File contents Skill: obscura-browser
Motor headless nativo de Savia: Rust + V8, motor de render propio, CDP drop-in
para Playwright/Puppeteer, MCP nativo, anti-detect opcional, guard SSRF,
cero telemetria. Adoptado 2026-09-04 tras benchmark vs Playwright (3x menos
RAM, arranque frio 76ms). Depreca a lightpanda-browser.
Authoritative Paths
Lee estos paths antes de actuar. NUNCA asumas firmas, NUNCA inventes paths.
Para
Lee este path
Binario instalado
~/.local/bin/obscura (v0.2.1) + ~/.local/bin/obscura-worker
Fork interno (pin 8a3d048)
~/tools/obscura — remotes: upstream (h4ckf0r0day/obscura), fork
Referencia CLI completa
~/tools/obscura/README.md (secciones CLI Reference, CDP API, MCP)
Variables de entorno
~/tools/obscura/docs/Environment-variables.md
Benchmark de adopcion
output/research/obscura-vs-playwright-20260904.md
Skill deprecada (referencia)
../lightpanda-browser/SKILL.md
Reglas duras :
Si ~/.local/bin/obscura no existe, ABORTA y reporta — no lo descargues sin confirmar.
Fork interno NUNCA recibe push a upstream. Sync solo: git -C ~/tools/obscura fetch upstream.
Apache-2.0: conserva LICENSE/NOTICE del fork. Bundling permitido (a diferencia de Lightpanda/AGPL).
Cuándo usar
Fetch con JS de paginas/SPAs y dump markdown/text/links (uso principal: research y digests)
Scraping paralelo de N URLs (obscura scrape)
Servidor CDP para automation ligera (Playwright connectOverCDP)
Servidor MCP para agentes (stdio o HTTP)
Anti-detect/stealth para sitios con bloqueo headless
Cuándo NO usar
E2E testing de web apps → Playwright + Chromium real (fidelidad de motor)
Visual QA / screenshots de fidelidad pixel → Playwright
HTML estatico simple → curl (mas ligero)
PDF de fidelidad print → Playwright (Obscura exporta PDF pero motor propio)
Decision Checklist
¿Es un test e2e o validacion visual? → SI: Playwright. NO: continuar.
¿Pagina estatica sin JS? → SI: curl. NO: continuar.
¿Datos N3+ en la sesion? → verifica que solo fluyen al sitio objetivo (motor local, cero telemetria).
Abort Conditions
Binario ausente o corrupto → reportar, no reinstalar autonomamente
Sitio exige Chromium real (DRM, WebGL pesado, media) → delegar a Playwright
Workflow
URL(s) + objetivo (markdown/text/eval)
↓
obscura fetch / scrape / serve (según volumen)
↓
output estructurado → digest / research / informe
Detalle de cada paso
Un URL : obscura fetch "$URL" --dump markdown --quiet
Titulo rapido: --eval "document.title"
Screenshot: --screenshot out.png
Dev server local: --allow-private-network
Proxy: --proxy socks5://... (global, antes del subcomando)
Volumen : obscura scrape url1 url2 ... --concurrency 10 --format json --quiet
Sesion interactiva : obscura serve --port 9222 [--stealth] [--obey-robots]
Playwright: chromium.connectOverCDP({ endpointURL: 'ws://127.0.0.1:9222' })
Outputs esperados
stdout: markdown/text/html/json segun --dump/--format
Ficheros: screenshots/PDFs en output/ con naming YYYYMMDD
Memory hooks
Fallos de sitio (Obscura no renderiza X) → bash scripts/memory-store.sh save --type pattern --title "obscura-fallo:<dominio>" --content "<detalle>" --source skill:obscura-browser
Related
Skill: ../lightpanda-browser/SKILL.md (DEPRECATED — sustituida por esta)
Skill: ../tier3-probes/SKILL.md
Research: output/research/obscura-vs-playwright-20260904.md
Upstream: https://github.com/h4ckf0r0day/obscura (Apache-2.0)
1 --- 2 name: obscura-browser 3 description: "Skill: obscura-browser" 4 --- 5 6 # Skill: obscura-browser 7 8 Motor headless nativo de Savia: Rust + V8, motor de render propio, CDP drop-in 9 para Playwright/Puppeteer, MCP nativo, anti-detect opcional, guard SSRF, 10 cero telemetria. Adoptado 2026-09-04 tras benchmark vs Playwright (3x menos 11 RAM, arranque frio 76ms). Depreca a lightpanda-browser. 12 13 ## Authoritative Paths 14 15 > **Lee estos paths antes de actuar. NUNCA asumas firmas, NUNCA inventes paths.** 16 17 | Para | Lee este path | 18 |---|---| 19 | Binario instalado | `~/.local/bin/obscura` (v0.2.1) + `~/.local/bin/obscura-worker` | 20 | Fork interno (pin 8a3d048) | `~/tools/obscura` — remotes: `upstream` (h4ckf0r0day/obscura), `fork` | 21 | Referencia CLI completa | `~/tools/obscura/README.md` (secciones CLI Reference, CDP API, MCP) | 22 | Variables de entorno | `~/tools/obscura/docs/Environment-variables.md` | 23 | Benchmark de adopcion | `output/research/obscura-vs-playwright-20260904.md` | 24 | Skill deprecada (referencia) | `../lightpanda-browser/SKILL.md` | 25 26 **Reglas duras**: 27 - Si `~/.local/bin/obscura` no existe, ABORTA y reporta — no lo descargues sin confirmar. 28 - Fork interno NUNCA recibe push a upstream. Sync solo: `git -C ~/tools/obscura fetch upstream`. 29 - Apache-2.0: conserva LICENSE/NOTICE del fork. Bundling permitido (a diferencia de Lightpanda/AGPL). 30 31 ## Cuándo usar 32 33 - Fetch con JS de paginas/SPAs y dump markdown/text/links (uso principal: research y digests) 34 - Scraping paralelo de N URLs (`obscura scrape`) 35 - Servidor CDP para automation ligera (Playwright `connectOverCDP`) 36 - Servidor MCP para agentes (stdio o HTTP) 37 - Anti-detect/stealth para sitios con bloqueo headless 38 39 ## Cuándo NO usar 40 41 - E2E testing de web apps → Playwright + Chromium real (fidelidad de motor) 42 - Visual QA / screenshots de fidelidad pixel → Playwright 43 - HTML estatico simple → `curl` (mas ligero) 44 - PDF de fidelidad print → Playwright (Obscura exporta PDF pero motor propio) 45 46 ## Decision Checklist 47 48 1. ¿Es un test e2e o validacion visual? → SI: Playwright. NO: continuar. 49 2. ¿Pagina estatica sin JS? → SI: curl. NO: continuar. 50 3. ¿Datos N3+ en la sesion? → verifica que solo fluyen al sitio objetivo (motor local, cero telemetria). 51 52 ### Abort Conditions 53 54 - Binario ausente o corrupto → reportar, no reinstalar autonomamente 55 - Sitio exige Chromium real (DRM, WebGL pesado, media) → delegar a Playwright 56 57 ## Workflow 58 59 ``` 60 URL(s) + objetivo (markdown/text/eval) 61 ↓ 62 obscura fetch / scrape / serve (según volumen) 63 ↓ 64 output estructurado → digest / research / informe 65 ``` 66 67 ### Detalle de cada paso 68 69 1. **Un URL**: `obscura fetch "$URL" --dump markdown --quiet` 70 - Titulo rapido: `--eval "document.title"` 71 - Screenshot: `--screenshot out.png` 72 - Dev server local: `--allow-private-network` 73 - Proxy: `--proxy socks5://...` (global, antes del subcomando) 74 2. **Volumen**: `obscura scrape url1 url2 ... --concurrency 10 --format json --quiet` 75 3. **Sesion interactiva**: `obscura serve --port 9222 [--stealth] [--obey-robots]` 76 + Playwright: `chromium.connectOverCDP({ endpointURL: 'ws://127.0.0.1:9222' })` 77 78 ## Outputs esperados 79 80 - stdout: markdown/text/html/json segun `--dump`/`--format` 81 - Ficheros: screenshots/PDFs en `output/` con naming YYYYMMDD 82 83 ## Memory hooks 84 85 - Fallos de sitio (Obscura no renderiza X) → `bash scripts/memory-store.sh save --type pattern --title "obscura-fallo:<dominio>" --content "<detalle>" --source skill:obscura-browser` 86 87 ## Related 88 89 - Skill: `../lightpanda-browser/SKILL.md` (DEPRECATED — sustituida por esta) 90 - Skill: `../tier3-probes/SKILL.md` 91 - Research: `output/research/obscura-vs-playwright-20260904.md` 92 - Upstream: https://github.com/h4ckf0r0day/obscura (Apache-2.0)
gonzalezpazmonica/pm-workspace commit 9031598168
Frequently asked questions How do I install the Obscura Browser skill? Run npx get-skill add gonzalezpazmonica/obscura-browser in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
What does the Obscura Browser skill do? Skill: obscura-browser It is listed under Coding & Dev Tools on SkillMD.
Is Obscura Browser safe to use? This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: PASS, Skill Scanner: PASS. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
Which AI agents work with Obscura Browser? This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Is Obscura Browser free to use? Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
Who published Obscura Browser? gonzalezpazmonica (@gonzalezpazmonica) published this skill. Their other Agent Skills are listed on their SkillMD profile.