apps/desktop/src/utils/createPresets.ts (1)

31-31: Preset configs should exclude clip data at the type level.

Instead of mutating config.clips to undefined, omit it in the type to prevent accidental inclusion and avoid runtime mutation.

Apply:

-export type CreatePreset = {
-  name: string;
-  config: Omit<ProjectConfiguration, "timeline">;
-  default: boolean;
-};
+export type CreatePreset = {
+  name: string;
+  config: Omit<ProjectConfiguration, "timeline" | "clips">;
+  default: boolean;
+};
...
-      // @ts-expect-error we reeeally don't want the timeline in the preset
-      config.timeline = undefined;
-      config.clips = undefined;
+      // @ts-expect-error we reeeally don't want the timeline in the preset
+      config.timeline = undefined;

apps/desktop/src/routes/editor/Timeline/ClipTrack.tsx (1)

358-374: Confirm UX: always opening clip configuration on mouseup.

This removes the previous “single-segment no-open” behavior. If intentional, LGTM; if not, gate on segments().length > 1 as before.

- setEditorState("timeline", "selection", {
-   type: "clip",
-   index: i(),
- });
+// If desired to keep prior UX:
+// if (segments().length > 1) {
+setEditorState("timeline", "selection", { type: "clip", index: i() });
+// }

crates/rendering/src/lib.rs (1)

139-149: Guard camera time against negatives.

If offsets.camera can be negative, request a frame at < 0 may return None or waste work. Safer to clamp to 0.

- .map(|d| d.get_frame(segment_time + offsets.camera)))
+ .map(|d| {
+     let camera_time = (segment_time + offsets.camera).max(0.0);
+     d.get_frame(camera_time)
+ }))

crates/editor/src/playback.rs (1)

253-258: Remove now-unused prev_audio_config.

After always calling set_playhead, prev_audio_config is written but never read.

-        let mut prev_audio_config = project.borrow().audio.clone();
...
-                audio_renderer.set_playhead(
+                audio_renderer.set_playhead(
                     playhead + elapsed as f64 / output_info.sample_rate as f64,
                     &project,
                 );
-                prev_audio_config = project.audio.clone();

apps/desktop/src/routes/editor/ConfigSidebar.tsx (1)

2426-2479: Keep NumberField state in sync with props and avoid wraparound on empty/NaN input.

• Local value doesn’t react if props.value changes (switching clips). Add a sync effect.
• Use an explicit empty/NaN check on blur; !rawValue() also triggers for 0.

Apply this diff:

 const rawValue = () => Math.round((props.value ?? 0) * 1000);
 
 const [value, setValue] = createSignal(rawValue().toString());
+createEffect(() => {
+  setValue(rawValue().toString());
+});
@@
-            onBlur={() => {
-              if (!rawValue() || value() === "" || Number.isNaN(rawValue())) {
-                setValue("0");
-                props.onChange(0);
-              }
-            }}
+            onBlur={() => {
+              const v = value();
+              const n = Number.parseInt(v, 10);
+              if (Number.isNaN(n) || v.trim() === "") {
+                setValue("0");
+                props.onChange(0);
+              }
+            }}

crates/audio/src/renderer.rs (1)

24-30: Consider bounding samples by the maximum positive per-track offset.

Tracks delayed by a positive track.offset reduce the readable window from the shared offset. Without accounting for this, the mixer may emit trailing silence where some tracks haven’t reached valid data yet.

You can precompute max_right_shift = tracks.iter().map(|t| t.offset.max(0)).max().unwrap_or(0) as usize and cap samples by (frames_per_track - offset - max_right_shift) to better reflect the tightest window.

crates/project/src/configuration.rs (3)

600-606: Make offsets tolerant to omission

If a clip entry is created with only an index, deserialization will currently fail because offsets is required. Default it to zeroed offsets.

 #[derive(Type, Serialize, Deserialize, Clone, Debug, Default)]
 #[serde(rename_all = "camelCase")]
 pub struct ClipConfiguration {
     pub index: u32,
-    pub offsets: ClipOffsets,
+    #[serde(default)]
+    pub offsets: ClipOffsets,
 }

---

603-604: Consider a stable clip identifier instead of positional index

Indices can shift on reorder/delete; a stable id (e.g., UUID or recording segment id) avoids misapplying offsets after edits.

---

590-598: Clarify units/sign and harden (alias + struct-level serde default) for ClipOffsets

• Add short field docs (seconds; positive = delay, negative = advance), move per-field serde defaults to a struct-level #[serde(default)], and accept camelCase input for system_audio via #[serde(alias = "systemAudio")]. Repo search shows snake_case "system_audio" is used widely, so alias is backward-compatible.

-#[derive(Type, Serialize, Deserialize, Clone, Copy, Debug, Default)]
-pub struct ClipOffsets {
-    #[serde(default)]
-    pub camera: f32,
-    #[serde(default)]
-    pub mic: f32,
-    #[serde(default)]
-    pub system_audio: f32,
-}
+#[derive(Type, Serialize, Deserialize, Clone, Copy, Debug, Default)]
+#[serde(default)]
+pub struct ClipOffsets {
+    /// Offset in seconds. Positive delays the source relative to screen; negative advances it.
+    pub camera: f32,
+    /// Offset in seconds. Positive delays the source relative to screen; negative advances it.
+    pub mic: f32,
+    /// Offset in seconds. Positive delays the source relative to screen; negative advances it.
+    #[serde(alias = "systemAudio")]
+    pub system_audio: f32,
+}

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

Learnt from: CR
PR: CapSoftware/Cap#0
File: AGENTS.md:0-0
Timestamp: 2025-09-08T16:48:20.727Z
Learning: Applies to **/tauri.ts : Do not edit auto-generated file: tauri.ts

crates/audio/src/renderer.rs (1)

36-41: Fix negative-offset wrapping and deref bugs (silence due to wrap; compile risk on &f32 ops).

• wrapping_add_signed on usize underflows for negative offsets, producing huge indices and silencing tracks.
• samples().get(...) returns &f32; multiplying references without deref is invalid; use pattern Some(&x) or *ref.

Apply:

-        for track in tracks {
-            let i = i.wrapping_add_signed(track.offset);
-
-            let data = track.data;
-            let gain = gain_for_db(track.gain);
+        for track in tracks {
+            // translate frame index by per-track offset in signed space
+            let ti = i as isize + track.offset;
+            if ti < 0 {
+                continue; // not yet reached this track’s first sample
+            }
+            let ti = ti as usize;
+
+            let data = track.data;
+            let gain = gain_for_db(track.gain);
@@
-            if data.channels() == 1 {
-                if let Some(sample) = data.samples().get(offset + i) {
-                    left += sample * 0.707 * gain;
-                    right += sample * 0.707 * gain;
-                }
+            if data.channels() == 1 {
+                if let Some(&sample) = data.samples().get(offset + ti) {
+                    left += sample * 0.707 * gain;
+                    right += sample * 0.707 * gain;
+                }
             } else if data.channels() == 2 {
-                let base_idx = offset * 2 + i * 2;
-                let Some(l_sample) = data.samples().get(base_idx) else {
+                let base_idx = (offset + ti) * 2;
+                let Some(&l_sample) = data.samples().get(base_idx) else {
                     continue;
                 };
-                let Some(r_sample) = data.samples().get(base_idx + 1) else {
+                let Some(&r_sample) = data.samples().get(base_idx + 1) else {
                     continue;
                 };

Also applies to: 46-58, 60-73

crates/audio/src/renderer.rs (3)

24-30: Clamp output length using per-track positive offsets.

Current cap uses only frames - offset, ignoring positive per-track offsets that shorten usable frames. Incorporate max(0, track.offset) (in frames) so we don’t overrun logical track ends and waste cycles mixing silence.

-    let samples = samples.min(
-        tracks
-            .iter()
-            .flat_map(|t| (t.data.samples().len() / t.data.channels() as usize).checked_sub(offset))
-            .max()
-            .unwrap_or(0),
-    );
+    let samples = samples.min(
+        tracks
+            .iter()
+            .flat_map(|t| {
+                let frames = t.data.samples().len() / t.data.channels() as usize;
+                let poff = if t.offset > 0 { t.offset as usize } else { 0 };
+                frames.checked_sub(offset + poff)
+            })
+            .max()
+            .unwrap_or(0),
+    );

---

9-14: Document offset units and sign semantics.

Add a short docstring to prevent unit drift (seconds vs samples) elsewhere.

 pub struct AudioRendererTrack<'a> {
     pub data: &'a AudioData,
     pub gain: f32,
     pub stereo_mode: StereoMode,
-    pub offset: isize,
+    /// Offset in sample frames. Positive = delay start; negative = start earlier (silence padded).
+    pub offset: isize,
 }

---

39-41: Micro-opt: avoid recomputing linear gain per sample.

gain_for_db(track.gain) is invariant across i; precompute once per track before the inner loop or store linear gain in AudioRendererTrack.

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro