Works with the AI Agents you already use
Quote Lock: Check Article Quotes Against Transcripts
**The pain.** AI drafting tools quietly reshape quotes: a number is rounded, a "not" disappears, two remarks become one sentence, one person's line is credited to another.
$12
Quote Lock: Check Article Quotes Against Transcripts
Example session with this skill installed
Check the quotes in my article draft (draft-sample.md) against the two interview transcripts: the English panel interview captions (interview-ortiz.srt.txt) and the Arabic radio interview with Dr. Salem (interview-salem-ar.txt). Tell me which quotes I can publish and what to fix.
- Read your context and instructions
- Compiled the quote lock
| # | Verdict | Draft line | Speaker (draft -> transcript) | Where | Quote |
|---|---|---|---|---|---|
| 1 | ALTERED (CRITICAL) | 3 | Lena Ortiz -> Lena Ortiz | interview-ortiz.srt.txt 00:00:09 | 40 million for the river wall this year, |
| 2 | VERBATIM | 6 | Lena Ortiz -> Lena Ortiz | interview-ortiz.srt.txt 00:00:41 | The wall comes first because it protects the most homes, |
| 3 | CLEANED | 8 | Lena Ortiz -> Lena Ortiz | interview-ortiz.srt.txt 00:00:16 | We think that covers the first phase, but not the pumping... |
| 4 | WRONG SPEAKER | 11 | Lena Ortiz -> Tom Becker | interview-ortiz.srt.txt 00:00:33 | The old ones failed twice in 2024 and nobody was held res... |
| 5 | STITCHED | 14 | Lena Ortiz -> Lena Ortiz | interview-ortiz.srt.txt 00:00:49 +1 | The pumps could start next spring. Construction starts in... |
| 6 | ELIDED (HIGH) | 17 | Tom Becker -> Tom Becker | interview-ortiz.srt.txt 00:00:27 +1 | Honestly I … think the pumps can wait another winter. |
| 7 | NOT FOUND | 20 | Lena Ortiz -> ? | - | will protect every single home in the valley by next autumn. |
| 8 | VERBATIM | 27 | Dr. Salem -> Dr. Salem | interview-salem-ar.txt 00:00:14 | عندنا 32 مدرسة تحتاج صيانة، وسنبدأ بالمدارس الأقدم |
| 9 | ALTERED (CRITICAL) | 29 | Dr. Salem -> Dr. Salem | interview-salem-ar.txt 00:00:06 | المدارس ستغلق هذا الشتاء |
| 10 | TRANSLATED | 31 | Dr. Salem -> ? | - | Teachers' salaries will rise by 10 percent from January. |
1. ALTERED - CRITICAL
Draft (line 3): "40 million for the river wall this year,"
Transcript (interview-ortiz.srt.txt, 00:00:09, Lena Ortiz): "14 million for the river wall this year."
- CRITICAL: draft "40,000,000" vs spoken "14,000,000" (number changed)
Fix: quote the exact words - "14 million for the river wall this year." - or remove the quotation marks and paraphrase. If the transcript is machine-made, listen at 00:00:09 first.
6. ELIDED - HIGH
Draft (line 17): "Honestly I … think the pumps can wait another winter."
Transcript (interview-ortiz.srt.txt, 00:00:27, Tom Becker): "Honestly I"
Transcript (interview-ortiz.srt.txt, 00:00:27, Tom Becker): "think the pumps can wait another winter."
- Omitted by the ellipsis: "don't" - RISK: contains a negation - check the meaning did not flip
Fix: read the omitted words above; if leaving them out changes the meaning, quote the full passage.
9. ALTERED - CRITICAL
Draft (line 29): "المدارس ستغلق هذا الشتاء"
Transcript (interview-salem-ar.txt, 00:00:06, Dr. Salem): "المدارس لن تُغلق هذا الشتاء."
- CRITICAL: draft "ستغلق" vs spoken "لن تغلق" (negation changed)
Fix: quote the exact words - "المدارس لن تُغلق هذا الشتاء." - or remove the quotation marks and paraphrase. If the transcript is machine-made, listen at 00:00:06 first.
Connects securely to your tools. The creator never sees your data.
About this skill
The pain. AI drafting tools quietly reshape quotes: a number is rounded, a "not" disappears, two remarks become one sentence, one person's line is credited to another. Newsrooms in 2025-2026 printed corrections and suspended staff over exactly this. Checking quotes by hand means scrubbing through an hour of transcript for every line.
The outcome. One report that lists every quotation in your draft with a verdict, the file and timestamp where it was said, who said it, the exact word-level differences, and a fix line giving the words to use instead. Listen to the timestamp, fix the quote, publish.
How it works. A standard-library Python script (no network, no packages) extracts quotes from your .md or .txt draft ("...", “...”, «...», blockquotes), finds each one in your transcripts (SRT, VTT, timed plain text, Otter-style exports, ElevenLabs Scribe or Whisper JSON) and classifies it:
- VERBATIM / CLEANED (fillers and stutters dropped)
- ELIDED: shows what an ellipsis, a [bracket] or a partial quote left out, with a RISK line when that was a negation, a condition or a number
- ALTERED: CRITICAL when a number or negation changed, HIGH for scope words (only, always, may)
- STITCHED: real words from different moments or two speakers joined into one quote
- WRONG SPEAKER, TRANSLATED, NOT FOUND (closest passage shown)
Arabic is handled properly on both sides: tashkeel, hamza/alef forms, ta marbuta, alef maqsura, tatweel and Arabic-Indic digits are normalised, and Arabic negations (لا، لم، لن، ما، مش، ليس) count as CRITICAL changes. English number words and contractions are normalised too ("twenty five" = 25, "don't" = "do not"). Exit codes (0/1/2) make it usable in a pre-publish check.
Works better with a speech-to-text tool when you only have audio: ElevenLabs Scribe through Agensi (no API key, per-minute from your balance) gives a word-timed, speaker-labelled transcript the skill reads directly; Whisper, Otter, Descript or YouTube captions work too. Without one, the skill still checks any transcript you already have.
Limits. It checks wording and attribution against the transcripts you supply, not whether a statement is true. Translated quotes are flagged for hand-checking, not matched. Attribution is read from the draft's own sentence; "she said" with no name nearby is reported as not checked. Run it with no arguments to see a full labelled demo first.
Checks wording and attribution only against the transcripts you supply; a quote from an unrecorded conversation shows NOT FOUND. Translated quotes are flagged, not matched word for word. Attribution is read from the draft's own sentence; a pronoun with no name in the paragraph is reported as not checked, and names in another script need --speaker-map. No synonym matching on purpose ("close" vs "shut" is ALTERED). Single-quoted ('...') quotes are not extracted. Arabic number words are compared as words, not values. In very long transcripts, a short quote made only of very common words may come back NOT FOUND instead of ALTERED.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 3 days ago
Requires
Permissions
File Scopes
The script makes no network calls. The optional ElevenLabs Scribe transcription runs through Agensi's tool, priced per minute from your balance, only if you choose to transcribe audio.