.
cto-loop (semi-auto with 2 gates) or sonzai-sdk wizard (spec + plan, no build)sonzai-sdk wizard/full-auto <path> with a transcript file pathversion-search.md..full-auto/BLOCKED.md and exit.git push, no gh pr create, no remote deploys (Fly / Cloud Run / etc.). Operator owns those decisions.SONZAI_API_KEY. If absent, abort with a clear message (autonomous pipeline cannot prompt for a key).https://api.sonz.ai/docs/openapi.json).→ Full detail: pipeline.md
| # | Phase | Reference | Output |
|---|---|---|---|
| 0-pre | Notify-setup | notify-setup.md | Detect Gmail+Slack MCPs, load recipient config |
| 0a | Transcript analysis | transcript-analysis.md | in-memory synthesis |
| 0b | Project type detection | project-type-detection.md | greenfield or brownfield |
| 0c-G | Tech-stack derivation (greenfield) | tech-stack-derivation.md | tech_stack |
| 0c-B | Brownfield audit | brownfield-audit.md + subagent-prompts/auditor.md | brownfield-context.md |
| 0d | Wizard answer derivation | answer-derivation.md | 8 sonzai wizard answers + rationales |
| 1 | Masterplan assembly | masterplan-assembly.md + templates/masterplan.md.template | docs/cto-review/<date>-masterplan.md |
| 2 | Build | builder-dispatch.md + subagent-prompts/{builder,version-checker}.md | local commits, files in CWD |
| 3 | Local deploy + QA | local-deploy.md + qa-loop.md + templates/ | running app + QA report |
| 4 | Feedback iteration (auto-fixer) | feedback-iteration.md + subagent-prompts/fixer.md | bounded retry loop |
| 5 | Final report | final-report.md.template | docs/cto-review/<date>-final-report.md |
sonzai-sdk (sibling) — archetype playbooks, feature references, spec templates, drift-detection guidecto-loop (sibling) — only if operator opted into the semi-auto path; otherwise unusedsonzai-internal-staff (sibling, install-time gated) — adds workspace awareness when present; full-auto runs fine without it on public sources alonemcp__chrome-devtools__* or mcp__playwright__*
status: blocked — do NOT re-dispatch; exit with the blockerSONZAI_API_KEY missing from environment AND from .env — halt with "set the key, then re-run"Operator gets:
docs/cto-review/<date>-masterplan.md (the autonomous masterplan; approval section emitted but not marked)docs/cto-review/<date>-final-report.md summarizing what was built, what was tested, what defaults were used, and what's left for the operatordocker compose stack on localhost — operator can docker compose down when ready.full-auto/BLOCKED.md if something couldn't be resolved within 5 cyclesOperator does NOT get prompted at all during the run.
Ships with the sonzai-sdk plugin in sonzai-claude-skill repo. See repo CHANGELOG for release history.