Soundings 0.3.0

0.3.0 — 2026-10-02

• Add an optional Transcript display setting: Folded callout places the full transcript in a collapsed Full Transcript quote callout beneath the Transcript heading, with speaker headings and retained timestamps inside.
• Keep plain output as the default, including for existing saved settings. Existing notes are never rewritten.
• Include transcript display in the review summary and invalidate reviewed plans when it changes.
• Verify containment with CommonMark structural tests, adversarial transcripts, and the 5 MB source limit.

Soundings 0.2.3

0.2.3 — 2026-10-01

Soundings 0.2.3 is a small corrective release. It behaves exactly like 0.2.2 for you, and it clears the one warning Obsidian's automated Community review reported.

🔍 Community review

• Type-safety warning fixed: Obsidian's scanner flagged that settings validation passed an untyped value into a typed check (no-unsafe-argument in src/core/settings.ts). The values were already validated, so nothing was unsafe at runtime, but the check now proves the type instead of assuming it. Accepted and rejected settings are unchanged.
• Same review rules, run locally: every change is now linted with Obsidian's official ESLint rules and type-aware TypeScript checks before it can be merged or released, so this kind of warning cannot slip through again.

📄 Project

• Added a contributing guide covering setup, safety rules, the spec-first workflow, and private security reporting.

Notes

• No change to how transcripts are found, converted, or written. Existing notes are never modified.

Soundings 0.2.2

0.2.2 — 2026-09-30

Soundings 0.2.2 is a hardening patch for 0.2.1, from a full code and security review. It adds no features, no network access, and no new way of changing your vault. Every new refusal leaves your files untouched and explains why.

🔒 Safety and stability

• No more freezes on hostile captions: A small crafted .vtt file could freeze Obsidian during a scan, or, with observation on, as soon as it synced in. Caption markup is now read in a single pass, and 1 MB adversarial files classify in milliseconds. A cue tag longer than 256 characters, or a tag containing a second < (for example a <<i>x</i>), is refused. A lone < is still kept as text.
• One bad file no longer stops a batch: A transcript that fails to render is reported as failed and the remaining selected files still convert.
• Letter case and accents count as the same name: Bar.txt next to an existing bar.md, or Foo.txt next to foo.vtt, is now blocked instead of shown as ready. The same applies to names that differ only in how accented characters are encoded. Soundings also checks storage directly just before writing, so a file Obsidian has not indexed still blocks conversion.
• No hidden notes: A name that would become a dot-file after cleanup (for example ?.env.txt → .env.md) is refused, because Obsidian would hide it.
• Disabling Soundings stops everything: Dialogs close when the plugin unloads, and a review or enrichment form left open can no longer write afterwards. Closing the progress dialog with Escape now cancels the run.

🗣️ WebVTT faithfulness

• Text after a closing </v> no longer goes to the previous speaker.
• Tags that merely start with v, such as <video>, are refused like other unknown tags instead of being silently removed.
• A second, genuine U+FEFF character at the start of a file is kept.

👀 Observation

• Observation starts after Obsidian finishes loading the vault, so existing transcripts are no longer re-queued on every launch.
• New notes, images, and other non-transcript files are filtered by path alone, so sync bursts no longer make Soundings "busy".
• Changing settings other than observation on/off or observation roots keeps the inbox.

📝 Notes and enrichment

• Speaker and title headings escape %, $, =, ~, and ^, so a speaker named %% cannot hide the rest of the transcript and $5 … $10 does not render as math. Generated Markdown source changes only for headings that contain these characters.
• Invisible control characters in filenames are written as \uXXXX escapes in frontmatter.
• In manual enrichment, typed lines such as ~~~, a backtick fence, ---, or === stay visible text and no longer change the companion note's sections.
• Manual enrichment is refused for a note whose name contains [ ] | # ^ < >, because the backlink could not point at it safely. Rename the note first.

⚙️ Settings

• Saved settings with the wrong type (for example "false" as text) reset to safe defaults field by field, with a notice naming them. A malformed settings file no longer prevents the plugin from loading.

🧰 Release tooling

• The runtime audit now requires the bundle to load only obsidian, and blocks more network, code-loading, and destructive vault patterns.
• Release staging writes only to release/<version> and refuses symbolic links.

Notes

• Existing notes are never modified. Only new conversions and new companion notes are affected.

Soundings 0.2.1

0.2.1 — 2026-09-29

Soundings 0.2.1 is a corrective patch for 0.2.0. It makes WebVTT speaker attribution faithful, keeps manual-enrichment drafts safe, and ships the enrichment hardening that was verified for, but missing from, the published 0.2.0 assets.

🗣️ WebVTT faithfulness

• Per-speaker attribution: A caption cue containing several voices (for example <v Alice>… then <v Bob>…) now produces one block per speaker. In 0.2.0 every line in such a cue was attributed to the first speaker.
• Header metadata: Files with Kind: or Language: lines under WEBVTT (common in YouTube captions) are now eligible instead of unreadable.
• Cue separators: A line containing only spaces between cues now separates them; it no longer merges two cues into one block.
• Character references: < now renders as < instead of being decoded twice.

✍️ Manual enrichment

• If publishing a companion note does not succeed (for example because the transcript note changed or the destination appeared), the form stays open with everything you typed. After a source change, Soundings re-checks the note before you review again.
• Source notes are identified only from their exact Soundings frontmatter (type, source, and a supported soundings_version), including notes whose metadata contains ---.
• The companion backlink uses the full vault path, so same-named notes in different folders link correctly.

🔒 Release integrity

• Releases are built by a read-only job and published by a separate job with GitHub artifact attestations for main.js, manifest.json, and styles.css.
• Only bare version tags (for example 0.2.1) publish releases, matching what Obsidian installs.

Notes

• Existing notes are never modified. Only new conversions of multi-voice cues use the new per-speaker layout.

Soundings 0.2.0

🚀 Soundings v0.2.0

Soundings 0.2.0 introduces configurable note output profiles, enhanced WebVTT transcript compatibility, pre-scan classification, and local manual companion notes.

✨ New Features

🎛️ Configurable Output Profiles

You now have full control over how your companion notes are created. Instead of a fixed format, customize output in Settings → Soundings:

• Flexible Naming: Choose how your notes are titled and named. Use the source name alone or prefix it with the immediate parent folder to keep your vault organized.
• Selective Content: Toggle specific enrichment sections on or off (Summary, Decisions, Action Items, Follow-ups).
• Custom Tags: Add static tags to every companion note you generate, making them easy to find via Obsidian search.
• Timestamp Control: For WebVTT transcripts, choose whether to omit or retain normalized start and end times in your notes.

✍️ Local Manual Companion Notes

Add structured review notes beside any generated transcript without external AI or network calls:

• Run Soundings: Add manual enrichment on any active Soundings transcript note.
• Record local summaries, decisions, action items, and follow-ups.
• Preview the exact destination ( - Enrichment.md) and rendered Markdown before writing.
• 100% local, create-only, and non-mutating.

🛠️ Improved WebVTT Support & Discovery

We've overhauled the WebVTT parser to be more robust and an even better "first-pass" filter:

• Greater Compatibility: The parser handles a wider variety of WebVTT formats, including optional hour timestamps (mm:ss.ttt), unclosed voice tags, and multiple voice lines per cue.
• Smarter Classification: Parsing now happens during the discovery phase. The Review plan accurately tells you if a file is Eligible (ready for conversion), Unsupported (unsupported cue syntax), or Unreadable (malformed) before you start the process.

🔒 Safety & Performance

• Zero changes to original transcripts (preserved byte-for-byte).
• Zero overwriting of existing notes.
• Sub-millisecond scan overhead across large vaults.

Soundings 0.1.3

Implement robust enrichment evidence and draft validation for manual enrichment notes.

Soundings 0.1.2

Corrective follow-up for Obsidian Community source review. Removes the unnecessary saved-settings type assertion, uses the scanner-required window timer for cooperative yielding, preserves desktop-only local create-only behavior, and adds local regression guards for both findings.

Soundings 0.1.1

Corrective patch for Obsidian Community review feedback. Uses host-provided browser capabilities, respects custom Obsidian configuration directories, adopts searchable declarative settings, preserves the desktop-only safety boundary, and retains create-only transcript conversion.

Soundings 0.1.0

Initial desktop-only Community release. Soundings converts reviewed .txt and .vtt transcripts into create-only Markdown notes beside their preserved sources. Requires Obsidian desktop 1.13.7 or later.