feat(short): generación y render de Shorts vía shortsmith

Añade /generate short_en y /short_spec. El pipeline genera un shot spec
con Haiku, verifica cada cifra, fecha y cita contra los chunks de la
sesión, lo renderiza en shortsmith y entrega el MP4 por Telegram junto
a un informe de claims.

- ShortsmithClient con sondeo y fallback al spec JSON si el render falla
- Contrato de plantillas obtenido de GET /templates, no codificado
- Comprobación de fundamento determinista, sin LLM
- outputs.published_url para enlazar el artículo de Ghost
- Normalización de comillas rectas a tipográficas (ver KNOWN-ISSUES.md)

Lo que no aparece en los chunks se contrasta contra el ejemplo del
prompt: si casa ahí es fuga, no invención, y se informa como tal. El
purgado de sesiones se lleva también su MP4.

La subida a YouTube queda fuera a propósito: fase 3.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

README.md

@@ -39,9 +39,50 @@ OutputGenerator (Ollama)
 | `/status` | Check progress |
 | `/finish` | Stop early, proceed to generation |
 | `/generate podcast\|blog\|report\|thread` | Generate output |
+| `/generate short_en` | Vertical Short: shot spec → grounding check → MP4 |
+| `/short_spec` | Last shot spec as a JSON file, to hand-edit and re-render |
 | `/sources` | List all sources found |
 | `/cancel` | Cancel current research |
 
+## Shorts (`/generate short_en`)
+
+Claude writes a **shot spec** — typed JSON, not prose — which
+[shortsmith](https://git.chemavx.xyz/chemavx/shortsmith) renders into a 1080×1920
+MP4. The bot sends the video and, in a separate message, a **claims report**.
+
+```
+/research JAL 1628 Alaska 1986 …
+/generate blog en          → Ghost draft, article URL stored on the output row
+/generate short_en         → spec → grounding → render → video + claims report
+                             (YouTube upload is deliberately manual)
+```
+
+Three things make this different from generating text, and each has its own
+mitigation:
+
+- **It is a contract, not prose.** The template schemas are fetched live from
+  `GET /templates` and never copied here, so a template added to shortsmith is
+  available immediately. A spec is validated locally against those schemas
+  before anything renders, and the exact error paths
+  (`shots.0.radar_sweep.props.sweeeps`) go back to the model verbatim — up to 3
+  attempts.
+- **It contains figures and quotes.** `grounding.py` extracts every quote,
+  figure, date and proper noun and checks it against the exact chunks the model
+  was given. No LLM in that path: normalisation plus substring, deterministic
+  and free. Whatever is not in the chunks is checked against the worked example
+  that travels in the prompt, so a figure lifted from it is reported as a
+  **prompt leak**, not as an invention — different diagnosis, different fix.
+  Neither ever blocks the render: both are surfaced next to the video and a
+  human decides.
+- **It becomes a published video.** Nothing is uploaded anywhere. The MP4 lands
+  in Telegram for review, and in `/data/shorts/{session_id}.mp4`.
+
+Fallbacks hold throughout: if shortsmith is unreachable, the job errors, or the
+spec never validates, the spec JSON comes back as a file. The expensive part is
+the generation, not the render.
+
+Full spec of the phase: `docs/shortsmith-phase2-spec.md`.
+
 ## Local Development
 
 ```bash