Privacy-Safe Sapat Transcription in Daytona

[zergzorg]

/claim #13

Summary

• Add a long-form Daytona guide for running Sapat as a privacy-aware transcript review workflow.
• Cover Daytona workspace setup, Sapat provider configuration, single-file and directory transcription, local redaction, review checklist, raw-vs-redacted diff review, and safe handoff artifacts.
• Add a companion definition for transcript redaction workflow, a minimal first-time author profile, and an original SVG workflow illustration.

Scope and non-overlap

• This is a provider-agnostic privacy/redaction workflow for Sapat in Daytona, not a new Sapat transcription provider.
• The guide deliberately focuses on workspace guardrails, provider-upload decision notes, redaction smoke testing, raw transcript handling, and reviewed artifact handoff.
• It complements provider-specific Sapat/Daytona guides and companion provider implementations by documenting the safety gate that should wrap any provider path.

Validation

• ./node_modules/.bin/markdownlint guides/20260520_privacy_safe_sapat_transcription_daytona.md definitions/20260520_definition_transcript_redaction_workflow.md authors/zergzorg.md
• xmllint --noout guides/assets/20260520_privacy_safe_sapat_transcription_daytona_img1.svg
• git diff --check
• npm run lint:md was also run; it still fails on pre-existing legacy markdown issues outside this PR, not on the new files.

Notes

• Source-checked against the Sapat README, CLI source, and documented provider behavior.
• No API keys, recordings, payment details, private data, or third-party account credentials are included.
• Transparency: AI-assisted with Codex, then edited and checked against the referenced source material before submission.

[zergzorg]

Follow-up after checking newer #13 attempts:

• Strengthened the guide as a source-checked privacy/review runbook for existing Sapat providers instead of adding another provider slice.
• Added explicit scope/non-overlap notes, no-key CLI/source checks, an artifact boundary table, a reusable local redaction script, a synthetic redaction smoke test, and extra troubleshooting grounded in Sapat source behavior.
• This stays complementary to provider-specific PRs such as Together AI, fal.ai, Gladia, and similar slices because it does not claim or add a provider integration; it wraps the existing Sapat path with reproducibility and review controls.
• No credentials, private data, payment details, KYC/account data, or secrets are included.

Validation:

• ./node_modules/.bin/markdownlint guides/20260520_privacy_safe_sapat_transcription_daytona.md definitions/20260520_definition_transcript_redaction_workflow.md authors/zergzorg.md — passed (only Node punycode deprecation warning).
• xmllint --noout guides/assets/20260520_privacy_safe_sapat_transcription_daytona_img1.svg — passed.
• git diff --check — passed.
• Guide word count: 2,289 words.

[zergzorg]

Follow-up after checking newer #13 guide attempts:

• Added a provider-upload decision note before the first real transcription run, so the guide captures whether the recording is approved for provider upload before any external call.
• Added *.mp3 to the local ignore guidance and a raw-vs-redacted diff -u review gate in the checklist.
• This keeps the PR positioned as a source-checked privacy/review runbook around existing Sapat providers, not another provider-specific integration.
• No credentials, recordings, private data, payment details, KYC/account data, or secrets are included.

Validation:

• ./node_modules/.bin/markdownlint guides/20260520_privacy_safe_sapat_transcription_daytona.md definitions/20260520_definition_transcript_redaction_workflow.md authors/zergzorg.md
• xmllint --noout guides/assets/20260520_privacy_safe_sapat_transcription_daytona_img1.svg
• git diff --check
• Guide word count: 2,425 words.