diff --git a/docs/superpowers/plans/2026-09-30-quick-switcher-and-inline-ai.md b/docs/superpowers/plans/2026-09-30-quick-switcher-and-inline-ai.md
new file mode 100644
index 0000000..16db99c
--- /dev/null
+++ b/docs/superpowers/plans/2026-09-30-quick-switcher-and-inline-ai.md
@@ -0,0 +1,47 @@
+# SP-1: Quick-switcher fuzzy overlay + Inline AI assist
+
+**Branch:** master
+**Started:** 2026-09-30 IST
+**Status:** in progress
+
+## Scope
+
+Replace the existing Cmd+P "Recent Files" menu with a fuzzy quick-switcher
+overlay, and add Cmd+K inline AI assist (Rewrite / Shorten / Expand) on
+selected text using the existing `AiProviders` infrastructure.
+
+## Design decisions (user-approved)
+
+| Question | Answer |
+|---|---|
+| Cmd+P behavior | Replace Recent Files menu with overlay |
+| AI response mode | Typewriter stream into selection |
+| AI action set | Rewrite + Shorten + Expand only (no custom prompts) |
+
+## Commit plan
+
+### SP-1A — Quick-switcher fuzzy overlay
+- [ ] **C1**: `feat(quick-switcher): pure fuzzy matcher + ranker` — `src/quick-switcher/fuzzy-matcher.js` + tests
+- [ ] **C2**: `feat(quick-switcher): IPC for listing workspace files` — main.js + preload.js wiring
+- [ ] **C3**: `feat(quick-switcher): overlay UI + keyboard nav` — `src/quick-switcher/quick-switcher-overlay.js` + tests
+- [ ] **C4**: `feat(quick-switcher): wire Cmd+P into renderer, drop old Recent Files menu`
+
+### SP-1B — Inline AI assist
+- [ ] **C5**: `feat(ai-assist): pure prompt builder + result applier` — `src/ai-assist/inline-assist.js` + tests
+- [ ] **C6**: `feat(ai-assist): IPC streaming wrapper around AiProviders` — main.js handler
+- [ ] **C7**: `feat(ai-assist): floating popover + cancel/stop controls` — `src/renderer/inline-ai-popover.js` + tests
+- [ ] **C8**: `feat(ai-assist): Cmd+K keymap + popover mount in renderer`
+
+## TDD discipline
+
+- All pure modules test: integration tests first, then implementation
+- All renderer UI test: DOM contract tests using @testing-library/dom
+- All IPC handlers: tests using injected fetch stub
+- Lint + format green at every commit
+- Grep forbidden markers before each commit claim
+
+## Out of scope (deferred)
+
+- Per-action configurable prompts
+- Multi-selection batch assist
+- AI inline assist without specific settings (built on AI Assistant plugin config)
\ No newline at end of file
diff --git a/src/quick-switcher/fuzzy-matcher.js b/src/quick-switcher/fuzzy-matcher.js
new file mode 100644
index 0000000..a1fa6f9
--- /dev/null
+++ b/src/quick-switcher/fuzzy-matcher.js
@@ -0,0 +1,200 @@
+/**
+ * Quick-switcher fuzzy matcher.
+ *
+ * Pure module — no DOM, no IPC, no Node globals — so it can run in the
+ * renderer (for instant client-side filtering) or the main process (for
+ * precomputed indices). Tested under @jest-environment node.
+ *
+ * Scoring tiers:
+ *   exact match        → 1.00  (query === target, case-insensitive)
+ *   stem match         → 1.00  (query === target without first extension)
+ *   prefix match       → 0.80
+ *   subsequence match  → 0.50 + bonuses, capped at 0.79
+ *   no match           → 0.00
+ *
+ * The stem tier exists because a user typing `idea` against a file named
+ * `idea.md` is signalling the same intent as typing the full name — they
+ * have the unique file in mind and aren't fuzzy-matching. Treating it as
+ * 1.0 (rather than 0.8 prefix) means recent/open-tabs boosts work as
+ * expected: 1.0 × 2.0 × 1.5 = 3.0 for an exact, open, recent file.
+ *
+ * Subsequence matching tries BOTH leftmost-greedy and rightmost-greedy
+ * alignments and keeps the higher-scoring one. Rightmost wins for queries
+ * that target a file's extension or suffix ("md" → "readme.md"); leftmost
+ * wins for queries that target a leading prefix ("re" → "readme.md").
+ *
+ * Bonuses (each +0.05, capped):
+ *   - word-boundary hit: previous char is one of  / - _ space .
+ *     OR position is at start-of-target (i=0)
+ *     OR position is at end-of-target (i=length-1)
+ *   - consecutive run: each pair of adjacent match positions in target
+ *
+ * Subsequence score is capped below the prefix tier (0.79) so the ranking
+ * is stable: prefix > subsequence even when the subsequence is a perfect
+ * substring match.
+ *
+ * @module fuzzy-matcher
+ */
+
+const SEPARATORS = /[\/\-_. ]/;
+const SUBSEQ_CAP = 0.79;
+const BOUNDARY_BONUS = 0.05;
+const CONSECUTIVE_BONUS = 0.1;
+
+/**
+ * Score `query` against `target` (typically a filename basename).
+ *
+ * @param {string} query
+ * @param {string} target
+ * @returns {number} score in [0, 1]
+ */
+function fuzzyScore(query, target) {
+  if (typeof query !== 'string' || typeof target !== 'string') return 0;
+  if (query.length === 0 || target.length === 0) return 0;
+
+  const q = query.toLowerCase();
+  const t = target.toLowerCase();
+
+  if (q === t) return 1.0;
+
+  // Stem match: query equals target with the first extension stripped.
+  // 'idea' against 'idea.md' → stem 'idea' → 1.0
+  // 'archive' against 'archive.tar.gz' → stem 'archive' → 1.0
+  // '.gitignore' (no real extension) → stem '.gitignore', query must match whole name
+  const firstDot = t.indexOf('.');
+  const stem = firstDot >= 1 ? t.slice(0, firstDot) : t;
+  if (q === stem) return 1.0;
+
+  if (t.startsWith(q)) return 0.8;
+
+  const leftMatch = greedyMatch(q, t, 'left');
+  const rightMatch = greedyMatch(q, t, 'right');
+
+  const leftScore = leftMatch ? scoreMatch(leftMatch, t) : 0;
+  const rightScore = rightMatch ? scoreMatch(rightMatch, t) : 0;
+
+  const best = Math.max(leftScore, rightScore);
+  return best > 0 ? Math.min(SUBSEQ_CAP, best) : 0;
+}
+
+/**
+ * Greedy subsequence match. Returns the array of target indices where
+ * each query character was matched, or null if the query isn't a
+ * subsequence.
+ *
+ * 'left'  → for each query char, take the first match at or after the
+ *           previous match position.
+ * 'right' → for each query char (processed right-to-left), take the
+ *           last match at or before the next-needed position.
+ */
+function greedyMatch(query, target, direction) {
+  const positions = [];
+  if (direction === 'left') {
+    let ti = 0;
+    for (let qi = 0; qi < query.length; qi++) {
+      let found = -1;
+      for (let j = ti; j < target.length; j++) {
+        if (target[j] === query[qi]) {
+          found = j;
+          break;
+        }
+      }
+      if (found === -1) return null;
+      positions.push(found);
+      ti = found + 1;
+    }
+    return positions;
+  }
+  // right
+  let ti = target.length;
+  for (let qi = query.length - 1; qi >= 0; qi--) {
+    let found = -1;
+    for (let j = ti - 1; j >= 0; j--) {
+      if (target[j] === query[qi]) {
+        found = j;
+        break;
+      }
+    }
+    if (found === -1) return null;
+    positions.unshift(found);
+    ti = found;
+  }
+  return positions;
+}
+
+/**
+ * Score a match by counting boundary hits and consecutive pairs.
+ */
+function scoreMatch(positions, target) {
+  let boundaryHits = 0;
+  let consecutivePairs = 0;
+  const n = target.length;
+
+  for (let i = 0; i < positions.length; i++) {
+    const p = positions[i];
+    if (p === 0 || p === n - 1 || SEPARATORS.test(target[p - 1] || '')) {
+      boundaryHits++;
+    }
+    if (i > 0 && positions[i] === positions[i - 1] + 1) {
+      consecutivePairs++;
+    }
+  }
+
+  return 0.5 + boundaryHits * BOUNDARY_BONUS + consecutivePairs * CONSECUTIVE_BONUS;
+}
+
+/**
+ * Rank `items` against `query`, applying per-path boosts. Items with no
+ * match (and not boosted) are excluded.
+ *
+ * Boosts are multiplicative on the base score:
+ *   - recent   → ×2.0
+ *   - openTabs → ×1.5
+ *
+ * An empty query bypasses scoring entirely and returns items purely by
+ * boost: recent first (×2.0), then open tabs (×1.5), then nothing.
+ *
+ * @param {string} query
+ * @param {Array<{path:string, name?:string}>} items
+ * @param {object} [boosts]
+ * @param {Set<string>} [boosts.recent]
+ * @param {Set<string>} [boosts.openTabs]
+ * @returns {Array<{item:object, score:number}>} sorted descending by score
+ */
+function rankResults(query, items, boosts = {}) {
+  if (!Array.isArray(items)) return [];
+  const { recent = new Set(), openTabs = new Set() } = boosts;
+
+  const isEmptyQuery = typeof query !== 'string' || query.length === 0;
+  const out = [];
+
+  for (const item of items) {
+    if (!item || typeof item.path !== 'string') continue;
+
+    let score;
+    if (isEmptyQuery) {
+      if (recent.has(item.path)) score = 2.0;
+      else if (openTabs.has(item.path)) score = 1.5;
+      else continue;
+    } else {
+      const base = fuzzyScore(query, item.name || '');
+      if (base === 0) continue;
+      score = base;
+      if (recent.has(item.path)) score *= 2.0;
+      if (openTabs.has(item.path)) score *= 1.5;
+    }
+    out.push({ item, score });
+  }
+
+  out.sort((a, b) => b.score - a.score);
+  return out;
+}
+
+module.exports = {
+  fuzzyScore,
+  rankResults,
+  // exported for tests / advanced callers
+  greedyMatch,
+  scoreMatch,
+  SUBSEQ_CAP,
+};
diff --git a/tests/quick-switcher-fuzzy-matcher.test.js b/tests/quick-switcher-fuzzy-matcher.test.js
new file mode 100644
index 0000000..4f97e10
--- /dev/null
+++ b/tests/quick-switcher-fuzzy-matcher.test.js
@@ -0,0 +1,168 @@
+/**
+ * @jest-environment node
+ *
+ * Quick-switcher fuzzy matcher.
+ *
+ * Tiered scoring:
+ *   - exact match         → 1.00
+ *   - prefix match        → 0.80
+ *   - subsequence match   → 0.50 + bonuses (word-boundary, consecutive)
+ *   - no match            → 0.00
+ *
+ * Empty query returns 0 for every target — the caller (rankResults) falls
+ * back to recency/open-tabs boosts alone to rank empty-query results.
+ *
+ * All scoring is case-insensitive. The original target casing is preserved
+ * for display; only the matching logic lowercases.
+ */
+
+const { fuzzyScore, rankResults } = require('../src/quick-switcher/fuzzy-matcher');
+
+describe('fuzzyScore', () => {
+  test('exact match returns 1.0', () => {
+    expect(fuzzyScore('readme.md', 'readme.md')).toBe(1.0);
+  });
+
+  test('exact match is case-insensitive', () => {
+    expect(fuzzyScore('README.md', 'readme.md')).toBe(1.0);
+    expect(fuzzyScore('readme.MD', 'README.md')).toBe(1.0);
+  });
+
+  test('prefix match returns ~0.8', () => {
+    const score = fuzzyScore('read', 'readme.md');
+    expect(score).toBeGreaterThanOrEqual(0.8);
+    expect(score).toBeLessThan(0.9);
+  });
+
+  test('prefix match is case-insensitive', () => {
+    expect(fuzzyScore('READ', 'readme.md')).toBeCloseTo(0.8, 2);
+  });
+
+  test('subsequence match returns ~0.5 + bonuses', () => {
+    const score = fuzzyScore('rmd', 'readme.md');
+    expect(score).toBeGreaterThanOrEqual(0.5);
+    expect(score).toBeLessThan(0.8);
+  });
+
+  test('subsequence match on word boundary scores higher than mid-word', () => {
+    const boundary = fuzzyScore('md', 'readme.md'); // 'md' is a suffix → still prefix-like
+    const mid = fuzzyScore('ea', 'readme.md'); // 'ea' is mid-word subsequence
+    expect(boundary).toBeGreaterThan(mid);
+  });
+
+  test('consecutive matches score higher than scattered matches', () => {
+    const consecutive = fuzzyScore('re', 'readme.md'); // contiguous
+    const scattered = fuzzyScore('rd', 'readme.md'); // not contiguous
+    expect(consecutive).toBeGreaterThan(scattered);
+  });
+
+  test('no match returns 0', () => {
+    expect(fuzzyScore('xyz', 'readme.md')).toBe(0);
+    expect(fuzzyScore('', '')).toBe(0);
+  });
+
+  test('empty query returns 0 (caller decides what to do)', () => {
+    expect(fuzzyScore('', 'readme.md')).toBe(0);
+  });
+
+  test('query longer than target returns 0', () => {
+    expect(fuzzyScore('readme.md.bak', 'readme.md')).toBe(0);
+  });
+
+  test('non-string inputs return 0', () => {
+    expect(fuzzyScore(null, 'readme.md')).toBe(0);
+    expect(fuzzyScore(undefined, 'readme.md')).toBe(0);
+    expect(fuzzyScore('rm', null)).toBe(0);
+    expect(fuzzyScore('rm', undefined)).toBe(0);
+  });
+
+  test('preserves display casing (does not mutate target)', () => {
+    const target = 'README.md';
+    fuzzyScore('readme', target);
+    expect(target).toBe('README.md');
+  });
+});
+
+describe('rankResults', () => {
+  const items = [
+    { path: '/notes/readme.md', name: 'readme.md' },
+    { path: '/notes/rust/intro.md', name: 'intro.md' },
+    { path: '/notes/idea.md', name: 'idea.md' },
+    { path: '/archive/old-readme.md', name: 'old-readme.md' },
+    { path: '/totally/unrelated.txt', name: 'unrelated.txt' },
+  ];
+
+  test('sorts by score descending', () => {
+    const ranked = rankResults('readme', items);
+    expect(ranked.length).toBeGreaterThan(0);
+    for (let i = 1; i < ranked.length; i++) {
+      expect(ranked[i - 1].score).toBeGreaterThanOrEqual(ranked[i].score);
+    }
+  });
+
+  test('excludes zero-score items', () => {
+    const ranked = rankResults('xyz', items);
+    expect(ranked).toEqual([]);
+  });
+
+  test('applies recency boost (×2.0) to recent paths', () => {
+    const recent = new Set(['/archive/old-readme.md']);
+    const ranked = rankResults('readme', items, { recent });
+    // old-readme is a prefix match → base 0.8; with ×2.0 boost = 1.6
+    // readme.md is a prefix match → base 0.8; no boost
+    // So old-readme should rank first despite being older
+    expect(ranked[0].item.path).toBe('/archive/old-readme.md');
+  });
+
+  test('applies open-tabs boost (×1.5)', () => {
+    const openTabs = new Set(['/notes/idea.md']);
+    const ranked = rankResults('idea', items, { openTabs });
+    // idea.md is exact match → base 1.0; with ×1.5 boost = 1.5
+    expect(ranked[0].item.path).toBe('/notes/idea.md');
+    expect(ranked[0].score).toBeCloseTo(1.5, 2);
+  });
+
+  test('combines recency and open-tabs boosts (multiplicative)', () => {
+    const recent = new Set(['/notes/idea.md']);
+    const openTabs = new Set(['/notes/idea.md']);
+    const ranked = rankResults('idea', items, { recent, openTabs });
+    // exact 1.0 × 2.0 × 1.5 = 3.0
+    expect(ranked[0].score).toBeCloseTo(3.0, 2);
+  });
+
+  test('empty query falls back to boosts only — recent first, then open tabs', () => {
+    const recent = new Set(['/archive/old-readme.md']);
+    const openTabs = new Set(['/notes/rust/intro.md']);
+    const ranked = rankResults('', items, { recent, openTabs });
+    // Recent items get base × 2.0; open tabs × 1.5; nothing else
+    expect(ranked[0].item.path).toBe('/archive/old-readme.md');
+    expect(ranked[1].item.path).toBe('/notes/rust/intro.md');
+  });
+
+  test('empty query with no boosts returns empty array', () => {
+    const ranked = rankResults('', items, {});
+    expect(ranked).toEqual([]);
+  });
+
+  test('returns empty array for empty items', () => {
+    expect(rankResults('readme', [])).toEqual([]);
+  });
+
+  test('exact match beats prefix match beats subsequence match', () => {
+    const ranked = rankResults('idea', items);
+    // idea.md is exact match (1.0)
+    // old-readme.md is subsequence (matches 'i', 'd', 'e', 'a' across path chars)
+    expect(ranked[0].item.path).toBe('/notes/idea.md');
+  });
+
+  test('handles identical names in different paths via path-aware boost', () => {
+    const dupItems = [
+      { path: '/a/readme.md', name: 'readme.md' },
+      { path: '/b/readme.md', name: 'readme.md' },
+    ];
+    const recent = new Set(['/b/readme.md']);
+    const ranked = rankResults('readme', dupItems, { recent });
+    expect(ranked[0].item.path).toBe('/b/readme.md');
+    expect(ranked[1].item.path).toBe('/a/readme.md');
+  });
+});