> ## Documentation Index
> Fetch the complete documentation index at: https://api-docs.recnote.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI agent skill

> Drop-in skill so your AI coding agent integrates Recnote for you.

If your team builds with an AI coding agent (Claude Code, Cursor, or any agent
that loads a `SKILL.md` / rules file), you can hand it this skill so it knows
the Recnote API end to end — endpoints, auth, payloads, debugging — and can
scaffold a client or test calls without you re-explaining the API each time.

## Install

Save the block below as `.claude/skills/recnote-api/SKILL.md` in your project
(or paste it into your agent's rules / context file). Then ask your agent
things like *"send consultation.wav to Recnote and show me the note"* or
*"scaffold a Python client for the chunked flow"*.

<CodeGroup>
  ```markdown SKILL.md theme={null}
  ---
  name: recnote-api
  description: "Integrate or test the Recnote API (transcription + AI clinical-note generation). Use when calling /api/v1, scaffolding a client, sending audio for a note, streaming chunks, or debugging a 401/502."
  ---

  # Recnote API

  Base URL: https://api.recnote.ai
  Auth: `Authorization: Bearer rec_sk_...`  (or header `X-API-Key: rec_sk_...`)
  Set the key as env RECNOTE_API_KEY before running examples.

  ## Endpoints
  - GET  /api/v1/ping                         verify key
  - POST /api/v1/transcriptions               audio -> { transcript }
  - POST /api/v1/notes                         audio OR transcript -> { transcript, note }
  - POST /api/v1/sessions                      open chunked session -> { sessionId }
  - POST /api/v1/sessions/:id/chunks           push one audio chunk
  - POST /api/v1/sessions/:id/notes            generate from accumulated chunks

  Metadata for note endpoints (multipart fields or JSON keys):
  language (sv/en/fi/nb/da/de/tr/ar), templateSections (JSON string),
  template (id), sideNote, primaryProfession, customTemplate.

  ## One-shot note from audio
  curl -X POST "$RECNOTE_BASE/api/v1/notes" \
    -H "Authorization: Bearer $RECNOTE_API_KEY" \
    -F "file=@consultation.wav" -F "language=sv"

  ## Note from transcript (JSON)
  curl -X POST "$RECNOTE_BASE/api/v1/notes" \
    -H "Authorization: Bearer $RECNOTE_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{"transcript":"...","language":"sv"}'

  ## Custom template
  Add field: templateSections='[{"sectionName":"Anamnes","aiRules":{"autoFill":true}}]'
  Returned note.sections mirrors it; note.combined is full text.

  ## Chunked streaming
  SID=$(curl -s -X POST "$RECNOTE_BASE/api/v1/sessions" -H "Authorization: Bearer $RECNOTE_API_KEY" | jq -r .sessionId)
  curl -X POST "$RECNOTE_BASE/api/v1/sessions/$SID/chunks" -H "Authorization: Bearer $RECNOTE_API_KEY" -F "file=@chunk.webm" -F "language=sv"
  curl -X POST "$RECNOTE_BASE/api/v1/sessions/$SID/notes" -H "Authorization: Bearer $RECNOTE_API_KEY" -H "Content-Type: application/json" -d '{"language":"sv"}'

  ## Response
  { "transcript": "...", "note": { cause, subjective, objective, assessment,
    plan, diagnoses[], coding_codes[], sections[], combined, tasks } }
  note.combined drops straight into an EHR text field.

  ## Errors
  401 invalid key · 400 no file/transcript · 413 audio >500MB · 502 upstream failed
  ```
</CodeGroup>

## What the agent can then do

* Generate a typed client (Node, Python, C#, …) for either flow.
* Wire the chunked session loop to your recorder.
* Map the `note` JSON onto your EHR fields.
* Reproduce and explain failed calls (`401`, `502`, …) against your key.

<Note>
  The skill contains no secrets — set your `RECNOTE_API_KEY` in the environment
  your agent runs in, never inside the skill file.
</Note>