Skill instructions
traspion-account-settings1.77 KB
View saved version →
---
name: traspion-account-settings
description: Inspect Traspion plan usage, billing history, spelling replacements, or active vocabulary. Use for read-only questions about the connected account, recent payments, and language settings.
---
# Inspect Traspion account settings
Use the narrowest read-only tool that answers the request:
- Use `get_account_status` for plan, wallet-balance, and transcription-usage questions. It returns a link to the Traspion billing page.
- Use `get_billing_history` when the user asks about payments, purchases, charges, credits, or transaction history. Request only as many recent records as needed, up to 100. If `hasMore` is true, explain that older records are available in the Traspion billing dashboard.
- Use `get_spelling_replacements` for custom spelling replacement pairs.
- Use `get_vocabulary` for active vocabulary entries.
Keep these boundaries clear:
- Report only the fields returned by the tool. Do not expose or infer payment, order, subscription, or user identifiers, payment methods, failure details, purchase totals, or other private billing data.
- These tools do not change account or language settings. Do not claim that a setting was added, removed, or updated.
- Do not claim that vocabulary or spelling settings were applied to an upload unless an upload result explicitly confirms it.
- Present the exact billing-page link returned by `get_account_status` whenever reporting plan or wallet status, regardless of the available minutes.
- If an entitlement is unavailable, explain the account status and include the billing-page link returned by the tool. Do not claim to complete purchases or subscription changes in ChatGPT.
- Billing changes, purchases, subscription management, settings changes, and deletion are unavailable through this plugin.
traspion-transcribe-media5.3 KB
View saved version →
---
name: traspion-transcribe-media
description: Transcribe an audio or video file with Traspion, whether it is attached, shared as a link (Google Drive, Dropbox, OneDrive, or a direct https link), or still on the user's device (upload panel), including speech-to-text, audio-to-text, video-to-text, speaker detection, subtitle, or translation requests. Use when the user wants a new recording converted into a transcript.
---
# Transcribe audio or video to text with Traspion
Pick one tool for the request:
- **An attached audio or video file** (from the ChatGPT library or the composer): call `upload_and_transcribe` once for it.
- **A link in the message** (Google Drive, Dropbox, OneDrive or SharePoint, or a direct https link to the media file): call `transcribe_from_link` once with it.
- **Neither**: call `upload_media` once, straight away, without asking anything first. It shows the Traspion upload panel in the chat, where the user drops or chooses a file from their device, or pastes a share link, and picks speaker labels, subtitles (SRT, WebVTT) and PII redaction. In one short sentence, tell the user to finish there; don't list steps or options, since the panel shows them. The panel uploads or imports the file, shows the transcription's progress itself, and shows the transcript itself in its viewer, and opens exports itself. It posts into the chat only when a button needs you (for example, if an export link can't open). Don't check the status or call `get_transcription` for a panel file until the user asks. Each panel holds one file; when the panel asks for a new upload panel, call `upload_media` once with no preset settings. The panel handles its own refusals: it offers the billing page, Retry, and Replace waiting file, which deletes the waiting file only when the user presses it. Don't call `upload_media` again for the same file, and don't call `upload_and_transcribe` or `transcribe_from_link` for a file the panel handles.
- If the panel doesn't appear, ask the user to attach the file or share a link to it.
- If several attachments or links make the target ambiguous, ask the user to identify one file.
- `transcribe_from_link` doesn't accept YouTube, Zoom, or folder links. If it refuses a link, relay its message and suggestion; don't retry the same link or try another tool with it.
- For an attachment or a link in the message, if the user has not specified speaker handling, ask one combined question before uploading: “Would you like speaker labels? If yes, how many speakers are there (2–30), or should Traspion detect the number automatically?” Do not ask a second confirmation after the answer; the original transcription request already authorizes the upload or import.
- If the answer is no, set `speakerPartitioning: false` and omit `speakersCount`. If the answer is yes with a count, set `speakerPartitioning: true` and pass that count. If the answer is yes but the count is unknown, set `speakerPartitioning: true` and omit `speakersCount` for automatic detection.
- Skip the speaker question when the user already specified whether speaker labelling is wanted, and always for `upload_media`: the panel asks it. Pass to `upload_media` only what the user already said (for example `speakerPartitioning`, `speakersCount`, `piiRedaction`, `srtSubtitles`); the panel starts from those values. Never guess the number of speakers or attempt to determine it before transcription.
- Set `channelIdentification: true` only when the user requests channel identification. Speaker diarization and channel identification may both be enabled when requested; omit `speakersCount` in that combination so AssemblyAI detects speakers automatically within each channel.
- Use the tool defaults for other unspecified options. Pass a language, translation target, transcription prompt, PII redaction, or profanity filtering only when the user requests it or supplies the value.
- Do not claim that advanced options unavailable in these tools were enabled.
- Do not aut
Preview of the first 4,000 characters. Read full instructions →
traspion-work-with-content6.29 KB
View saved version →
---
name: traspion-work-with-content
description: Find, retrieve, translate, edit, save, rename, or export an existing Traspion speech-to-text transcript. Use when the user wants to work with audio-to-text or video-to-text content already stored in Traspion.
---
# Work with Traspion content
Resolve the intended file before using a file-specific tool.
- Reuse a `fileId` already returned for the clearly identified file.
- Otherwise call `list_files`, using its filename, date, or status filters when the user supplied those details. If several files could match, ask the user to choose; do not silently infer an ID. A request for the latest file may use the most recent matching list result.
- When showing files returned by `list_files`, present them as a Markdown table with row number, filename, upload time, and status. Keep the returned order and use the row number to resolve a user's follow-up selection. Do not expose a pagination cursor unless it is needed for the next page.
- To show or open a transcript, or to edit it without saying what to change, call only `open_transcript`: it shows Traspion's panel for that file with View transcript, Edit, and Export. Don't fetch the text first, and reply with one short sentence pointing to the panel. Do the same when the user switches to another file later in the conversation.
- Call `get_transcription` when the text itself is needed: to read, quote, summarize, or translate it, to make a change the user describes, or to check a file's status.
## Retrieve and translate
- Use the returned pagination flags and next offsets when more transcript content is required. Continue only until the user's task has enough content or the complete requested transcript has been retrieved.
- Clearly identify partial results. Never describe a truncated transcript as complete.
- When the user asks for a transcript's text in the chat, or for a stored translation, retrieve all pages needed and show the complete text right away in ChatGPT's native editable view when available, without asking first or offering to split it; length is not a reason to hold back. Otherwise show the text in the conversation, continuing in the next reply if it doesn't fit. If client limits prevent showing it all, identify the result as partial and offer a complete export. Apart from the Edit button in Traspion's upload panel, the plugin does not supply its own editor.
- Follow the returned `presentationMode`. When both speaker and channel identification are enabled, present each segment once in a combined form such as `[00:00–00:08] Channel 1 · Speaker A: text`. Preserve the returned labels and timestamps, omit a label only when it is absent, and do not duplicate the same content as a separate plain transcript unless the user explicitly asks for plain text.
- Use `get_translation` first when the user asks for a translation, because a stored Traspion translation may have been requested during transcription. If it is still processing, report the status and stop after that check.
- If `get_translation` reports `TRANSLATION_NOT_REQUESTED`, tell the user: "You didn't opt for translation while transcribing, so I'll use ChatGPT to translate the transcription." Retrieve the complete transcription and translate it directly in ChatGPT into the user's requested language. If a requested stored translation is still processing, report that status instead of substituting ChatGPT translation.
- Present that fallback as **Transcript translation**. Translate faithfully and preserve the transcript's meaning, paragraph structure, speaker labels, and timestamps when present. Do not summarize, paraphrase, polish, add commentary inside the translation, or claim that Traspion generated or stored it.
- After translating the complete transcript in ChatGPT, present it in a native editable view if the client offers one. Otherwise present the complete translation in the conversation under **Transcript translation**.
- Never save a ChatGPT-produced translation
Preview of the first 4,000 characters. Read full instructions →