chore: bump version 0.11.0 → 0.11.1

dogfood semantic cleanup (PR #150) lands: document-centric fetch_span +
assets.workspace_path 'last-registered' semantic explicitly documented.

patch bump 사유: 외부 wire / CLI / config surface 변경 없음. 새 internal
trait method (get_asset) + caller refactor + doc-comment 갱신. twin file
의 fetch_span 잘못 분기 가능성 fix (rare).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Cargo.lock

@@ -4127,7 +4127,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-app"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "base64 0.22.1",
@@ -4172,7 +4172,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-chunk"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "blake3",
@@ -4187,7 +4187,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-cli"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "clap",
@@ -4208,7 +4208,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-config"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "dirs 5.0.1",
@@ -4223,7 +4223,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-core"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "blake3",
@@ -4237,7 +4237,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-embed"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "blake3",
@@ -4251,7 +4251,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-embed-local"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "fastembed",
@@ -4264,7 +4264,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-eval"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "kebab-app",
@@ -4283,7 +4283,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-llm"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "kebab-core",
@@ -4292,7 +4292,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-llm-local"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "kebab-config",
@@ -4309,7 +4309,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-mcp"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "kebab-app",
@@ -4327,7 +4327,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-normalize"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "kebab-core",
@@ -4342,7 +4342,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-parse-code"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "gix",
@@ -4360,7 +4360,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-parse-image"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "ab_glyph",
  "anyhow",
@@ -4384,7 +4384,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-parse-md"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "kebab-core",
@@ -4401,7 +4401,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-parse-pdf"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "blake3",
@@ -4414,7 +4414,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-parse-types"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "kebab-core",
  "serde",
@@ -4422,7 +4422,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-rag"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "blake3",
@@ -4443,7 +4443,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-search"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "globset",
@@ -4462,7 +4462,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-source-fs"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "blake3",
@@ -4481,7 +4481,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-store-sqlite"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "blake3",
@@ -4502,7 +4502,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-store-vector"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "arrow",
@@ -4526,7 +4526,7 @@ dependencies = [
 
 [[package]]
 name = "kebab-tui"
-version = "0.8.3"
+version = "0.11.1"
 dependencies = [
  "anyhow",
  "crossterm",

Cargo.toml

@@ -31,7 +31,7 @@ edition       = "2024"
 rust-version  = "1.85"
 license       = "MIT OR Apache-2.0"
 repository    = "https://github.com/altair823/kebab"
-version       = "0.8.3"
+version       = "0.11.1"
 
 [workspace.dependencies]
 anyhow       = "1"

README.md

@@ -34,7 +34,7 @@ cargo install --git https://gitea.altair823.xyz/altair823-org/kebab.git --bin ke
 
 업데이트는 `git pull && cargo install --path crates/kebab-cli --locked --force` 또는 git URL 형식의 경우 `cargo install --git ... --force`.
 
-제거는 `cargo uninstall kebab-cli`. 이 명령은 binary 만 지우고 워크스페이스 데이터는 그대로 남는다. 데이터까지 정리하려면 `kebab reset --all --yes` (config + data + cache + state 4 개 XDG 경로 모두 wipe — **irreversible**, 재시작 시 `kebab init` 다시 실행). 부분 wipe 는 `kebab reset --data-only` (config 보존), `kebab reset --vector-only` (Lance + `embedding_records` 만, 다음 ingest 가 re-embed) 등.
+제거는 `cargo uninstall kebab-cli`. 이 명령은 binary 만 지우고 워크스페이스 데이터는 그대로 남는다. 데이터까지 정리하려면 `kebab reset --all --yes` (config + data + cache + state 4 개 XDG 경로 모두 wipe — **irreversible**, 재시작 시 `kebab init` 다시 실행). 부분 wipe 는 `kebab reset --data-only` (config 보존), `kebab reset --vector-only` (Lance + `embedding_records` 만, 다음 ingest 가 re-embed), **`kebab reset --orphans-only`** (현재 walker scope 밖에 있는 stored doc 만 정리 — `config.workspace.include` 좁히거나 sub-dir 옮긴 후 explicit reconcile; fs 의 file 은 건드리지 않음) 등.
 
 ## Quick start
 

crates/kebab-app/src/fetch.rs

@@ -189,10 +189,12 @@ fn fetch_span(
     // (markdown / note / paper / reference / inbox) is the *user-facing*
     // category, not the rendering format — the actual byte-level format
     // lives on the source `RawAsset.media_type`. Look it up via
-    // workspace_path (unique key per asset).
-    if let Some(asset) = <kebab_store_sqlite::SqliteStore as DocumentStore>::get_asset_by_workspace_path(
+    // doc.source_asset_id (PRIMARY KEY) so twin files (identical content
+    // at different paths) always read *this* document's own asset row,
+    // not whichever twin last wrote `assets.workspace_path`.
+    if let Some(asset) = <kebab_store_sqlite::SqliteStore as DocumentStore>::get_asset(
         &app.sqlite,
-        &doc.workspace_path,
+        &doc.source_asset_id,
     )? {
         if matches!(
             asset.media_type,

crates/kebab-app/src/lib.rs

@@ -71,7 +71,7 @@ mod staleness;
 
 pub use app::{App, SearchResponse};
 pub use ingest_progress::{AggregateCounts, IngestEvent, render_skipped_breakdown};
-pub use reset::{ResetReport, ResetScope};
+pub use reset::{ResetReport, ResetScope, enumerate_orphans};
 pub use error_wire::{ERROR_V1_ID, ErrorV1, StructuredError, classify};
 pub use fetch::fetch_with_config;
 #[doc(hidden)]
@@ -375,6 +375,28 @@ pub fn ingest_with_config_opts(
         .map(|d| d.doc_id.0)
         .collect();
 
+    // Dogfood: post-walker sweep to remove stored docs whose source
+    // file has been deleted from the filesystem. Must run BEFORE the
+    // per-asset loop so the loop's New/Updated labelling is based on
+    // the post-purge store state (the purged doc_ids won't be in
+    // `existing_doc_ids` above — they were already removed, OR the
+    // sweep here removes them before we start counting).
+    //
+    // Critical design invariant: only purge when the file is TRULY
+    // absent from disk. A file that is still on disk but outside the
+    // current walker scope (config narrowing / include-glob change) is
+    // NOT purged — we leave it in place to protect against accidental
+    // data loss via config edits.
+    let scanned_paths: std::collections::HashSet<kebab_core::WorkspacePath> = assets
+        .iter()
+        .map(|a| a.workspace_path.clone())
+        .collect();
+    let purged_deleted_files = sweep_deleted_files(
+        &app,
+        &scanned_paths,
+        vector_store.as_ref().map(|v| v.as_ref()),
+    )?;
+
     let started_at = time::OffsetDateTime::now_utc();
 
     let mut items: Vec<kebab_core::IngestItem> = Vec::new();
@@ -647,11 +669,11 @@ pub fn ingest_with_config_opts(
     crate::ingest_progress::emit(progress, terminal_event);
 
     // p9-fb-19: bump the persistent corpus_revision counter when a
-    // commit landed (any new / updated). This invalidates every
+    // commit landed (any new / updated / purged). This invalidates every
     // entry in any in-process LRU search cache (in this process or
     // a sibling) on the next lookup. No-op when nothing changed
     // (skipped-only run) — the cache stays valid.
-    if new_count > 0 || updated_count > 0 {
+    if new_count > 0 || updated_count > 0 || purged_deleted_files > 0 {
         match app.sqlite.bump_corpus_revision() {
             Ok(rev) => tracing::debug!(
                 target: "kebab-app",
@@ -682,6 +704,7 @@ pub fn ingest_with_config_opts(
         skipped_generated: fs_skips.skipped_generated,
         skipped_size_exceeded: fs_skips.skipped_size_exceeded,
         skip_examples: fs_skips.skip_examples,
+        purged_deleted_files,
         items: if summary_only { None } else { Some(items) },
     })
 }
@@ -1453,6 +1476,120 @@ fn purge_vector_orphans_for_workspace_path(
     Ok(())
 }
 
+/// Dogfood: post-walker sweep that purges stored documents whose source
+/// file has been physically deleted from the filesystem.
+///
+/// Algorithm:
+/// 1. Query `documents` for every `workspace_path` currently stored.
+/// 2. Compute `orphan_candidates = stored_paths - scanned_paths`.
+/// 3. For each candidate: resolve to an absolute path and call
+///    `Path::try_exists().unwrap_or(true)` — transient FS errors
+///    (EACCES, NFS hiccup, ownership change) conservatively count as
+///    "still present" so we never purge on uncertain signal. If the
+///    file still exists on disk it was merely out-of-scope this run
+///    (config narrowing / include-glob change) — leave it alone. Only
+///    files that are truly absent trigger a purge.
+/// 4. For absent files: call `purge_deleted_workspace_path` (SQLite
+///    cascade delete + optional copied-asset file removal) and, if a
+///    vector store is present, delete the associated vectors.
+///
+/// Returns the number of documents purged.
+///
+/// Non-fatal design: individual purge failures are logged and counted
+/// as errors on the per-file level but do NOT abort the sweep — a
+/// partial failure is preferable to blocking the rest of ingest. The
+/// return value only counts successful purges.
+fn sweep_deleted_files(
+    app: &App,
+    scanned_paths: &std::collections::HashSet<kebab_core::WorkspacePath>,
+    vector_store: Option<&kebab_store_vector::LanceVectorStore>,
+) -> anyhow::Result<u32> {
+    use kebab_core::DocumentStore as _;
+
+    let stored_paths = app
+        .sqlite
+        .all_workspace_paths()
+        .context("sweep_deleted_files: all_workspace_paths")?;
+
+    if stored_paths.is_empty() {
+        return Ok(0);
+    }
+
+    let workspace_root = app.config.resolve_workspace_root();
+    let mut purged: u32 = 0;
+
+    for stored_path in stored_paths {
+        if scanned_paths.contains(&stored_path) {
+            continue; // still in scope — skip
+        }
+
+        // Resolve to an absolute path and check existence on disk.
+        // Use `try_exists` + `unwrap_or(true)` so transient FS errors
+        // (EACCES on a path we lack read on, NFS hiccups, ownership
+        // change) are CONSERVATIVELY treated as "file still present" —
+        // never purge on uncertain signal (data-safety: PR #148 review).
+        // `exists()` would return false on Err and trigger a wrongful
+        // purge. Files whose path cannot be joined (theoretically
+        // impossible for non-empty workspace_path strings, but
+        // defense-in-depth) are likewise treated as still present.
+        let abs = workspace_root.join(&stored_path.0);
+        if abs.try_exists().unwrap_or(true) {
+            // File is on disk but not in this scan's scope (config
+            // narrowing). DO NOT purge — critical design constraint.
+            tracing::debug!(
+                target: "kebab-app",
+                path = %stored_path.0,
+                "sweep_deleted_files: file on disk but out of scope — leaving in store"
+            );
+            continue;
+        }
+
+        // File is truly absent → purge.
+        let chunk_ids = match kebab_store_sqlite::purge_deleted_workspace_path(
+            &app.sqlite,
+            &stored_path,
+        ) {
+            Ok(ids) => ids,
+            Err(e) => {
+                tracing::warn!(
+                    target: "kebab-app",
+                    path = %stored_path.0,
+                    error = %e,
+                    "sweep_deleted_files: purge failed; skipping this path"
+                );
+                continue;
+            }
+        };
+
+        // Purge associated vectors (best-effort; partial failure
+        // acceptable — orphan vectors get cleaned by `kebab reset
+        // --vector-only` if they accumulate).
+        if let Some(vec) = vector_store {
+            if !chunk_ids.is_empty() {
+                use kebab_core::VectorStore as _;
+                if let Err(e) = vec.delete_by_chunk_ids(&chunk_ids) {
+                    tracing::warn!(
+                        target: "kebab-app",
+                        path = %stored_path.0,
+                        count = chunk_ids.len(),
+                        error = %e,
+                        "sweep_deleted_files: vector delete failed; SQLite side already cleaned"
+                    );
+                }
+            }
+        }
+
+        tracing::info!(
+            target: "kebab-app",
+            path = %stored_path.0,
+            "sweep_deleted_files: purged document for deleted file"
+        );
+        purged = purged.saturating_add(1);
+    }
+
+    Ok(purged)
+}
+
 /// P7-3: process one `MediaType::Pdf` asset end-to-end.
 ///
 /// - Reads bytes from disk.

crates/kebab-app/src/reset.rs

@@ -9,13 +9,19 @@
 //!
 //! `--vector-only` additionally truncates `embedding_records` in SQLite
 //! so the next `kebab ingest` re-embeds cleanly without orphan rows.
+//!
+//! `--orphans-only` purges stored docs that are outside the current walker
+//! scope (config narrowing / removed sub-directory). No filesystem paths are
+//! removed — this is purely a store-level reconciliation.
 
+use std::collections::HashSet;
 use std::path::PathBuf;
 
 use anyhow::{Context, Result};
 use serde::{Deserialize, Serialize};
 
 use kebab_config::{Config, expand_path};
+use kebab_core::WorkspacePath;
 
 /// What the user asked to remove. Mutually exclusive — picked by the CLI
 /// from a clap `ArgGroup`.
@@ -32,6 +38,13 @@ pub enum ResetScope {
     VectorOnly,
     /// Wipe only the config dir.
     ConfigOnly,
+    /// Purge stored docs that are outside the current walker scope (no
+    /// filesystem paths are removed). Filesystem existence is NOT checked —
+    /// anything the current walker would not visit is considered an orphan.
+    /// The explicit complement to the conservative `sweep_deleted_files`
+    /// that runs during ingest (which leaves on-disk-but-out-of-scope docs
+    /// alone for data safety).
+    OrphansOnly,
 }
 
 /// Result of a successful wipe — emitted as `reset_report.v1` by the
@@ -41,6 +54,16 @@ pub struct ResetReport {
     pub scope: ResetScope,
     pub removed_paths: Vec<PathBuf>,
     pub embedding_rows_truncated: u64,
+    /// Number of stored docs purged because they are outside the current
+    /// walker scope. Non-zero only when `scope == OrphansOnly`.
+    /// `#[serde(default)]` preserves back-compat with older callers that
+    /// do not include this field.
+    #[serde(default)]
+    pub orphans_purged: u32,
+    /// Paths of the orphaned docs that were purged. Sorted for deterministic
+    /// output. Non-empty only when `scope == OrphansOnly`.
+    #[serde(default)]
+    pub purged_paths: Vec<WorkspacePath>,
 }
 
 /// Compute the absolute on-disk paths a given scope will wipe, given a
@@ -67,6 +90,10 @@ pub fn enumerate_paths(scope: ResetScope, cfg: &Config) -> Vec<PathBuf> {
             vec![vector_dir]
         }
         ResetScope::ConfigOnly => vec![cfg_dir],
+        // OrphansOnly operates purely at the store level — no filesystem paths
+        // are removed. Return empty so `estimate_size_bytes` stays zero and
+        // the existing confirm UI path for directory wipes is skipped.
+        ResetScope::OrphansOnly => vec![],
     }
 }
 
@@ -96,16 +123,82 @@ pub fn estimate_size_bytes(paths: &[PathBuf]) -> u64 {
     paths.iter().map(|p| walk(p)).sum()
 }
 
+/// Compute the workspace paths stored in SQLite that are NOT visited by
+/// the current walker scope (i.e. they are "orphans" — on disk but
+/// outside the configured include/exclude rules, or from a sub-directory
+/// that has since been removed from the workspace).
+///
+/// Does NOT check filesystem existence — `OrphansOnly` is the explicit
+/// "I know what I'm doing" variant; callers that want the conservative
+/// fs-aware sweep should use `sweep_deleted_files` inside ingest.
+///
+/// Returns the list sorted for deterministic output. Called twice by the
+/// CLI path (once for the confirm UI preview, once inside `execute`);
+/// the double scan is acceptable for a rare destructive operation.
+pub fn enumerate_orphans(cfg: &Config) -> Result<Vec<WorkspacePath>> {
+    use kebab_core::DocumentStore as _;
+    use kebab_source_fs::FsSourceConnector;
+    use kebab_core::SourceScope;
+
+    let store = kebab_store_sqlite::SqliteStore::open(cfg)
+        .context("enumerate_orphans: open SqliteStore")?;
+
+    let stored = store
+        .all_workspace_paths()
+        .context("enumerate_orphans: all_workspace_paths")?;
+
+    if stored.is_empty() {
+        return Ok(Vec::new());
+    }
+
+    // Build the same SourceScope the CLI's ingest path uses: root from
+    // config, exclude list from config, no include override (full scope).
+    let root = cfg.resolve_workspace_root();
+    let scope = SourceScope {
+        root: root.clone(),
+        exclude: cfg.workspace.exclude.clone(),
+        ..Default::default()
+    };
+
+    let connector = FsSourceConnector::new(cfg)
+        .context("enumerate_orphans: build FsSourceConnector")?;
+    let (assets, _skips) = connector
+        .scan_with_skips(&scope)
+        .context("enumerate_orphans: scan workspace")?;
+
+    let scanned: HashSet<WorkspacePath> = assets
+        .into_iter()
+        .map(|a| a.workspace_path)
+        .collect();
+
+    let mut orphans: Vec<WorkspacePath> = stored
+        .into_iter()
+        .filter(|p| !scanned.contains(p))
+        .collect();
+    orphans.sort_by(|a, b| a.0.cmp(&b.0));
+    Ok(orphans)
+}
+
 /// Wipe every path from `enumerate_paths(scope, cfg)`. For
 /// `ResetScope::VectorOnly`, also truncates the SQLite
 /// `embedding_records` table so the store doesn't point at the Lance
 /// rows we just removed off-disk.
 ///
+/// For `ResetScope::OrphansOnly`, no filesystem directories are removed.
+/// Instead the store is reconciled: stored docs outside the current walker
+/// scope are purged from SQLite (+ vector store when configured). The
+/// caller is expected to have already shown the confirm UI using
+/// `enumerate_orphans`.
+///
 /// Idempotent: a missing path is treated as already-removed (success).
 /// Returns a `ResetReport` listing exactly what was removed (paths that
 /// existed before the call) so `--json` callers see the truth, not the
 /// request.
 pub fn execute(scope: ResetScope, cfg: &Config) -> Result<ResetReport> {
+    if matches!(scope, ResetScope::OrphansOnly) {
+        return execute_orphans_only(cfg);
+    }
+
     let paths = enumerate_paths(scope, cfg);
     let mut removed = Vec::new();
 
@@ -128,9 +221,100 @@ pub fn execute(scope: ResetScope, cfg: &Config) -> Result<ResetReport> {
         scope,
         removed_paths: removed,
         embedding_rows_truncated,
+        orphans_purged: 0,
+        purged_paths: Vec::new(),
     })
 }
 
+/// Execute the `OrphansOnly` variant: reconcile stored docs against the
+/// current walker scope without touching any filesystem directory.
+fn execute_orphans_only(cfg: &Config) -> Result<ResetReport> {
+    let orphans = enumerate_orphans(cfg)
+        .context("execute_orphans_only: enumerate orphans")?;
+
+    if orphans.is_empty() {
+        return Ok(ResetReport {
+            scope: ResetScope::OrphansOnly,
+            removed_paths: Vec::new(),
+            embedding_rows_truncated: 0,
+            orphans_purged: 0,
+            purged_paths: Vec::new(),
+        });
+    }
+
+    let store = std::sync::Arc::new(
+        kebab_store_sqlite::SqliteStore::open(cfg)
+            .context("execute_orphans_only: open SqliteStore")?,
+    );
+
+    // Open vector store if configured. Mirror the same guard the ingest
+    // path uses: only construct when the provider is not "none" / dims > 0.
+    let vector_store: Option<kebab_store_vector::LanceVectorStore> =
+        open_vector_store_if_configured(cfg, store.clone())?;
+
+    let mut purged_paths: Vec<WorkspacePath> = Vec::new();
+
+    for path in &orphans {
+        let chunk_ids = kebab_store_sqlite::purge_deleted_workspace_path(&store, path)
+            .with_context(|| format!("execute_orphans_only: purge {}", path.0))?;
+
+        if let Some(ref vs) = vector_store {
+            if !chunk_ids.is_empty() {
+                use kebab_core::VectorStore as _;
+                if let Err(e) = vs.delete_by_chunk_ids(&chunk_ids) {
+                    tracing::warn!(
+                        target: "kebab-app",
+                        path = %path.0,
+                        count = chunk_ids.len(),
+                        error = %e,
+                        "reset --orphans-only: vector delete failed; SQLite side already cleaned"
+                    );
+                }
+            }
+        }
+
+        tracing::info!(
+            target: "kebab-app",
+            path = %path.0,
+            "reset --orphans-only: purged orphan document"
+        );
+        purged_paths.push(path.clone());
+    }
+
+    let orphans_purged = u32::try_from(purged_paths.len()).unwrap_or(u32::MAX);
+
+    Ok(ResetReport {
+        scope: ResetScope::OrphansOnly,
+        removed_paths: Vec::new(),
+        embedding_rows_truncated: 0,
+        orphans_purged,
+        purged_paths,
+    })
+}
+
+/// Open the Lance vector store if the configured embedding provider is
+/// active (non-"none", dimensions > 0). Returns `None` for lexical-only
+/// configs. Mirrors the guard in `App::vector`.
+fn open_vector_store_if_configured(
+    cfg: &Config,
+    store: std::sync::Arc<kebab_store_sqlite::SqliteStore>,
+) -> Result<Option<kebab_store_vector::LanceVectorStore>> {
+    if cfg.models.embedding.provider == "none" || cfg.models.embedding.dimensions == 0 {
+        return Ok(None);
+    }
+    match kebab_store_vector::LanceVectorStore::new(cfg, store) {
+        Ok(vs) => Ok(Some(vs)),
+        Err(e) => {
+            tracing::warn!(
+                target: "kebab-app",
+                error = %e,
+                "reset --orphans-only: could not open vector store; skipping vector delete"
+            );
+            Ok(None)
+        }
+    }
+}
+
 /// Open the SQLite store at the configured path and run
 /// `truncate_embedding_records`. Returns the count of truncated rows
 /// (the helper itself reports `DELETE` rowcount). If the SQLite file
@@ -200,4 +384,14 @@ mod tests {
         let bytes = estimate_size_bytes(&[dir.path().to_path_buf()]);
         assert_eq!(bytes, 5 + 6);
     }
+
+    #[test]
+    fn enumerate_orphans_only_returns_empty_paths() {
+        let cfg = Config::defaults();
+        let paths = enumerate_paths(ResetScope::OrphansOnly, &cfg);
+        assert!(
+            paths.is_empty(),
+            "OrphansOnly must return empty vec from enumerate_paths"
+        );
+    }
 }

crates/kebab-app/tests/file_deletion_auto_purge.rs

@@ -0,0 +1,178 @@
+//! Dogfood: auto-purge stored docs for filesystem-deleted files.
+//!
+//! Two tests:
+//!
+//! 1. `file_deletion_auto_purge` — ingest 2 files, delete one, re-ingest.
+//!    The re-ingest must report `purged_deleted_files = 1`, the deleted
+//!    file must no longer appear in `list_docs`, and lexical search for
+//!    its unique content must return no hits.
+//!
+//! 2. `include_scope_narrowing_does_not_purge` — ingest 2 files under a
+//!    wide glob, narrow the walker scope to only one file, re-ingest.
+//!    The narrowed ingest must NOT purge the out-of-scope file because
+//!    the file is still on disk (just excluded from this run). Protects
+//!    users against accidental data loss via config edits.
+
+mod common;
+
+use common::TestEnv;
+use kebab_app::ingest_with_config_opts;
+use kebab_app::IngestOpts;
+use kebab_core::{DocFilter, DocumentStore, SearchMode, SearchQuery, SourceScope};
+
+/// Helper: open the store via `TestEnv` and run `list_documents`.
+fn list_doc_paths(env: &TestEnv) -> Vec<String> {
+    use kebab_store_sqlite::SqliteStore;
+    let store = SqliteStore::open(&env.config).unwrap();
+    store.run_migrations().unwrap();
+    store
+        .list_documents(&DocFilter::default())
+        .unwrap()
+        .into_iter()
+        .map(|d| d.doc_path.0)
+        .collect()
+}
+
+#[test]
+fn file_deletion_auto_purge() {
+    let env = TestEnv::lexical_only();
+
+    // Write two .rs files into the workspace.
+    let a_path = env.workspace_root.join("a.rs");
+    let b_path = env.workspace_root.join("b.rs");
+    std::fs::write(&a_path, "// file a\nfn alpha() {}\n").unwrap();
+    std::fs::write(&b_path, "// file b\nfn bravo() {}\n").unwrap();
+
+    // First ingest — both must be New.
+    let first = ingest_with_config_opts(
+        env.config.clone(),
+        env.scope(),
+        false,
+        IngestOpts::default(),
+    )
+    .expect("first ingest must succeed");
+    // Only count the .rs files we added (there may be fixture files too).
+    let first_new = first.new;
+    assert!(first_new >= 2, "expected at least 2 new docs: {first:?}");
+    assert_eq!(
+        first.purged_deleted_files, 0,
+        "no purges on first ingest: {first:?}"
+    );
+    assert_eq!(first.errors, 0, "no errors on first ingest: {first:?}");
+
+    // Delete one file from the filesystem.
+    std::fs::remove_file(&b_path).expect("remove b.rs");
+
+    // Second ingest — scanned count drops by 1; b.rs should be purged.
+    let second = ingest_with_config_opts(
+        env.config.clone(),
+        env.scope(),
+        false,
+        IngestOpts::default(),
+    )
+    .expect("second ingest must succeed");
+
+    assert_eq!(
+        second.purged_deleted_files, 1,
+        "exactly 1 file should be purged: {second:?}"
+    );
+    assert_eq!(second.new, 0, "no new docs after deletion: {second:?}");
+    assert_eq!(second.updated, 0, "no updated docs: {second:?}");
+    assert_eq!(second.errors, 0, "no errors: {second:?}");
+
+    // b.rs must no longer appear in list_docs.
+    let doc_paths = list_doc_paths(&env);
+    let b_ws_path = "b.rs";
+    assert!(
+        !doc_paths.iter().any(|p| p == b_ws_path),
+        "b.rs must be gone from list_docs; got: {doc_paths:?}"
+    );
+    // a.rs must still be present.
+    let a_ws_path = "a.rs";
+    assert!(
+        doc_paths.iter().any(|p| p == a_ws_path),
+        "a.rs must still be in list_docs; got: {doc_paths:?}"
+    );
+
+    // Lexical search for b.rs's unique content returns no hits.
+    let app = env.app();
+    let query = SearchQuery {
+        text: "bravo".to_string(),
+        mode: SearchMode::Lexical,
+        k: 10,
+        filters: kebab_core::SearchFilters::default(),
+    };
+    let hits = app.search(query).expect("search must not error");
+    assert!(
+        hits.is_empty(),
+        "search for deleted file's content must return no hits; got: {hits:?}"
+    );
+}
+
+#[test]
+fn include_scope_narrowing_does_not_purge() {
+    let env = TestEnv::lexical_only();
+
+    // Write two .rs files.
+    let a_path = env.workspace_root.join("a_narrow.rs");
+    let b_path = env.workspace_root.join("b_narrow.rs");
+    std::fs::write(&a_path, "// narrow a\nfn alpha_narrow() {}\n").unwrap();
+    std::fs::write(&b_path, "// narrow b\nfn bravo_narrow() {}\n").unwrap();
+
+    // Wide scope: first ingest — both must be New.
+    let wide_scope = SourceScope {
+        root: env.workspace_root.clone(),
+        include: vec!["**/*.rs".to_string()],
+        exclude: env.config.workspace.exclude.clone(),
+    };
+    let first = ingest_with_config_opts(
+        env.config.clone(),
+        wide_scope,
+        false,
+        IngestOpts::default(),
+    )
+    .expect("first ingest (wide) must succeed");
+    assert!(
+        first.new >= 2,
+        "expected at least 2 new docs: {first:?}"
+    );
+    assert_eq!(
+        first.purged_deleted_files, 0,
+        "no purges on first ingest: {first:?}"
+    );
+
+    // Narrow scope: only a_narrow.rs in include — b_narrow.rs is still
+    // on disk but excluded from the walker scope.
+    let narrow_scope = SourceScope {
+        root: env.workspace_root.clone(),
+        include: vec!["a_narrow.rs".to_string()],
+        exclude: env.config.workspace.exclude.clone(),
+    };
+    let second = ingest_with_config_opts(
+        env.config.clone(),
+        narrow_scope,
+        false,
+        IngestOpts::default(),
+    )
+    .expect("second ingest (narrow) must succeed");
+
+    // CRITICAL: b_narrow.rs is still on disk — must NOT be purged.
+    assert_eq!(
+        second.purged_deleted_files, 0,
+        "scope-narrowing must NOT purge on-disk files; got: {second:?}"
+    );
+    assert_eq!(second.errors, 0, "no errors: {second:?}");
+
+    // b_narrow.rs must still exist in the store.
+    let doc_paths = list_doc_paths(&env);
+    let b_ws_path = "b_narrow.rs";
+    assert!(
+        doc_paths.iter().any(|p| p == b_ws_path),
+        "b_narrow.rs must still be in list_docs after scope narrowing; got: {doc_paths:?}"
+    );
+    // And the file must still be on disk.
+    assert!(
+        b_path.exists(),
+        "b_narrow.rs must still be on disk (we didn't delete it)"
+    );
+}

crates/kebab-app/tests/reset_orphans.rs

@@ -0,0 +1,141 @@
+//! Integration test for `kebab reset --orphans-only`.
+//!
+//! Verifies that stored docs outside the current walker scope are purged
+//! from the store without removing any files from the filesystem.
+//!
+//! Test outline:
+//! 1. Ingest 3 .rs files (a.rs, b.rs, c.rs) — all New.
+//! 2. Narrow the config `include` to `["a.rs"]` only; b.rs and c.rs are
+//!    still on disk but outside the walker scope.
+//! 3. Run `execute(ResetScope::OrphansOnly, &cfg)` — report must show
+//!    `orphans_purged == 2` and `purged_paths` contains b.rs + c.rs.
+//! 4. `list docs` must show only a.rs.
+//! 5. b.rs and c.rs must still exist on disk (no filesystem removal).
+//! 6. Second reset → `orphans_purged == 0` (idempotent).
+
+mod common;
+
+use common::TestEnv;
+use kebab_app::IngestOpts;
+use kebab_app::reset::{ResetScope, execute};
+use kebab_core::{DocFilter, DocumentStore, SourceScope};
+
+/// Open the SqliteStore and list all `workspace_path` values.
+fn list_doc_paths(env: &TestEnv) -> Vec<String> {
+    use kebab_store_sqlite::SqliteStore;
+    let store = SqliteStore::open(&env.config).unwrap();
+    store.run_migrations().unwrap();
+    store
+        .list_documents(&DocFilter::default())
+        .unwrap()
+        .into_iter()
+        .map(|d| d.doc_path.0)
+        .collect()
+}
+
+#[test]
+fn reset_orphans_only_purges_out_of_scope_docs() {
+    let env = TestEnv::lexical_only();
+
+    // Write three .rs files into the workspace.
+    let a_path = env.workspace_root.join("a.rs");
+    let b_path = env.workspace_root.join("b.rs");
+    let c_path = env.workspace_root.join("c.rs");
+    std::fs::write(&a_path, "// file a\nfn alpha() {}\n").unwrap();
+    std::fs::write(&b_path, "// file b\nfn bravo() {}\n").unwrap();
+    std::fs::write(&c_path, "// file c\nfn charlie() {}\n").unwrap();
+
+    // Ingest all three with a wide scope.
+    let wide_scope = SourceScope {
+        root: env.workspace_root.clone(),
+        include: vec!["**/*.rs".to_string()],
+        exclude: env.config.workspace.exclude.clone(),
+    };
+    let first = kebab_app::ingest_with_config_opts(
+        env.config.clone(),
+        wide_scope,
+        false,
+        IngestOpts::default(),
+    )
+    .expect("first ingest must succeed");
+    // The fixture workspace may contain other .rs files — just assert we
+    // got at least 3 new docs (our a.rs, b.rs, c.rs).
+    assert!(first.new >= 3, "expected at least 3 new docs: {first:?}");
+    assert_eq!(first.errors, 0, "no errors on first ingest");
+
+    // Narrow config to include only a.rs; b.rs + c.rs are still on disk.
+    let mut narrow_cfg = env.config.clone();
+    narrow_cfg.workspace.exclude.clear();
+    // Re-point workspace root (already correct) and restrict include via
+    // the SourceScope in the connector. The config's `workspace.root` is
+    // used by `enumerate_orphans` to build its scope — we keep that
+    // pointing at the workspace root. We simulate narrowing by setting a
+    // glob that only matches a.rs.
+    //
+    // NOTE: `kebab_config::WorkspaceCfg` does not have an `include` field
+    // (it was removed in p9-fb-25). We narrow the scope via the walker
+    // exclude list: exclude b.rs and c.rs explicitly.
+    narrow_cfg.workspace.exclude = vec!["b.rs".to_string(), "c.rs".to_string()];
+
+    // Run orphans-only reset.
+    let report = execute(ResetScope::OrphansOnly, &narrow_cfg)
+        .expect("orphans-only reset must succeed");
+
+    assert_eq!(
+        report.orphans_purged, 2,
+        "expected 2 orphans purged (b.rs + c.rs): {report:?}"
+    );
+
+    let mut purged: Vec<String> = report
+        .purged_paths
+        .iter()
+        .map(|p| p.0.clone())
+        .collect();
+    purged.sort();
+    assert_eq!(
+        purged,
+        vec!["b.rs".to_string(), "c.rs".to_string()],
+        "purged_paths must list b.rs and c.rs in sorted order: {purged:?}"
+    );
+
+    // list docs must show only a.rs (and any pre-existing fixture files
+    // that are not excluded by the narrow config).
+    let doc_paths = list_doc_paths(&env);
+    // The narrow_cfg excludes b.rs + c.rs — they must no longer be in store.
+    assert!(
+        !doc_paths.iter().any(|p| p == "b.rs"),
+        "b.rs must be gone from store after orphans-only reset; got: {doc_paths:?}"
+    );
+    assert!(
+        !doc_paths.iter().any(|p| p == "c.rs"),
+        "c.rs must be gone from store after orphans-only reset; got: {doc_paths:?}"
+    );
+    assert!(
+        doc_paths.iter().any(|p| p == "a.rs"),
+        "a.rs must still be in store; got: {doc_paths:?}"
+    );
+
+    // Both b.rs and c.rs must still exist on the filesystem — no file
+    // removal is performed by orphans-only.
+    assert!(
+        b_path.exists(),
+        "b.rs must still be on disk after orphans-only reset"
+    );
+    assert!(
+        c_path.exists(),
+        "c.rs must still be on disk after orphans-only reset"
+    );
+
+    // Second reset must be idempotent: nothing left to purge.
+    let second = execute(ResetScope::OrphansOnly, &narrow_cfg)
+        .expect("second orphans-only reset must succeed");
+    assert_eq!(
+        second.orphans_purged, 0,
+        "second reset must be idempotent (orphans_purged == 0): {second:?}"
+    );
+    assert!(
+        second.purged_paths.is_empty(),
+        "second reset purged_paths must be empty: {:?}",
+        second.purged_paths
+    );
+}

crates/kebab-app/tests/twin_files_fetch_span.rs

@@ -0,0 +1,176 @@
+//! Regression test for the twin-file fetch_span media-type lookup bug.
+//!
+//! Twin files (identical content at different workspace paths) share one
+//! `assets` row whose PRIMARY KEY is the blake3 content hash. The old
+//! `fetch_span` implementation called
+//! `get_asset_by_workspace_path(&doc.workspace_path)` to check whether the
+//! media type was PDF/audio (and therefore reject span fetch). For a twin
+//! file that lookup could silently return the *other* twin's asset row if
+//! `assets.workspace_path` had been overwritten on the most recent ingest of
+//! the sibling — making the media-type branch decision incorrect.
+//!
+//! Fix: `fetch_span` now uses the 2-step lookup
+//!   `get_document_by_workspace_path` → `doc.source_asset_id` → `get_asset`
+//! so the result is always anchored to the requesting document, not
+//! whichever twin last updated `assets.workspace_path`.
+//!
+//! This test builds a twin-file scenario (two .md files at different paths
+//! with identical content), ingests both, then calls `fetch_span` on each
+//! twin's `doc_id` and asserts it succeeds. Before the fix, if the asset
+//! row's workspace_path happened to point at the wrong twin the span could
+//! return an incorrect `span_not_supported` for a non-PDF/audio file, or
+//! conversely allow span on a PDF twin by accident. After the fix, the
+//! lookup is always doc-specific.
+
+mod common;
+
+use common::TestEnv;
+use kebab_app::ingest_with_config;
+use kebab_core::{DocumentStore, FetchKind, FetchOpts, FetchQuery, IngestItemKind};
+
+#[test]
+fn twin_files_fetch_span_uses_correct_asset() {
+    let env = TestEnv::lexical_only();
+
+    // Write two markdown files with identical content at different paths.
+    let dir_a = env.workspace_root.join("src_a");
+    let dir_b = env.workspace_root.join("src_b");
+    std::fs::create_dir_all(&dir_a).unwrap();
+    std::fs::create_dir_all(&dir_b).unwrap();
+
+    // The content must produce at least 1 line so span fetch is non-trivial.
+    let content = "# Twin\n\nLine one.\n\nLine two.\n\nLine three.\n";
+    std::fs::write(dir_a.join("note.md"), content).unwrap();
+    std::fs::write(dir_b.join("note.md"), content).unwrap();
+
+    // Ingest all files (fixture workspace + our two new twins).
+    let report = ingest_with_config(env.config.clone(), env.scope(), false)
+        .expect("ingest must succeed");
+    assert_eq!(report.errors, 0, "no ingest errors; report={report:?}");
+
+    // Both twin paths must appear as New in the report.
+    let items = report.items.as_ref().expect("items must be present");
+    let twin_items: Vec<_> = items
+        .iter()
+        .filter(|i| {
+            i.doc_path.0.ends_with("src_a/note.md")
+                || i.doc_path.0.ends_with("src_b/note.md")
+        })
+        .collect();
+    assert_eq!(
+        twin_items.len(),
+        2,
+        "exactly 2 twin items expected; items={items:?}"
+    );
+    for item in &twin_items {
+        assert_eq!(
+            item.kind,
+            IngestItemKind::New,
+            "each twin must be New; item={item:?}"
+        );
+    }
+
+    // Resolve doc_ids for both workspace paths.
+    // The ingest layer normalises workspace_path to the path relative to
+    // workspace_root (e.g. "src_a/note.md"), so we look up by that form.
+    let store = kebab_store_sqlite::SqliteStore::open(&env.config).unwrap();
+    store.run_migrations().unwrap();
+
+    // Find the twin items by matching on suffix so the test is robust to
+    // however the workspace root is represented.
+    let items = report.items.as_ref().expect("items must be present");
+    let path_a_str = items
+        .iter()
+        .find(|i| i.doc_path.0.ends_with("src_a/note.md"))
+        .map(|i| i.doc_path.0.clone())
+        .expect("src_a/note.md must appear in ingest report");
+    let path_b_str = items
+        .iter()
+        .find(|i| i.doc_path.0.ends_with("src_b/note.md"))
+        .map(|i| i.doc_path.0.clone())
+        .expect("src_b/note.md must appear in ingest report");
+
+    let path_a = kebab_core::WorkspacePath(path_a_str);
+    let path_b = kebab_core::WorkspacePath(path_b_str);
+
+    let doc_a = store
+        .get_document_by_workspace_path(&path_a)
+        .expect("get_document_by_workspace_path path_a")
+        .expect("doc_a must exist after ingest");
+    let doc_b = store
+        .get_document_by_workspace_path(&path_b)
+        .expect("get_document_by_workspace_path path_b")
+        .expect("doc_b must exist after ingest");
+
+    // Both twins share one asset_id (same content hash).
+    assert_eq!(
+        doc_a.source_asset_id, doc_b.source_asset_id,
+        "twin files must share one asset_id"
+    );
+
+    // Open App and issue span fetch on each twin's doc_id.
+    let app = env.app();
+
+    let result_a = app
+        .fetch(
+            FetchQuery::Span {
+                doc_id: doc_a.doc_id.clone(),
+                line_start: 1,
+                line_end: 2,
+            },
+            FetchOpts::default(),
+        )
+        .expect("fetch_span on twin A must succeed for a markdown file");
+    assert_eq!(result_a.kind, FetchKind::Span);
+    assert!(
+        result_a.text.as_deref().is_some_and(|t| !t.is_empty()),
+        "span text for twin A must not be empty"
+    );
+
+    let result_b = app
+        .fetch(
+            FetchQuery::Span {
+                doc_id: doc_b.doc_id.clone(),
+                line_start: 1,
+                line_end: 2,
+            },
+            FetchOpts::default(),
+        )
+        .expect("fetch_span on twin B must succeed for a markdown file");
+    assert_eq!(result_b.kind, FetchKind::Span);
+    assert!(
+        result_b.text.as_deref().is_some_and(|t| !t.is_empty()),
+        "span text for twin B must not be empty"
+    );
+
+    // Ingest again to force the asset.workspace_path flip-flop, then
+    // re-check. Pre-fix this was the scenario that triggered the bug:
+    // after the second ingest the asset row's workspace_path could point
+    // at either twin, making one twin's span fetch behave incorrectly.
+    let report2 = ingest_with_config(env.config.clone(), env.scope(), false)
+        .expect("second ingest must succeed");
+    assert_eq!(report2.errors, 0, "no ingest errors on second run; report={report2:?}");
+
+    // Re-open app after second ingest and verify span still works on both.
+    let app2 = env.app();
+
+    app2.fetch(
+        FetchQuery::Span {
+            doc_id: doc_a.doc_id.clone(),
+            line_start: 1,
+            line_end: 3,
+        },
+        FetchOpts::default(),
+    )
+    .expect("fetch_span on twin A after flip-flop must still succeed");
+
+    app2.fetch(
+        FetchQuery::Span {
+            doc_id: doc_b.doc_id.clone(),
+            line_start: 1,
+            line_end: 3,
+        },
+        FetchOpts::default(),
+    )
+    .expect("fetch_span on twin B after flip-flop must still succeed");
+}

crates/kebab-cli/src/main.rs

@@ -275,6 +275,14 @@ enum Cmd {
         #[arg(long, group = "reset_scope")]
         config_only: bool,
 
+        /// Purge stored docs that are outside the current walker scope
+        /// (config narrowing / removed sub-directory). No filesystem paths
+        /// are removed — this is purely a store-level reconciliation.
+        /// Filesystem existence is NOT checked; anything the current walker
+        /// would not visit is considered an orphan and removed from the store.
+        #[arg(long, group = "reset_scope")]
+        orphans_only: bool,
+
         /// Skip the interactive confirm. Required in non-interactive
         /// contexts (CI, pipes).
         #[arg(long)]
@@ -595,14 +603,20 @@ fn run(cli: &Cli) -> anyhow::Result<()> {
                 println!("{}", serde_json::to_string(&wire::wire_ingest(&report))?);
             } else {
                 let skipped_breakdown = kebab_app::render_skipped_breakdown(&report.skipped_by_extension);
+                let purged_suffix = if report.purged_deleted_files > 0 {
+                    format!("  purged {}", report.purged_deleted_files)
+                } else {
+                    String::new()
+                };
                 println!(
-                    "scanned {}  new {}  updated {}  skipped {}{}  errors {}  ({} ms)",
+                    "scanned {}  new {}  updated {}  skipped {}{}  errors {}{}  ({} ms)",
                     report.scanned,
                     report.new,
                     report.updated,
                     report.skipped,
                     skipped_breakdown,
                     report.errors,
+                    purged_suffix,
                     report.duration_ms
                 );
             }
@@ -1088,6 +1102,7 @@ fn run(cli: &Cli) -> anyhow::Result<()> {
             data_only: _,
             vector_only,
             config_only,
+            orphans_only,
             yes,
         } => {
             use kebab_app::ResetScope;
@@ -1101,11 +1116,50 @@ fn run(cli: &Cli) -> anyhow::Result<()> {
                 ResetScope::VectorOnly
             } else if *config_only {
                 ResetScope::ConfigOnly
+            } else if *orphans_only {
+                ResetScope::OrphansOnly
             } else {
                 ResetScope::DataOnly
             };
 
             let cfg = kebab_config::Config::load(cli.config.as_deref())?;
+
+            if matches!(scope, ResetScope::OrphansOnly) {
+                // OrphansOnly: confirm UI shows orphan count + sample paths
+                // rather than on-disk directory sizes.
+                let orphan_paths = kebab_app::enumerate_orphans(&cfg)?;
+
+                if !*yes {
+                    use std::io::IsTerminal;
+                    if !std::io::stdin().is_terminal() {
+                        anyhow::bail!(
+                            "reset --orphans-only is destructive and stdin is non-interactive — pass --yes to proceed"
+                        );
+                    }
+                    if !confirm_orphans_only(&orphan_paths)? {
+                        if !cli.quiet {
+                            eprintln!("aborted.");
+                        }
+                        return Ok(());
+                    }
+                }
+
+                let report = kebab_app::reset::execute(scope, &cfg)?;
+                if cli.json {
+                    println!("{}", serde_json::to_string(&wire::wire_reset(&report))?);
+                } else {
+                    if report.orphans_purged > 0 {
+                        println!("orphans purged: {}", report.orphans_purged);
+                        for p in &report.purged_paths {
+                            println!("  - {}", p.0);
+                        }
+                    } else {
+                        println!("no orphaned docs found — store is already in sync with walker scope");
+                    }
+                }
+                return Ok(());
+            }
+
             let paths = kebab_app::reset::enumerate_paths(scope, &cfg);
             let bytes = kebab_app::reset::estimate_size_bytes(&paths);
 
@@ -1444,6 +1498,46 @@ fn confirm_destructive(
     Ok(matches!(s.as_str(), "y" | "yes"))
 }
 
+/// Confirm prompt for `--orphans-only`: shows the orphan count + a
+/// sample of up to 5 paths so the user knows what will be purged before
+/// committing. No filesystem paths are removed — only store records.
+fn confirm_orphans_only(
+    orphan_paths: &[kebab_core::WorkspacePath],
+) -> anyhow::Result<bool> {
+    use std::io::Write;
+    let n = orphan_paths.len();
+    let mut out = std::io::stderr().lock();
+
+    if n == 0 {
+        writeln!(out, "no orphaned docs found — nothing to purge.")?;
+        out.flush()?;
+        // Nothing to do; treat as confirmed so the caller can emit the
+        // "no orphans" report without prompting.
+        return Ok(true);
+    }
+
+    let sample: Vec<&str> = orphan_paths
+        .iter()
+        .take(5)
+        .map(|p| p.0.as_str())
+        .collect();
+    let sample_str = sample.join(", ");
+    let ellipsis = if n > 5 { ", …" } else { "" };
+
+    writeln!(
+        out,
+        "Purge {n} stored doc(s) outside the current walker scope? (no filesystem paths removed)"
+    )?;
+    writeln!(out, "  sample: {sample_str}{ellipsis}")?;
+    write!(out, "[y/N] ")?;
+    out.flush()?;
+
+    let mut line = String::new();
+    std::io::stdin().read_line(&mut line)?;
+    let s = line.trim().to_ascii_lowercase();
+    Ok(matches!(s.as_str(), "y" | "yes"))
+}
+
 /// p9-fb-35: human-friendly plain output for `kebab fetch`.
 fn render_fetch_plain(r: &kebab_core::FetchResult) {
     println!("# {} ({})", r.doc_path.0, format_kind(r.kind));

crates/kebab-cli/src/wire.rs

@@ -260,6 +260,7 @@ mod tests {
             skipped_generated: 0,
             skipped_size_exceeded: 0,
             skip_examples: SkipExamples::default(),
+            purged_deleted_files: 0,
             items: None,
         };
         let v = wire_ingest(&r);
@@ -364,6 +365,8 @@ mod tests {
             scope: kebab_app::ResetScope::DataOnly,
             removed_paths: vec![std::path::PathBuf::from("/tmp/x")],
             embedding_rows_truncated: 0,
+            orphans_purged: 0,
+            purged_paths: vec![],
         };
         let v = wire_reset(&r);
         assert_eq!(schema_of(&v), Some("reset_report.v1"));

crates/kebab-core/src/ingest.rs

@@ -47,6 +47,12 @@ pub struct IngestReport {
     /// p10-1A-1: sample file paths per skip category (≤ 5 each).
     #[serde(default)]
     pub skip_examples: SkipExamples,
+    /// Dogfood: docs whose on-disk file was deleted since the last ingest
+    /// and were therefore removed from the store. Additive field — older
+    /// wire consumers that pre-date this field read it as 0 via
+    /// `#[serde(default)]`.
+    #[serde(default)]
+    pub purged_deleted_files: u32,
     /// `None` ↔ wire `items: null` (`--summary-only`).
     pub items: Option<Vec<IngestItem>>,
 }
@@ -136,6 +142,7 @@ mod tests {
                 builtin_blacklist: vec!["node_modules/x.js".into()],
                 gitignore: vec![],
             },
+            purged_deleted_files: 0,
             items: None,
         };
         let v = serde_json::to_value(&r).unwrap();

crates/kebab-core/src/traits.rs

@@ -8,7 +8,7 @@ use serde_json::Value;
 use crate::asset::{RawAsset, WorkspacePath};
 use crate::chunk::Chunk;
 use crate::document::{Block, CanonicalDocument};
-use crate::ids::{ChunkId, DocumentId};
+use crate::ids::{AssetId, ChunkId, DocumentId};
 use crate::jobs::{JobFilter, JobId, JobKind, JobRow, JobStatus};
 use crate::media::MediaType;
 use crate::search::{DocFilter, DocSummary, SearchFilters, SearchHit, SearchQuery};
@@ -161,10 +161,23 @@ pub trait DocumentStore {
     fn get_document(&self, id: &DocumentId) -> anyhow::Result<Option<CanonicalDocument>>;
     fn get_chunk(&self, id: &ChunkId) -> anyhow::Result<Option<Chunk>>;
     fn list_documents(&self, filter: &DocFilter) -> anyhow::Result<Vec<DocSummary>>;
+    /// Look up an asset row by its `asset_id` (PRIMARY KEY = blake3
+    /// content hash). Twin-file safe: asset_id is PK so there is
+    /// exactly one row per unique content hash, regardless of how many
+    /// `documents` rows share it. Use this instead of
+    /// `get_asset_by_workspace_path` when you already have a
+    /// `CanonicalDocument` (which carries `source_asset_id`).
+    fn get_asset(&self, id: &AssetId) -> anyhow::Result<Option<RawAsset>>;
+
     /// p9-fb-23: look up an asset row by its workspace path. Used by
     /// the incremental-ingest skip path to compare the freshly
     /// computed blake3 checksum against what's already in SQLite. The
     /// schema enforces a unique workspace_path per asset.
+    ///
+    /// NOTE: for twin files (identical content at different paths),
+    /// `assets.workspace_path` is "last-registered path" — it
+    /// flip-flops on every ingest. Prefer `get_asset` (by asset_id)
+    /// when you have a `CanonicalDocument.source_asset_id`.
     fn get_asset_by_workspace_path(
         &self,
         path: &WorkspacePath,
@@ -183,6 +196,16 @@ pub trait DocumentStore {
         &self,
         path: &WorkspacePath,
     ) -> anyhow::Result<Option<CanonicalDocument>>;
+
+    /// Return every `workspace_path` stored in the `documents` table.
+    ///
+    /// Used by the post-walker sweep in `kebab-app::ingest` to detect
+    /// documents whose source file has been deleted from the filesystem.
+    /// The set difference `(stored - scanned)` yields orphan candidates;
+    /// each candidate is then existence-checked on disk so that
+    /// out-of-scope files (config narrowing) are NOT purged — only truly
+    /// absent files trigger the purge.
+    fn all_workspace_paths(&self) -> anyhow::Result<Vec<WorkspacePath>>;
 }
 
 pub trait VectorStore {

crates/kebab-parse-code/src/lang.rs

@@ -24,7 +24,7 @@ pub fn code_lang_for_path(path: &Path) -> Option<&'static str> {
     match ext.as_str() {
         "rs" => Some("rust"),
         "py" | "pyi" => Some("python"),
-        "ts" | "tsx" => Some("typescript"),
+        "ts" | "tsx" | "mts" | "cts" => Some("typescript"),
         "js" | "mjs" | "cjs" | "jsx" => Some("javascript"),
         "go" => Some("go"),
         "java" => Some("java"),
@@ -82,7 +82,7 @@ pub fn module_path_for_python(workspace_path: &str) -> String {
 /// (no slash replacement, no source-root strip). See plan §Task C.
 pub fn module_path_for_tsjs(workspace_path: &str) -> String {
     let p = workspace_path;
-    for ext in [".tsx", ".ts", ".jsx", ".mjs", ".cjs", ".js"] {
+    for ext in [".tsx", ".mts", ".cts", ".ts", ".jsx", ".mjs", ".cjs", ".js"] {
         if let Some(stripped) = p.strip_suffix(ext) {
             return stripped.to_string();
         }
@@ -110,7 +110,7 @@ mod tests {
 
     #[test]
     fn module_path_for_tsjs_keeps_slashes_and_strips_ext() {
-        for ext in ["ts", "tsx", "js", "jsx", "mjs", "cjs"] {
+        for ext in ["ts", "tsx", "mts", "cts", "js", "jsx", "mjs", "cjs"] {
             let p = format!("src/search/retriever/Retriever.{ext}");
             assert_eq!(module_path_for_tsjs(&p), "src/search/retriever/Retriever");
         }

crates/kebab-parse-code/src/typescript.rs

@@ -173,8 +173,9 @@ impl Extractor for TypescriptAstExtractor {
 }
 
 /// Select the tree-sitter grammar based on the workspace path's
-/// extension. `.tsx` → TSX grammar; everything else (`.ts`, `.d.ts`,
-/// missing extension) → TypeScript grammar.
+/// extension. `.tsx` → TSX grammar; everything else (`.ts`, `.mts`,
+/// `.cts`, `.d.ts`, missing extension) → TypeScript grammar (the JSX-
+/// agnostic variants all share one grammar in tree-sitter-typescript 0.23).
 fn select_grammar(workspace_path: &str) -> tree_sitter::Language {
     if workspace_path.ends_with(".tsx") {
         tree_sitter_typescript::LANGUAGE_TSX.into()

crates/kebab-parse-code/tests/lang.rs

@@ -9,6 +9,8 @@ fn known_extensions_map_to_canonical_identifiers() {
         ("foo.pyi", Some("python")),
         ("foo.ts", Some("typescript")),
         ("foo.tsx", Some("typescript")),
+        ("foo.mts", Some("typescript")),  // ESM TS — same grammar
+        ("foo.cts", Some("typescript")),  // CommonJS TS — same grammar
         ("foo.js", Some("javascript")),
         ("foo.mjs", Some("javascript")),
         ("foo.cjs", Some("javascript")),

crates/kebab-source-fs/src/media.rs

@@ -19,7 +19,9 @@ pub(crate) fn media_type_for(path: &Path) -> MediaType {
         .unwrap_or_default();
 
     match ext.as_str() {
-        "md" => MediaType::Markdown,
+        // Markdown + MDX (markdown + JSX, treated as plain markdown — the
+        // JSX islands are folded into raw passthrough by the md parser).
+        "md" | "mdx" => MediaType::Markdown,
         "pdf" => MediaType::Pdf,
 
         "png" => MediaType::Image(ImageType::Png),
@@ -40,7 +42,8 @@ pub(crate) fn media_type_for(path: &Path) -> MediaType {
 
         // p10-1B: Python / TS / JS AST chunkers active.
         "py" | "pyi"               => MediaType::Code("python".into()),
-        "ts" | "tsx"               => MediaType::Code("typescript".into()),
+        // .mts / .cts are TypeScript ESM / CommonJS variants — same grammar.
+        "ts" | "tsx" | "mts" | "cts" => MediaType::Code("typescript".into()),
         "js" | "mjs" | "cjs" | "jsx" => MediaType::Code("javascript".into()),
 
         // Empty string (no extension) and any other extension: bucket as
@@ -102,6 +105,20 @@ mod tests {
         assert_eq!(media_type_for(Path::new("a/b.rs")),    MediaType::Code("rust".into()));
     }
 
+    #[test]
+    fn ts_variants_mts_cts() {
+        // .mts / .cts are TypeScript ESM / CommonJS — same grammar as .ts.
+        assert_eq!(media_type_for(Path::new("a/b.mts")), MediaType::Code("typescript".into()));
+        assert_eq!(media_type_for(Path::new("a/b.cts")), MediaType::Code("typescript".into()));
+    }
+
+    #[test]
+    fn mdx_routes_to_markdown() {
+        // MDX is markdown with JSX islands; the md parser folds the JSX
+        // through as raw passthrough.
+        assert_eq!(media_type_for(Path::new("docs/page.mdx")), MediaType::Markdown);
+    }
+
     #[test]
     fn unknown_and_missing_extension() {
         assert_eq!(

crates/kebab-store-sqlite/snapshots/ingest_report.snapshot.json

@@ -56,5 +56,6 @@
   "skipped_kebabignore": 0,
   "skipped_size_exceeded": 0,
   "unchanged": 0,
+  "purged_deleted_files": 0,
   "updated": 1
 }

crates/kebab-store-sqlite/src/documents.rs

@@ -264,6 +264,28 @@ impl kebab_core::DocumentStore for SqliteStore {
         }))
     }
 
+    fn get_asset(
+        &self,
+        id: &kebab_core::AssetId,
+    ) -> Result<Option<kebab_core::RawAsset>> {
+        let conn = self.lock_conn();
+        let result = conn.query_row(
+            r#"SELECT
+                asset_id, source_uri, workspace_path, media_type,
+                byte_len, checksum, storage_kind, storage_path,
+                discovered_at
+            FROM assets
+            WHERE asset_id = ?"#,
+            rusqlite::params![id.0.as_str()],
+            asset_from_row,
+        );
+        match result {
+            Ok(asset) => Ok(Some(asset)),
+            Err(rusqlite::Error::QueryReturnedNoRows) => Ok(None),
+            Err(e) => Err(e.into()),
+        }
+    }
+
     fn get_asset_by_workspace_path(
         &self,
         path: &kebab_core::WorkspacePath,
@@ -352,6 +374,22 @@ impl kebab_core::DocumentStore for SqliteStore {
         }))
     }
 
+    fn all_workspace_paths(&self) -> Result<Vec<kebab_core::WorkspacePath>> {
+        let conn = self.lock_conn();
+        let mut stmt = conn
+            .prepare("SELECT workspace_path FROM documents")
+            .map_err(StoreError::from)?;
+        let rows = stmt
+            .query_map([], |r| r.get::<_, String>(0))
+            .map_err(StoreError::from)?;
+        let mut out = Vec::new();
+        for row in rows {
+            let path = row.map_err(StoreError::from)?;
+            out.push(kebab_core::WorkspacePath(path));
+        }
+        Ok(out)
+    }
+
     fn list_documents(
         &self,
         filter: &kebab_core::DocFilter,
@@ -616,7 +654,8 @@ fn rows_optional<T>(err: rusqlite::Error) -> rusqlite::Result<Option<T>> {
 
 /// Reconstruct a [`kebab_core::RawAsset`] from one `assets` row.
 /// Row mapper for `RawAsset`. Column names are self-documenting; the
-/// SELECT in [`DocumentStore::get_asset_by_workspace_path`] must include
+/// SELECTs in [`DocumentStore::get_asset`] and
+/// [`DocumentStore::get_asset_by_workspace_path`] must both include
 /// all nine columns by their schema names.
 fn asset_from_row(row: &rusqlite::Row<'_>) -> rusqlite::Result<kebab_core::RawAsset> {
     use std::path::PathBuf;

crates/kebab-store-sqlite/src/lib.rs

@@ -35,4 +35,4 @@ pub use error::StoreError;
 pub use eval::{EvalQueryResultRecord, EvalRunRecord, EvalRunRow};
 pub use fts::rebuild_chunks_fts;
 pub use jobs::IngestRunRow;
-pub use store::{CountSummary, NotIndexed, SqliteStore};
+pub use store::{CountSummary, NotIndexed, SqliteStore, purge_deleted_workspace_path};

crates/kebab-store-sqlite/src/store.rs

@@ -540,10 +540,132 @@ pub(crate) fn purge_orphan_at_workspace_path(
     Ok(())
 }
 
+/// Purge all stored data for a document whose on-disk file has been
+/// deleted (as opposed to content-changed, which is handled by
+/// `purge_orphan_at_workspace_path`).
+///
+/// Returns the `chunk_id`s that were associated with the document so
+/// the caller can issue a matching `VectorStore::delete_by_chunk_ids`
+/// on the LanceDB side.
+///
+/// Deletion order:
+/// 1. Collect chunk_ids (before cascade removes them).
+/// 2. DELETE the `documents` row → CASCADE clears `blocks`, `chunks`,
+///    `embedding_records`.
+/// 3. DELETE the `assets` row **only if no other document still
+///    references it** (twin-file protection — `assets` can be shared
+///    across identical-content files via the blake3 PK).
+/// 4. If the asset was `storage_kind = 'copied'`, best-effort delete
+///    the on-disk byte file at `storage_path`.
+///
+/// Returns `Ok(vec![])` when no document exists at `workspace_path`
+/// (idempotent — caller doesn't need to pre-check).
+pub fn purge_deleted_workspace_path(
+    store: &SqliteStore,
+    workspace_path: &kebab_core::WorkspacePath,
+) -> anyhow::Result<Vec<kebab_core::ChunkId>> {
+    let conn = store.lock_conn();
+
+    // Look up the document + its asset_id.
+    let doc_row: Option<(String, String)> = conn
+        .query_row(
+            "SELECT doc_id, asset_id FROM documents WHERE workspace_path = ?",
+            rusqlite::params![workspace_path.0],
+            |r| Ok((r.get(0)?, r.get(1)?)),
+        )
+        .optional()
+        .map_err(StoreError::from)?;
+
+    let Some((doc_id, asset_id)) = doc_row else {
+        return Ok(Vec::new());
+    };
+
+    // 1. Collect chunk_ids before CASCADE removes them.
+    let mut stmt = conn
+        .prepare("SELECT chunk_id FROM chunks WHERE doc_id = ?")
+        .map_err(StoreError::from)?;
+    let rows = stmt
+        .query_map(rusqlite::params![doc_id], |r| r.get::<_, String>(0))
+        .map_err(StoreError::from)?;
+    let chunk_ids: Vec<kebab_core::ChunkId> = rows
+        .map(|r| r.map(kebab_core::ChunkId))
+        .collect::<rusqlite::Result<Vec<_>>>()
+        .map_err(StoreError::from)?;
+    drop(stmt);
+
+    // 2. DELETE the document row (CASCADE clears blocks / chunks /
+    //    embedding_records via the FK constraints in V001).
+    conn.execute(
+        "DELETE FROM documents WHERE doc_id = ?",
+        rusqlite::params![doc_id],
+    )
+    .map_err(StoreError::from)?;
+
+    // 3. Delete the asset row only when no other document still
+    //    references it (twin-file safety: two files with identical
+    //    bytes share a single asset row via the blake3 PK).
+    let remaining_refs: i64 = conn
+        .query_row(
+            "SELECT COUNT(*) FROM documents WHERE asset_id = ?",
+            rusqlite::params![asset_id],
+            |r| r.get(0),
+        )
+        .map_err(StoreError::from)?;
+
+    if remaining_refs == 0 {
+        // 4. Capture storage details before deleting the row.
+        let asset_storage: Option<(String, String)> = conn
+            .query_row(
+                "SELECT storage_kind, storage_path FROM assets WHERE asset_id = ?",
+                rusqlite::params![asset_id],
+                |r| Ok((r.get(0)?, r.get(1)?)),
+            )
+            .optional()
+            .map_err(StoreError::from)?;
+
+        conn.execute(
+            "DELETE FROM assets WHERE asset_id = ?",
+            rusqlite::params![asset_id],
+        )
+        .map_err(StoreError::from)?;
+
+        // 5. Best-effort: remove the on-disk copied asset file.
+        if let Some((storage_kind, storage_path)) = asset_storage {
+            if storage_kind == "copied" {
+                let _ = std::fs::remove_file(&storage_path);
+            }
+        }
+    }
+
+    tracing::debug!(
+        target: "kebab-store-sqlite",
+        workspace_path = %workspace_path.0,
+        doc_id = %doc_id,
+        chunk_count = chunk_ids.len(),
+        "purged deleted-file document from store"
+    );
+
+    Ok(chunk_ids)
+}
+
 /// UPSERT a row into `assets`. Used by both the `put_asset_with_bytes`
 /// path (which has bytes + computed `storage_kind/path`) and the
 /// `DocumentStore::put_asset` path (which only has the `RawAsset` and
 /// reads `storage_kind/path` from `asset.stored`).
+///
+/// **`assets.workspace_path` is "last-registered path" semantics for
+/// twin files** (two source files with identical content share one
+/// `assets` row keyed on `asset_id = blake3(content)`). Each ingest
+/// of either twin overwrites `workspace_path` with whichever path was
+/// seen most recently — this is intentional and correct after PR #146
+/// made `try_skip_unchanged` document-centric (uses
+/// `get_document_by_workspace_path`, not `get_asset_by_workspace_path`)
+/// and PR #149 made `reset --orphans-only` document-centric too.
+/// Do NOT "fix" the flip-flop by adding a UNIQUE constraint on
+/// `workspace_path` in the `assets` table — twin de-dup is load-bearing.
+/// When you need media_type for a known document, use the 2-step lookup
+/// `get_document_by_workspace_path` → `doc.source_asset_id` →
+/// `get_asset(asset_id)` so the result is twin-safe.
 pub(crate) fn upsert_asset_row(
     conn: &Connection,
     asset: &kebab_core::RawAsset,

crates/kebab-store-sqlite/tests/ingest_report_snapshot.rs

@@ -41,6 +41,7 @@ fn fixture_report() -> IngestReport {
         skipped_generated: 0,
         skipped_size_exceeded: 0,
         skip_examples: kebab_core::SkipExamples::default(),
+        purged_deleted_files: 0,
         items: Some(vec![
             IngestItem {
                 kind: IngestItemKind::New,

docs/wire-schema/v1/reset_report.schema.json

@@ -14,12 +14,18 @@
     "schema_version":           { "const": "reset_report.v1" },
     "scope":                    {
       "type": "string",
-      "enum": ["all", "data_only", "vector_only", "config_only"]
+      "enum": ["all", "data_only", "vector_only", "config_only", "orphans_only"]
     },
     "removed_paths":            {
       "type": "array",
       "items": { "type": "string" }
     },
-    "embedding_rows_truncated": { "type": "integer", "minimum": 0 }
+    "embedding_rows_truncated": { "type": "integer", "minimum": 0 },
+    "orphans_purged":           { "type": "integer", "minimum": 0, "default": 0 },
+    "purged_paths":             {
+      "type": "array",
+      "items": { "type": "string" },
+      "default": []
+    }
   }
 }