Standby.Design
CLAUDE.md1 starsChanged 15 days agoMITView on GitHub
Operational knowledge for agents working in this repo. Architecture lives in CONTEXT.md, product strategy in STRATEGY.md — read those for the what; this file is the how-to-not-get-stuck. - The npm account uses passkey 2FA — there are no TOTP codes. npm's browser auth flow requires a real TTY; from an agent session shell, npm publish fails (EOTP) and masks the auth URL. Never try to publish from the session shell, and don't send the user to a terminal. Instead, open…
In this repository2 other files it gives its agents.
- public/llms.txtllms.txt
- design-reviewSkill
The file
We split this CLAUDE.md into the rules it gives an agent
3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 66
Every export generator lives once in packages/core/src/*-code-export.ts.
- Very specific: line 88
Run npm install in packages/design-lint first.
- Very specific: line 91
Verify MCP server changes with npm run smoke (end-to-end stdio test).
≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CLAUDE.md
Operational knowledge for agents working in this repo. Architecture lives in CONTEXT.md, product strategy in STRATEGY.md — read those for the what; this file is the how-to-not-get-stuck.
npm publish (standby-design-mcp, standby-design-lint)
The npm account uses passkey 2FA — there are no TOTP codes. npm's browser auth flow requires a real TTY; from an agent session shell,
npm publishfails (EOTP) and masks the auth URL. Never try to publish from the session shell, and don't send the user to a terminal. Instead, open a visible window on the user's desktop:Start-Process powershell -ArgumentList '-NoExit','-Command','Set-Location "C:\Users\karst\repositories\Solopreneur\standby.design\packages\mcp-server"; npm publish'The browser opens, the user confirms the passkey. If they check "do not challenge for 5 minutes", follow-up publishes from the session shell work without a prompt — useful for quick re-publishes.
- Version bump touches three places:
package.json(version),src/server.ts(SERVER_VERSION),server.json(versionandpackages[0].version). mcpNamein package.json must stayio.github.KarstenKreh/standby-design-mcp— case-sensitive and must exactly match thenamein server.json, or the MCP registry rejects the publish.- MCP registry update after an npm release:
mcp-publisher login github(device flow, user confirms code) thenmcp-publisher publish. Binary: GitHub releases ofmodelcontextprotocol/registry(mcp-publisher_windows_amd64.tar.gz). Registry description max 100 chars. standby-design-lintis published the same way, withSet-Locationon...\packages\design-lint(first publish:npm publish --access public). Its version lives only inpackages/design-lint/package.json; the plugin meta reads it from there. Publish it before deploying the site or the MCP server, both point to it.
Deploys (website + remote MCP server)
bash deploy.shdeploys everything (site +standby-mcpcontainer for mcp.standby.design). SSH auth runs through the 1Password SSH agent — there is no key file on disk. The user may need to confirm a 1Password popup; if auth fails, checkssh-add -Land the whitelist in%LOCALAPPDATA%\1Password\config\ssh\agent.toml.- The script must use Windows OpenSSH (it prepends the PATH itself) — Git Bash's bundled ssh cannot reach the 1Password agent pipe.
- Never pipe
deploy.shthroughtail/headin a background task. Those buffer until EOF, so the log stays empty for the whole run and the deploy looks hung when it is only building images on the server (several minutes, normal). Run it bare and read the output file as it grows. Before blaming SSH, verify withssh -v -o BatchMode=yes -o ConnectTimeout=10 root@46.225.131.97 'echo ok'— that fails fast instead of waiting on a prompt. - New root-level static files (
public/*.html,*.txt) also need aCOPYline in the rootDockerfile, or they 404 in production while working locally.
Pitfalls
- The design rules at
/docs/rulesand the research pages at/docs/researchare generated. Editdocs/rules/*.mdordocs/research/*.md, then runnpm run buildindocs/and commit the output inpublic/docs/(seedocs/README.md). - Every export generator lives once in
packages/core/src/*-code-export.ts. The System page and the MCP server compose them throughpackages/core/src/system-export.tsand resolve a hash withresolveDesignSystem(design-system.ts). Defaults per tool sit next to the codec inpackages/core/src/url-state/*.ts. Rebuild the MCP server (packages/mcp-server: npm run build) after generator changes. - Each tool app syncs its store with the URL hash in
src/lib/hash-sync.ts(startHashSyncfrom@core/hash-sync), before the first render. color-reacthas snapshot tests covering export output: review the diff first, then update withnpx vitest run -u.- The apps build with
vite buildonly (no type checking). Runnpx tsc --noEmitwhere needed;packages/mcp-serverhas a propertypecheckscript. - On Windows the docs build appends a second copy of every entry to
public/sitemap.xmlwhen the file has CRLF line endings. Convert it to LF beforenpm run build(sed -i 's/\r$//' public/sitemap.xml) and check that the diff only adds the new pages. packages/design-lintenforces the design rules: a Claude Code hook (.claude/settings.json) lints every edited line,.githooks/pre-committhe staged lines and only warns (activate once withgit config core.hooksPath .githooks), CI the lines of a PR and fails. Other projects get all of it withnpx standby-design-lint@latest init. Runnpm installinpackages/design-lintfirst. Existing violations in the apps are known; they only show up when a line is touched.- Verify MCP server changes with
npm run smoke(end-to-end stdio test); the remote variant serves/healthfor a quick liveness check.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

