harness-native operator system for design, product and brand work.

agents, skills, hooks, rules and commands for ai coding agents like claude code, cursor, codex, opencode, gemini cli, zed, and vscode + copilot. one source of truth in markdown, six adapters for the rest. MIT licensed, free for everyone, designed to be edited freely.

studio/
├── agents/           15 specialist agents
├── skills/           33 skills across design, product, brand
├── commands/         29 slash commands
├── rules/            23 rules across 5 lanes
├── hooks/            working claude code hooks (pre-write, post-write, prompt-context)
├── mcp-configs/      figma, notion, linear, posthog, filesystem
├── memory/           operator instincts, lessons, decisions, glossary
├── adapters/         cursor, codex, opencode, gemini, zed, vscode
├── scripts/
│   ├── hooks/        hook runtime (node, executable)
│   ├── dashboard/    build.js (static html), log-viewer.js (cli)
│   └── util/         check-references.js (drift guard)
├── docs/             generated static dashboard (github pages ready)
└── tests/            run-all.js validator

git clone https://github.com/Dragoon0x/studio.git ~/.claude/studio
cd ~/.claude/studio
./install.sh

the installer links the plugin, writes the hooks config into ~/.claude/settings.json with absolute paths, and copies the mcp templates.

design-reviewer, design-system-auditor, accessibility-reviewer, brand-voice-keeper, copywriter, microcopy-writer, product-strategist, ux-research-synthesizer, competitor-analyst, naming-generator, narrative-architect, taxonomy-architect, release-narrator, case-study-writer, pitch-deck-writer.

design (10): design-review, design-system-audit, accessibility-audit, figma-handoff-spec, component-spec, motion-direction, responsive-rules, dark-mode-pairing, iconography-system, data-viz-design.

product (10): prd-writing, spec-writing, research-synthesis, jtbd-framing, roadmap-planning, feature-scoping, metric-design, ab-test-design, competitive-analysis, launch-planning.

brand (13): brand-voice-extraction, naming-generation, tagline-writing, positioning-statement, messaging-architecture, value-prop-writing, microcopy-writing, landing-copy, case-study-writing, release-narrative, brand-identity-audit, content-calendar, email-sequence.

design: /design-review, /a11y-scan, /system-audit, /handoff product: /prd, /spec, /research-synth, /jtbd, /roadmap, /scope, /metrics, /experiment, /competitor, /launch brand: /brand-check, /name, /copy-review, /voice-extract, /tagline, /position, /messaging, /value-prop, /microcopy, /landing, /case-study, /release, /brand-identity, /content-calendar, /email-sequence

pre-write flags ai-tone and rhythm flatness before writes. post-write logs every write. prompt-context surfaces relevant skills based on prompt keywords and logs each activation. all non-blocking by default. set STUDIO_HOOK_STRICT=1 to block on flags.

• common (5): principles, process, handoff, research, decisions
• design (5): spacing, type, color, motion, accessibility
• product (5): prd-structure, jtbd, metrics, specs, research-standards
• brand (5): voice, tone-matrix, banned-words, naming-conventions, messaging-hierarchy
• copy (3): sentence-rhythm, anti-ai-tone, active-voice

a static HTML site browsing every agent, skill, command and rule. searchable, filterable by type and lane. click a card to open the full source.

npm run dashboard        # writes to docs/
node scripts/dashboard/build.js --out custom-path
node scripts/dashboard/build.js --watch

deploy: push to github, enable github pages serving from docs/ on main.

inspect what STUDIO surfaced and what got written.

npm run logs                                    # today's activations
node scripts/dashboard/log-viewer.js --top      # most-activated keywords
node scripts/dashboard/log-viewer.js --writes   # write log
node scripts/dashboard/log-viewer.js --last 7   # last 7 days

scans for broken cross-references. runs as part of the test suite, also standalone.

npm run check-refs
node scripts/util/check-references.js --strict   # exit 1 on findings

cursor (.mdc rules), codex (AGENTS.md), opencode (commands + agents + opencode.json), gemini cli (GEMINI.md + @-imports), zed (agent profile), vscode + github copilot (.github/copilot-instructions.md).

each adapter has its own install.sh. run from STUDIO root:

./adapters/cursor/install.sh           # user-scope
./adapters/cursor/install.sh --project # project-scope

memory/instincts.md, lessons.md, glossary.md, decisions/, templates/. edit these as you learn what works for your team. instincts get auto-loaded by the prompt-context hook once edited.

figma (remote + dev mode stdio), notion, linear (remote), posthog, filesystem. see mcp-configs/README.md for env vars and setup.

./install.sh                        # default: claude code, all rules
./install.sh --rules common,design  # only common and design rules
./install.sh --target claude        # claude code (default)
./install.sh --no-hooks             # skip hooks setup
./install.sh --with-adapters cursor # also install cursor adapter

windows:

.\install.ps1 -Rules "common,design"

npm test

validates: JSON parsing, agent/skill/command frontmatter, hook execution on 8 edge cases, adapter install scripts, cross-references, dashboard build, log-viewer cli, and unit tests on the banned-word and rhythm checkers. 339 checks.

MIT. see LICENSE.

see DISCLAIMER.md. opinions in agents and skills are starting points; override them when context demands it. content generated through STUDIO is the responsibility of the operator running the harness.

Dragoon0x — 0xdragoon.xyz. issues and PRs welcome at github.com/Dragoon0x/studio.

• Dragoon0x — creator, maintainer.

full contributor list lives on the GitHub contributors page.

STUDIO is early-stage, experimental software. session 3 ships a dashboard, log viewer, reference checker, three new skills/commands, and a docs page; sessions 4–5 are unfinished. interfaces, agent contracts, skill descriptions, rule formats, hook signatures, adapter layouts, dashboard scripts, and install scripts may change without notice and without migration paths.

do your own research (DYOR) before relying on STUDIO for anything that matters:

• read every agent, skill, rule, hook, and dashboard script before running them. they encode opinions and they execute on your machine.
• review every output before shipping it. agents and skills will be wrong sometimes.
• validate generated copy, prds, audits, scopes, brand identities, content calendars, email sequences, and recommendations against your own context, audience, and legal/compliance requirements.
• treat anything STUDIO produces as a draft, not a deliverable.
• audit the hook scripts under scripts/hooks/, dashboard scripts under scripts/dashboard/, and each adapter's install.sh before running them — they execute locally, read prompts, write logs, and modify your harness config.

no warranty. STUDIO is provided "as is" under the MIT license. there is no guarantee of accuracy, originality, fitness for any purpose, security, or availability. the author and contributors accept no liability for losses, damages, missed deadlines, brand harm, leaked information, or any other consequence arising from use of this repo.

not professional advice. nothing in STUDIO constitutes legal, financial, medical, security, accessibility-compliance, or other professional advice. accessibility audits, brand guidance, product specs, content calendars, email sequences, and copy reviews here are starting points — not substitutes for qualified review.

operator owns the output. content generated through STUDIO via any harness (claude code, cursor, codex, opencode, gemini, zed, vscode, etc.) is the responsibility of the operator running the harness, not the author of STUDIO.

no affiliation. STUDIO is not affiliated with, endorsed by, or sponsored by Anthropic, Cursor, OpenAI, Google, Zed Industries, GitHub/Microsoft, Figma, Notion, Linear, PostHog, or any other company referenced in agents, skills, adapters, examples, or docs. all trademarks belong to their respective owners.

if any of this is a problem for your use case, do not install STUDIO. fork it, audit it, or wait for a stable release.

see DISCLAIMER.md for the long form.