/adobe-summarize-media

name: adobe-summarize-media
description: >
  Summarizes the spoken content of a video or audio recording (meeting, interview, lecture,
  podcast, voice memo) with Adobe and saves the summary as a PDF. Use for requests like "summarize
  this recording", "summarize this video/audio", "give me a summary of this meeting recording", or
  "turn this interview into a PDF summary". Do NOT use for full transcripts or captions, for
  silent or music-only footage, or for visual highlight reels.
license: Apache-2.0
compatibility: "Runs on widget-capable surfaces (Claude, which supports the asset_add_file picker) and on surfaces without widgets (Codex), which use the asset_initialize_file_upload → PUT → asset_finalize_file_upload staging path. Raw local paths are never passed to Adobe tools."
allowed-tools: adobe_mandatory_init asset_add_file asset_openai_file_upload asset_initialize_file_upload asset_finalize_file_upload media_summarize markdown_to_pdf pdf_operation_status
metadata:
  version: 1.0.0
  visibility: public
  surface: [claude, codex]

Adobe Summarize Media

Summarizes what is **said** in a video or audio recording and delivers the summary as a **PDF**, plus a short version in chat.

> **Scope:** summaries only. Full transcripts and captions are out of scope for this skill.

> **Async handling (`media_summarize`):** a `status: "working"` result is **pending — not a failure and > not missing output**. The finished result is delivered automatically when the job completes; do > not call any poll tool yourself. Submit each job **once** — never resubmit while one is pending. > If the result doesn't arrive, tell the user the job was submitted and is still processing, and > don't offer workarounds.

---

Tool Reference

| Step | Tool | Notes | |------|------|-------| | Initialize | `adobe_mandatory_init` | Returns file-handling rules and tool routing | | Select recording | `asset_add_file` | File picker widget; injects the asset ID on confirmation | | Upload file attached in ChatGPT | `asset_openai_file_upload` | ChatGPT only; returns a URL for the Adobe tools | | Upload attached/local file | `asset_initialize_file_upload` | Starts the upload; returns a PUT URL | | Finish upload | `asset_finalize_file_upload` | Completes the upload; returns the asset ID | | Summarize spoken content | `media_summarize` | Needs significant dialogue or narration | | Save summary as PDF | `markdown_to_pdf` | Converts inline Markdown to a PDF | | Wait for PDF | `pdf_operation_status` | Only when `markdown_to_pdf` returns a `tracking_id` |

---

Workflow

Step 0 — Initialize Adobe Tools

Call `adobe_mandatory_init` first. This returns file handling rules and tool routing guidance required for the rest of the workflow.

{ "skill_name": "adobe-summarize-media", "skill_version": "1.0.0" }

---

Step 1 — Get the Recording

`media_summarize` accepts a Creative Cloud asset ID (e.g. `urn:aaid:sc:...`) or a presigned HTTPS URL from an approved storage host. Never pass a raw local path.

| Source | Action | |---|---| | File attached in ChatGPT and `asset_openai_file_upload` is available | Call `asset_openai_file_upload` with the attached file and use the URL it returns. Don't pass the attached file's local path to `asset_upload_file_from_url` — it only accepts https URLs. | | File attached in chat (e.g. `/mnt/user-data/uploads/…`) | Check egress status from `adobe_mandatory_init` (Step 0). If egress is enabled, upload it directly: get its size and MIME type, call `asset_initialize_file_upload({ path, media_type })`, PUT the bytes to the returned URL, then call `asset_finalize_file_upload({ filename, transfer_document })` and use the returned asset ID. If egress is disabled, open the picker (`asset_add_file()`) instead. | | Creative Cloud asset ID or presigned URL in the message | Use it directly. | | No file provided yet | Open the picker — no need to ask first. |

Opening the picker:

> *"Let's summarize your recording. Start by selecting your file:"*

asset_add_file()

Wait for the widget to inject the asset ID on confirmation — do not poll.

> **No-widget fallback** *(only if `asset_add_file` is unavailable on this surface, e.g. Codex)* — > stage any local file with the same `asset_initialize_file_upload` → PUT → > `asset_finalize_file_upload` sequence. If egress is disabled, tell the user the file can't be > staged on this surface. If no file was provided, ask the user to attach their video or audio file.

If the selected file is not `video/*` or `audio/*`, say so and ask for a video or audio recording.

---

Step 2 — Summarize the Spoken Content

media_summarize({
  assetId: "<asset ID or presigned URL>",
  filename: "<recording-name>-summary"
})

Use a short, descriptive `filename` without an extension (e.g. `"team-sync-summary"`).

Acknowledge briefly while it runs: *"Summarizing your recording — this can take a moment for longer files."* If the result reports a pending/working status, wait for the completed result; never act on a pending status.

From the completed result, keep the **summary text**. Then continue to Step 3 right away, in the same reply — the job isn't done until the PDF is created. Don't stop and deliver only the summary text.

---

Step 3 — Save the Summary as PDF

Format the summary text from `media_summarize` as Markdown — a `# <recording name> — Summary` title followed by the summary content — and pass it to `markdown_to_pdf`. Use only content from the `media_summarize` result; do not add facts.

markdown_to_pdf({ /* Markdown summary, per the tool's input schema */ })

If `markdown_to_pdf` returns a `tracking_id`, call `pdf_operation_status` with that exact `tracking_id` and `operation_name: "markdown_to_pdf"` until it returns a final result.

Present only a PDF URL that the tools actually returned — never reconstruct or invent one.

---

Step 4 — Deliver

Reply with:

1. The PDF link: `<recording