What is wavecap-review?

The WaveCap Review skill streamlines the post-transcription quality assurance workflow by enabling users to verify, correct, and manage automated speech-to-text outputs. It solves the problem of transcription inaccuracies and manual oversight by providing a structured interface to filter low-confidence segments, update review statuses, and maintain a high-fidelity audit trail for all transcription edits.

When should I use wavecap-review?

wavecap-review is useful in the following scenarios: • Quality Assurance: Identify and retrieve transcriptions with low confidence scores (e.g., below 0.7) to manually correct errors and ensure text accuracy. • Workflow Management: Transition transcription statuses from 'pending' to 'verified' or 'corrected' to signal readiness for downstream publishing or archiving. • Audit Tracking: Log reviewer names and timestamps for every correction to maintain accountability and track progress in professional transcription projects. • Batch Review: Efficiently process large volumes of unreviewed audio data by filtering for pending items and performing bulk status updates.

name	wavecap-review
description	Review and correct WaveCap transcriptions. Use when the user wants to mark transcriptions as verified, add corrections, or manage the review workflow.

WaveCap Review Skill

Use this skill to manage the transcription review workflow in WaveCap.

Authentication Required

Review operations require editor authentication:

# Get auth token
TOKEN=$(curl -s -X POST http://localhost:8000/api/auth/login \
  -H "Content-Type: application/json" \
  -d '{"password": "YOUR_EDITOR_PASSWORD"}' | jq -r '.token')

Review Status Values

pending - Not yet reviewed (default)
corrected - Reviewed and corrected
verified - Reviewed and confirmed accurate

Update Transcription Review

Mark as Verified (transcript is accurate)

curl -s -X PATCH "http://localhost:8000/api/transcriptions/{TRANSCRIPTION_ID}/review" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "reviewStatus": "verified",
    "reviewer": "your_name"
  }' | jq

Mark as Corrected (provide corrected text)

curl -s -X PATCH "http://localhost:8000/api/transcriptions/{TRANSCRIPTION_ID}/review" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "correctedText": "The corrected transcription text here",
    "reviewStatus": "corrected",
    "reviewer": "your_name"
  }' | jq

Clear Review (reset to pending)

curl -s -X PATCH "http://localhost:8000/api/transcriptions/{TRANSCRIPTION_ID}/review" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "correctedText": null,
    "reviewStatus": "pending",
    "reviewer": null
  }' | jq

Find Transcriptions to Review

Get pending transcriptions (unreviewed)

curl -s "http://localhost:8000/api/streams/{STREAM_ID}/transcriptions?limit=50" | \
  jq '[.transcriptions[] | select(.reviewStatus == "pending")] | .[:10]'

Get low-confidence transcriptions (likely need review)

curl -s "http://localhost:8000/api/streams/{STREAM_ID}/transcriptions?limit=100" | \
  jq '[.transcriptions[] | select(.confidence < 0.7 and .reviewStatus == "pending")] | sort_by(.confidence) | .[:10] | .[] | {id, text, confidence}'

Count by review status

curl -s "http://localhost:8000/api/streams/{STREAM_ID}/transcriptions?limit=500" | \
  jq '.transcriptions | group_by(.reviewStatus) | map({status: .[0].reviewStatus, count: length})'

Review Response Format

After updating a review, the response contains the updated transcription:

{
  "id": "transcription-uuid",
  "streamId": "stream-id",
  "text": "original transcription",
  "correctedText": "corrected transcription",
  "reviewStatus": "corrected",
  "reviewedAt": "2025-01-15T12:30:00Z",
  "reviewedBy": "reviewer_name",
  ...
}

Batch Review Workflow

For reviewing multiple transcriptions:

# 1. Get pending transcriptions with low confidence
PENDING=$(curl -s "http://localhost:8000/api/streams/{STREAM_ID}/transcriptions?limit=100" | \
  jq -r '[.transcriptions[] | select(.confidence < 0.8 and .reviewStatus == "pending")] | .[].id')

# 2. Loop through and review each
for ID in $PENDING; do
  echo "Reviewing: $ID"
  # Get the transcription
  curl -s "http://localhost:8000/api/streams/{STREAM_ID}/transcriptions?limit=500" | \
    jq --arg id "$ID" '.transcriptions[] | select(.id == $id) | {text, confidence, recordingUrl}'

  # Mark as verified (or prompt for correction)
  # curl -s -X PATCH "http://localhost:8000/api/transcriptions/$ID/review" ...
done

Tips

Always include a reviewer name for audit trail
Use correctedText: null to clear a previous correction
Low confidence scores (< 0.7) often indicate transcription errors
The reviewedAt timestamp is set automatically by the server

wavecap-review

When & Why to Use This Skill

Use Cases