feat(history): add quick copy and hide silent taps - #85

crates/lumen-store/src/lib.rs (1)

1050-1067: 🗄️ Data Integrity & Integration | 🔵 Trivial | ⚡ Quick win

Corrupted metrics JSON is silently discarded.

map_session_history uses .ok() to drop the serde_json::from_str error for metrics_json, so a corrupted or unreadable metrics blob silently produces initial_attempt: None with no trace. This differs from map_attempt (line 1075), which surfaces a rusqlite::Error::FromSqlConversionFailure for the same kind of malformed JSON. The fallback path is safe (the session stays visible rather than being wrongly hidden), but a corrupted record passes unnoticed, which makes diagnosing bad data harder.

Add a tracing::warn! on the parse failure before falling back to None, so genuinely corrupted rows are observable.

🩹 Proposed fix to log a parse failure

     let initial_attempt = metrics_json
-        .and_then(|json| serde_json::from_str::<PipelineMetrics>(&json).ok())
+        .and_then(|json| {
+            serde_json::from_str::<PipelineMetrics>(&json)
+                .inspect_err(|error| {
+                    tracing::warn!(%error, "failed to parse initial-attempt pipeline metrics")
+                })
+                .ok()
+        })
         .map(|pipeline_metrics| SessionHistoryAttempt {

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/lumen-store/src/lib.rs` around lines 1050 - 1067, Update
map_session_history so serde_json::from_str failures for metrics_json emit a
tracing::warn! containing the parse error before falling back to
initial_attempt: None. Preserve the existing safe fallback and successful
parsing behavior, and do not change the row-mapping error contract.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

crates/lumen-store/src/lib.rs (1)

20-20: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Share the 2-second threshold with the v14 migration.

HIDDEN_SILENT_CAPTURE_MAX_MS defines the runtime hide threshold. The v14 backfill in crates/lumen-store/src/schema.rs at line 564 repeats the same value as the literal 2_000. If one value changes, migrated rows and newly written rows classify differently for the same evidence. Make schema.rs use this constant, or move the constant into schema and import it here.

♻️ Proposed change in crates/lumen-store/src/schema.rs

// crates/lumen-store/src/schema.rs
if metrics.audio_duration_ms >= super::HIDDEN_SILENT_CAPTURE_MAX_MS {
    continue;
}

Make HIDDEN_SILENT_CAPTURE_MAX_MS pub(crate) in crates/lumen-store/src/lib.rs.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/lumen-store/src/lib.rs` at line 20, Share the hidden silent-capture
threshold between runtime classification and the v14 migration: make
HIDDEN_SILENT_CAPTURE_MAX_MS accessible within the crate, then update the v14
backfill logic in schema.rs to compare against that constant instead of the
duplicated 2_000 literal.

crates/lumen-store/src/schema.rs (1)

507-581: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick win

Wrap the v14 backfill in one transaction.

The backfill runs one autocommitted UPDATE per hidden session. On a database with many silent captures, this produces one commit (and one WAL fsync) per row, so migration time grows with the number of hidden rows. A single transaction around the whole loop also makes the backfill atomic with the version marker at line 587, which removes the partial-backfill state after an interrupted upgrade.

The loop itself terminates correctly: the keyset cursor sessions.id > ?2 advances strictly, and the predicate does not depend on history_visible.

♻️ Proposed change: single transaction for the backfill and the version marker

     if !has_v14 {
+        let backfill = conn.unchecked_transaction()?;
         let mut after_session_id = String::new();
         loop {
             let candidates = {
-                let mut statement = conn.prepare(
+                let mut statement = backfill.prepare(
                     r#"

Apply the same substitution to the UPDATE sessions SET history_visible=0 call, then commit the transaction after the loop.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/lumen-store/src/schema.rs` around lines 507 - 581, Wrap the entire
!has_v14 backfill in a single transaction, beginning before the pagination loop
and keeping all candidate queries and history_visible updates within it. Update
the existing UPDATE sessions call to execute through the transaction, then
commit only after the loop completes and immediately before the v14 version
marker is written, preserving atomicity between the backfill and marker.

apps/desktop/src-tauri/src/dictation.rs (1)

362-379: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Persist a consistent message form for both capture issues.

record_capture_signal_issue stores the localized ABSOLUTE_SILENCE_MESSAGE for AbsoluteSilence, but stores the machine token INVALID_CAPTURE_ISSUE for InvalidSignal. CaptureSignalIssue::message() returns INVALID_CAPTURE_MESSAGE for the same variant, so the persisted message and the user-facing message diverge for one variant only. Choose one convention for PipelineStageIssue::message: either both localized, or both stable machine tokens with the localized text kept in the returned error.

♻️ One option: store stable tokens for both kinds

 fn record_capture_signal_issue(attempt: &mut DictationAttemptRecord, issue: CaptureSignalIssue) {
     let (kind, message) = match issue {
         CaptureSignalIssue::AbsoluteSilence => {
-            (PipelineIssueKind::AbsoluteSilence, ABSOLUTE_SILENCE_MESSAGE)
+            (PipelineIssueKind::AbsoluteSilence, ABSOLUTE_SILENCE_ISSUE)
         }
         CaptureSignalIssue::InvalidSignal => {
             (PipelineIssueKind::InputUnavailable, INVALID_CAPTURE_ISSUE)
         }
     };

Note: the test at lines 2142-2151 and the PipelineIssueKind-based filtering in crates/lumen-store/src/schema.rs would need the same token update.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/desktop/src-tauri/src/dictation.rs` around lines 362 - 379, Make
record_capture_signal_issue persist a consistent message convention for both
CaptureSignalIssue variants: use stable machine tokens for AbsoluteSilence and
InvalidSignal, or localized messages for both. Keep
CaptureSignalIssue::message() aligned with the chosen convention, and update the
related test and PipelineIssueKind-based filtering in schema.rs if changing the
stored token.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes