Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/agentlas-ai/agentlas-desktop/startup-studio-runtimenpx skills add agentlas-ai/agentlas-desktop --skill startup-studio-runtimegit clone --depth 1 https://github.com/agentlas-ai/agentlas-desktopWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/agentlas-ai/agentlas-desktop/startup-studio-runtime)<a href="https://agentmods.dev/skills/agentlas-ai/agentlas-desktop/startup-studio-runtime"><img src="https://agentmods.dev/badge/skills/agentlas-ai/agentlas-desktop/startup-studio-runtime.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00113 | $0.04932 |
| Opus 5 | $0.00056 | $0.02466 |
| Sonnet 5 | $0.00023 | $0.00986 |
| Haiku 4.5 | $0.00011 | $0.00493 |
Grade B, and why
startup-studio-runtime scanned grade B with 3 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Sends data to an external URLmediumData exfiltration
A POST to an outside endpoint may be telemetry or may be exfiltration; either way the mod talks to somewhere, and you should know where.
curl -s -XPOST http://127.0.0.1:4173/__studio/request -d '{"kind":"init","idea":"stub idea"}' Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
3. curl-verify the URL (see §4). macOS has no `timeout`. Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
- **FIX:** The slash captures the `route` JSON; if `action=="route"` + selected card has a local `gui_launcher` + `source` is a real dir → `subprocess.Popen([sys.executable, launcher], cwd=source, start_new_session=True, How it starts
The opening of the file, as written. The whole thing — 155 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Startup Studio Runtime — operator runbook
Deterministic resolution for the recurring GUI/runtime/publish saga. Cite the CODE, not the first guess. Line anchors below are verified against live source (open-studio-gui.py is 827 lines; forge-sync.mjs is 336 lines; caps live in agentlas.cjs). Re-verify any anchor before quoting it — files move.
1. When to use
Trigger on any of these symptoms (do not re-investigate from scratch — jump to §3 triage):
- "명령어 쳐도 GUI 안 뜸" / "Startup Studio GUI won't open"
- "클릭해도 / Run 눌러도 생성 안 됨, 행" / "clicking Run does nothing, hangs"
- "옛/레거시 GUI 뜸" / "old/legacy GUI shows"
- "내 api 쓰지마 / BYOK" concerns about external API keys
- "재게시 / publish 안 됨" / "re-publish the package fails"
2. Architecture invariants (memorize — these are fixed)
- BYOK = the user's local subscription CLI IS the runtime. Never an external/owner API key.
claude -p/codex execrunning in the session does the LLM work. An owner key baked into a public deploy = bankruptcy risk; the OpenAI-keyed Node server was a deleted mistake. No external API, no per-call cost, no Dockerfile/railway deploy (a public URL has no session). - The launcher serves SPA + a dev bridge — no LLM in the bridge.
scripts/open-studio-gui.pyservesweb/dist(SPA) and three bridge endpoints:GET /studio-data.json(live working copy or{},no-store;:442-450),GET /__studio/manifest({data:<mtime_ns>, idea, busy};:451-460/ embedded:540-549),POST /__studio/request(appends one line to.studio-runtime/requests.jsonl;:468-484). The bridge handler has ZERO LLM calls — see its docstring (open-studio-gui.py:412-419). - No embedded GUI fallback.
web/distis still excluded from the published bundle becausedistis a hardcoded packager skip-dir, but.htmlis included soweb/index.htmlships. A fresh Hub install must build the real React SPA on first launch (cd web && npm install && npm run build) and then serveweb/dist. If build prerequisites are missing, the launcher must fail loudly withgui_unavailable; it must NOT fall back toEMBEDDED_GUI_HTMLor any static-lite UI. - StudioRunner auto-spawns the local CLI per queued request and atomically commits.
class StudioRunner(open-studio-gui.py:222); daemon thread started at:264; pollsrequests.jsonlevery 0.8s (:276); cursor starts at current line count (:236) → skips stale pre-launch requests; serialized one-at-a-time.init= idea-stage-only;run= exactly one stage. CLI writes.studio-runtime/studio-data.next.json; Python validates thenos.replace→ GUI never sees a half-written file.StudioRunner(...).start()runs in BOTHserve_dirandserve_embedded. The runner spawns the CLI that MATCHES the host that launched the launcher (_runner_cmd): Codex host (CODEX_THREAD_ID/CODEX_SANDBOX/CODEX_CI) →codex; Claude host (CLAUDECODE=1/CLAUDE_CODE_ENTRYPOINT) →claude;STUDIO_RUNNER_CLIoverrides. It does NOT just pickclaudefirst. - The SPA derives all stage lock/done from verdict labels. No backend.
stageDone[s]=content.stages[s].verdict.labelis NOT in the NOT-DONE set{대기, 생성 중, 생성중, awaiting, 잠김, locked, "", —, -}(studio-context.tsx:78; derived:151-155).stageLocked[s]= predecessor not done (:158-165);ideais never locked. A placeholder verdict → stage (and successors) stay LOCKED.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 155 lines · 113 tokens per session scan B 89feadc0fed0
startup-studio-runtime is a skill published in the GitHub repository agentlas-ai/agentlas-desktop (5 stars, last pushed yesterday), licensed Apache-2.0. It adds 113 tokens to every session and 4,932 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it B with 3 findings (sends data to an external url, makes network calls, runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
create-workspace
Creates a .clave workspace file that defines groups, sessions, terminals, toolbar actions, and icons for the Clave desktop app. Use when the user wants to create, edit, or understand .clave workspace files.
electron-expert
Expert in Electron framework, desktop app development, IPC, and cross-platform packaging. Use when the user mentions desktop, Node.js, cross platform, Windows, macOS, or Linux, or when the task involves Electron Architecture, Process Types, IPC Communication, or App Lifecycle.
recover-sessions
Rebuilds a lost Clave workspace by finding the user's past Claude Code sessions on disk, mapping them to projects, and reopening them as grouped tabs with their full history. Use when the user has lost their tabs or groups (crash, restart, accidental close) and wants their working sessions back, or when they ask what…
ensemblr
How Ensemblr works from inside it — the workspace/git-worktree model, the ensemblr control tools and what their answers mean, the committed .ensemblr/settings.toml (setup and run scripts, [git], [prompts], [infisical], environment variables), /.config/ensemblr/config.json, the Changes panel and diff comments, the…
skills-cli-sync
Sync skills-desktop's AGENTDEFINITIONS and UNIVERSALAGENTIDS with the upstream vercel-labs/skills CLI, then update docs and validate. Use when bumping SKILLSCLIVERSION, mirroring a new upstream skills-CLI release, or when the agent list / universal set drifts from upstream agents.ts.
react-doctor
Use when finishing a feature, fixing a bug, before committing React code, or when the user types /doctor, asks to scan, triage, or clean up React diagnostics. Covers lint, accessibility, bundle size, architecture. Includes a regression check and a full local-triage workflow that fetches the canonical playbook.