@hover-dev/core 0.16.0 → 0.17.0

package/dist/scripts/bench-multi-tab.js

@@ -1,192 +0,0 @@
-/**
- * Benchmark agent success rate on the multi-tab "Pay with PayHover" flow.
- *
- * Why this exists — v0.10's central theme is "agent can drive cross-tab
- * flows in the wild." The system-prompt addendum (cdpHint.ts rule 5/6/7)
- * is the lever we're tuning. This script gives us a number to tune
- * against: across N iterations, how often does the agent get from
- * "browse the store" to "Order placed"?
- *
- * Per iteration the agent has to:
- *   1. Browse the e-commerce store, add 1+ items to cart, go to checkout.
- *   2. Fill the shipping form.
- *   3. Pick "Pay with PayHover" (opens a new tab at localhost:5177).
- *   4. Switch to the new tab, fill card number + CVV, click Continue.
- *   5. Wait ~600ms for the simulated 3DS pre-check.
- *   6. Fill the 6-digit OTP (always 123456 in the sandbox).
- *   7. Click Confirm. The provider tab closes itself.
- *   8. Switch back to the original tab, observe the "Order placed" view.
- *
- * Steps 3, 4, 7, 8 are the failure-prone ones.
- *
- * Assumes:
- *   - Debug Chrome on :9222 (run `pnpm smoke:chrome`).
- *   - e-commerce on :5174 AND payment-provider on :5177 both running
- *     (run `pnpm dev:example:e-commerce` and `pnpm dev:example:payment-provider`
- *     in two terminals before invoking this).
- *
- * Usage:
- *   pnpm --filter @hover-dev/core exec tsx src/scripts/bench-multi-tab.ts [n]
- *   pnpm bench-multi-tab [n]
- *
- * `n` defaults to 5. Per-iteration timeout is 5 minutes — multi-tab flows
- * are slow because the agent does a lot of browser_snapshot calls.
- *
- * Output: per-run pass/fail + final summary (success rate, median wall
- * time, median turns, median cost in $). A/B prompt changes by running
- * once on each branch and comparing.
- */
-import { WebSocket } from 'ws';
-import { startService } from '../service.js';
-const PROMPT = process.env.HOVER_BENCH_PROMPT ??
-    [
-        'Open http://localhost:5174 (Hover Store).',
-        'Add any item to the cart, go to checkout, fill the shipping form with',
-        'realistic values, then choose "Pay with PayHover". A new tab opens at',
-        'the payment provider — switch to it, fill in card 4242 4242 4242 4242',
-        'with CVV 123, click Continue, wait for the OTP step, enter 123456,',
-        'click Confirm. The popup will close. Switch back to the original tab',
-        'and verify the order shows as placed.',
-    ].join(' ');
-const ITERATIONS = Number(process.argv[2] ?? 5);
-const PER_RUN_TIMEOUT_MS = 5 * 60 * 1000;
-async function singleRun(idx) {
-    process.stderr.write(`\n[bench-multi-tab] run ${idx + 1}/${ITERATIONS}\n`);
-    const service = await startService({
-        port: 0,
-        agentId: 'claude',
-        model: 'sonnet',
-        cdpUrl: 'http://localhost:9222',
-        devRoot: process.cwd(),
-    });
-    return new Promise((resolve) => {
-        const ws = new WebSocket(`ws://127.0.0.1:${service.port}`);
-        const t0 = performance.now();
-        let turns = 0;
-        let costUsd = null;
-        let resolved = false;
-        const finish = (result) => {
-            if (resolved)
-                return;
-            resolved = true;
-            try {
-                ws.close(1000);
-            }
-            catch { /* already closed */ }
-            service.close().finally(() => resolve(result));
-        };
-        const timeout = setTimeout(() => {
-            finish({
-                ok: false,
-                wallMs: performance.now() - t0,
-                turns,
-                costUsd,
-                reason: `timed out after ${PER_RUN_TIMEOUT_MS / 1000}s`,
-            });
-        }, PER_RUN_TIMEOUT_MS);
-        ws.on('open', () => {
-            ws.send(JSON.stringify({ type: 'command', payload: { text: PROMPT } }));
-        });
-        ws.on('message', (raw) => {
-            let msg;
-            try {
-                msg = JSON.parse(raw.toString());
-            }
-            catch {
-                return;
-            }
-            if (msg.type !== 'event')
-                return;
-            const ev = msg.payload;
-            if (ev.kind === 'tool_use') {
-                turns += 1;
-                if (process.env.HOVER_BENCH_VERBOSE === '1') {
-                    const ev2 = ev;
-                    process.stderr.write(`    [turn ${turns}] ${ev2.name ?? '<tool>'}\n`);
-                }
-            }
-            if (ev.kind === 'session_end') {
-                clearTimeout(timeout);
-                const evAny = ev;
-                if (typeof evAny.costUsd === 'number')
-                    costUsd = evAny.costUsd;
-                finish({
-                    ok: !evAny.isError,
-                    wallMs: performance.now() - t0,
-                    turns,
-                    costUsd,
-                    reason: evAny.isError ? 'agent reported error' : undefined,
-                });
-            }
-        });
-        ws.on('error', (err) => {
-            clearTimeout(timeout);
-            finish({
-                ok: false,
-                wallMs: performance.now() - t0,
-                turns,
-                costUsd,
-                reason: `WS error: ${err.message}`,
-            });
-        });
-    });
-}
-function median(xs) {
-    if (xs.length === 0)
-        return 0;
-    const sorted = [...xs].sort((a, b) => a - b);
-    const mid = Math.floor(sorted.length / 2);
-    return sorted.length % 2 === 0
-        ? (sorted[mid - 1] + sorted[mid]) / 2
-        : sorted[mid];
-}
-function fmtMs(ms) {
-    return `${(ms / 1000).toFixed(1)}s`;
-}
-function fmtUsd(usd) {
-    return usd == null ? '–' : `$${usd.toFixed(4)}`;
-}
-async function main() {
-    process.stderr.write(`[bench-multi-tab] ${ITERATIONS} iterations, per-run timeout ${PER_RUN_TIMEOUT_MS / 1000}s\n`);
-    process.stderr.write(`[bench-multi-tab] prompt: ${PROMPT.slice(0, 80)}…\n`);
-    const results = [];
-    for (let i = 0; i < ITERATIONS; i++) {
-        try {
-            const r = await singleRun(i);
-            results.push(r);
-            const status = r.ok ? '✓ PASS' : '✗ FAIL';
-            process.stderr.write(`[bench-multi-tab] run ${i + 1}: ${status} · ${fmtMs(r.wallMs)} · ${r.turns} turns · ${fmtUsd(r.costUsd)}${r.reason ? ` · ${r.reason}` : ''}\n`);
-        }
-        catch (err) {
-            const msg = err instanceof Error ? err.message : String(err);
-            results.push({ ok: false, wallMs: 0, turns: 0, costUsd: null, reason: msg });
-            process.stderr.write(`[bench-multi-tab] run ${i + 1}: ✗ FAIL · setup error · ${msg}\n`);
-        }
-    }
-    const passes = results.filter((r) => r.ok);
-    const successRate = passes.length / results.length;
-    process.stderr.write('\n[bench-multi-tab] summary\n');
-    process.stderr.write(`  success rate: ${(successRate * 100).toFixed(0)}% (${passes.length}/${results.length})\n`);
-    if (passes.length > 0) {
-        process.stderr.write(`  median wall:  ${fmtMs(median(passes.map((r) => r.wallMs)))}\n`);
-        process.stderr.write(`  median turns: ${median(passes.map((r) => r.turns)).toFixed(0)}\n`);
-        const costs = passes.map((r) => r.costUsd).filter((c) => c != null);
-        if (costs.length > 0) {
-            process.stderr.write(`  median cost:  ${fmtUsd(median(costs))}\n`);
-        }
-    }
-    if (passes.length < results.length) {
-        process.stderr.write(`\n  failures:\n`);
-        results.forEach((r, i) => {
-            if (!r.ok)
-                process.stderr.write(`    run ${i + 1}: ${r.reason ?? 'unknown'}\n`);
-        });
-    }
-    // Exit non-zero if EVERY run failed — useful for CI plumbing later. A
-    // partial-pass run still exits 0 so we collect signal across branches.
-    process.exit(passes.length === 0 ? 1 : 0);
-}
-main().catch((err) => {
-    process.stderr.write(`[bench-multi-tab] fatal: ${err instanceof Error ? err.stack ?? err.message : String(err)}\n`);
-    process.exit(1);
-});

package/dist/scripts/bench-ttfb.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=bench-ttfb.d.ts.map

package/dist/scripts/bench-ttfb.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"bench-ttfb.d.ts","sourceRoot":"","sources":["../../src/scripts/bench-ttfb.ts"],"names":[],"mappings":""}

package/dist/scripts/bench-ttfb.js

@@ -1,127 +0,0 @@
-/**
- * Benchmark "time to first tool_use" for the LLM-driven loop.
- *
- * Assumes:
- *  - A debug Chrome is running on :9222 (start with `pnpm smoke:chrome`).
- *  - A dev server is running so the agent has something to drive
- *    (`pnpm dev:example:basic-app`).
- *
- * Per iteration:
- *  - Start a fresh Hover service (cold — kills any prior service to avoid
- *    cached MCP process state across iterations).
- *  - WS-connect, send a fixed command, mark t0 right before send().
- *  - Mark t1 on the first tool_use event from the agent.
- *  - Report (t1 - t0) in milliseconds. Close service + WS.
- *
- *   pnpm --filter @hover-dev/core exec tsx src/scripts/bench-ttfb.ts <n>
- *
- * `n` defaults to 5. Prints individual timings + median + min/max.
- */
-import { WebSocket } from 'ws';
-import { startService } from '../service.js';
-const PROMPT = process.env.HOVER_BENCH_PROMPT ?? 'Take a snapshot of the page.';
-const ITERATIONS = Number(process.argv[2] ?? 5);
-async function singleRun() {
-    const service = await startService({
-        // Use 0 to let the kernel pick — avoids cross-iter EADDRINUSE races.
-        port: 0,
-        agentId: 'claude',
-        model: 'sonnet',
-        cdpUrl: 'http://localhost:9222',
-        devRoot: process.cwd(),
-    });
-    return new Promise((resolve, reject) => {
-        const ws = new WebSocket(`ws://127.0.0.1:${service.port}`);
-        let t0 = 0;
-        let resolved = false;
-        const timeout = setTimeout(() => {
-            if (!resolved) {
-                ws.close(1000);
-                service.close();
-                reject(new Error('timed out waiting for first tool_use after 60s'));
-            }
-        }, 60_000);
-        ws.on('open', () => {
-            t0 = performance.now();
-            ws.send(JSON.stringify({ type: 'command', payload: { text: PROMPT } }));
-        });
-        ws.on('message', raw => {
-            let msg;
-            try {
-                msg = JSON.parse(raw.toString());
-            }
-            catch {
-                return;
-            }
-            if (process.env.HOVER_BENCH_VERBOSE === '1') {
-                process.stderr.write(`    [event] ${raw.toString().slice(0, 200)}\n`);
-            }
-            if (msg.type !== 'event')
-                return;
-            const ev = msg.payload;
-            if (ev.kind === 'tool_use' && !resolved) {
-                const t1 = performance.now();
-                const ms = t1 - t0;
-                resolved = true;
-                clearTimeout(timeout);
-                ws.close(1000);
-                service.close().finally(() => resolve(ms));
-            }
-            if (ev.kind === 'session_end' && !resolved) {
-                // Ran without any tool_use — agent went text-only or errored.
-                // Reject so the bench surfaces the issue instead of recording
-                // a misleadingly tiny "first tool_use" timing.
-                resolved = true;
-                clearTimeout(timeout);
-                const evAny = ev;
-                const reason = evAny.isError ? 'session_end (error)' : 'session_end without tool_use';
-                ws.close(1000);
-                service.close().finally(() => reject(new Error(reason)));
-            }
-        });
-        ws.on('error', err => {
-            if (resolved)
-                return;
-            resolved = true;
-            clearTimeout(timeout);
-            service.close().finally(() => reject(err));
-        });
-    });
-}
-function median(xs) {
-    const sorted = [...xs].sort((a, b) => a - b);
-    const mid = Math.floor(sorted.length / 2);
-    return sorted.length % 2 === 0
-        ? (sorted[mid - 1] + sorted[mid]) / 2
-        : sorted[mid];
-}
-async function main() {
-    console.log(`prompt: ${JSON.stringify(PROMPT)}`);
-    console.log(`iterations: ${ITERATIONS}`);
-    console.log('');
-    const results = [];
-    for (let i = 1; i <= ITERATIONS; i++) {
-        try {
-            const ms = await singleRun();
-            results.push(ms);
-            console.log(`  run ${i}: ${ms.toFixed(0).padStart(5)} ms`);
-        }
-        catch (err) {
-            console.error(`  run ${i}: FAILED — ${err instanceof Error ? err.message : String(err)}`);
-        }
-        // Small gap between runs so any process-cleanup tail can flush.
-        await new Promise(r => setTimeout(r, 500));
-    }
-    if (results.length === 0) {
-        console.error('\nNo successful runs.');
-        process.exit(1);
-    }
-    console.log('');
-    console.log(`min:    ${Math.min(...results).toFixed(0)} ms`);
-    console.log(`median: ${median(results).toFixed(0)} ms`);
-    console.log(`max:    ${Math.max(...results).toFixed(0)} ms`);
-}
-main().catch(err => {
-    console.error(err);
-    process.exit(1);
-});

package/dist/scripts/start-chrome.d.ts

@@ -1,3 +0,0 @@
-#!/usr/bin/env node
-export {};
-//# sourceMappingURL=start-chrome.d.ts.map

package/dist/scripts/start-chrome.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"start-chrome.d.ts","sourceRoot":"","sources":["../../src/scripts/start-chrome.ts"],"names":[],"mappings":""}

package/dist/scripts/start-chrome.js

@@ -1,23 +0,0 @@
-#!/usr/bin/env node
-/**
- * "Start debug Chrome on port 9222" CLI.
- *
- * Two entry points:
- *   - Repo dev:    `pnpm smoke:chrome`           → tsx src/scripts/start-chrome.ts
- *   - npm consumer: `pnpm exec hover-chrome`     → dist/scripts/start-chrome.js
- *                  (or `npx hover-chrome`, bin exposed by vite-plugin-hover)
- *
- * All actual launch logic lives in ../playwright/launchChrome.ts.
- */
-import { launchDebugChrome } from '../playwright/launchChrome.js';
-const result = await launchDebugChrome();
-if (!result.ok) {
-    console.error(`[hover:chrome] ${result.reason}`);
-    process.exit(1);
-}
-if (result.alreadyRunning) {
-    console.log(`[hover:chrome] already listening on ${result.port}`);
-}
-else {
-    console.log(`[hover:chrome] ready on ${result.port} (data-dir=${result.userDataDir})`);
-}

package/dist/skills/writeSkill.d.ts

@@ -1,27 +0,0 @@
-/**
- * Captured-step type.
- *
- * NOTE: Save-as-Skill (writing `.claude/skills/<slug>/SKILL.md` for agent
- * replay) was retired — `spec` + ⟳ Re-record covers intent-driven replay, and
- * "skill" collided with Claude Code's own skills concept. All that remains
- * here is `SkillStep`: the serialized message shape from the widget's
- * localStorage, which the whole spec pipeline (writeSpec, sidecar, listSpecs,
- * Page-Object extraction) consumes as `SpecStep`. The file keeps its path so
- * the many `import { SkillStep } from '../skills/writeSkill.js'` call sites
- * don't churn; renaming to a neutral module is a separate mechanical pass.
- */
-/**
- * Serialized message shape from the widget's localStorage. Matches the
- * `state.messages` schema in packages/widget-bootstrap/src/widget/client.js.
- */
-export interface SkillStep {
-    kind: 'user' | 'system' | 'step' | 'ai' | 'done';
-    text?: string;
-    tool?: string;
-    input?: unknown;
-    isError?: boolean;
-    turns?: number;
-    costUsd?: number;
-    summary?: string;
-}
-//# sourceMappingURL=writeSkill.d.ts.map

package/dist/skills/writeSkill.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"writeSkill.d.ts","sourceRoot":"","sources":["../../src/skills/writeSkill.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH;;;GAGG;AACH,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,MAAM,GAAG,QAAQ,GAAG,MAAM,GAAG,IAAI,GAAG,MAAM,CAAC;IACjD,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB"}

package/dist/skills/writeSkill.js

@@ -1,13 +0,0 @@
-/**
- * Captured-step type.
- *
- * NOTE: Save-as-Skill (writing `.claude/skills/<slug>/SKILL.md` for agent
- * replay) was retired — `spec` + ⟳ Re-record covers intent-driven replay, and
- * "skill" collided with Claude Code's own skills concept. All that remains
- * here is `SkillStep`: the serialized message shape from the widget's
- * localStorage, which the whole spec pipeline (writeSpec, sidecar, listSpecs,
- * Page-Object extraction) consumes as `SpecStep`. The file keeps its path so
- * the many `import { SkillStep } from '../skills/writeSkill.js'` call sites
- * don't churn; renaming to a neutral module is a separate mechanical pass.
- */
-export {};

package/dist/specs/listSpecs.d.ts

@@ -1,52 +0,0 @@
-import { type OptimizationSuggestion } from './optimizationSuggestion.js';
-export interface SpecSummary {
-    /** Path-relative slug, e.g. `login-and-counter`. Identifies the spec. */
-    slug: string;
-    /** Absolute path to the .spec.ts file. */
-    path: string;
-    /** `Original prompt:` parsed from the JSDoc header. `null` for
-     *  hand-authored specs that have no header — they list but can't be
-     *  re-recorded automatically. */
-    originalPrompt: string | null;
-    /** First line of `Outcome:` from the JSDoc header, if present. */
-    outcome: string | null;
-    /** Number of `Steps:` lines parsed (informational only). */
-    stepCount: number;
-    /** File mtime in ms — used to show "saved 2 hours ago" in the UI. */
-    mtimeMs: number;
-    /** Whether a structured `.hover/<slug>.json` sidecar exists. The widget
-     *  gates the optimization pass on this — without a captured session there's
-     *  no observed feedback for the LLM to add assertions from. */
-    hasSidecar: boolean;
-    /** Count of `// hover:optimizable` markers the deterministic translator left
-     *  — interactions it couldn't fully translate single-step. >0 is a strong
-     *  signal to run the optimization pass (or add a seed). */
-    optimizableCount: number;
-    /** The default-off "review optimization?" nudge (F7/D10): suggested + reasons,
-     *  derived from optimizable markers + relevant seeds. */
-    optimization: OptimizationSuggestion;
-}
-export interface SpecHeader {
-    /** Raw text of `Original prompt:` line, or null when absent. */
-    originalPrompt: string | null;
-    /** First line of `Outcome:`. */
-    outcome: string | null;
-    /** Step lines from the `Steps:` block, in order. */
-    steps: string[];
-    /** Lines from the `Expected:` block, in order. */
-    expected: string[];
-}
-/**
- * Parse the JSDoc header that `writeSpec.ts` emits. Tolerant of:
- *   - Specs without any JSDoc (returns all-null).
- *   - Hand-edited specs where users reordered or trimmed sections.
- *   - Long prompts that wrap across lines (we take only the first line).
- */
-export declare function parseSpecHeader(source: string): SpecHeader;
-/**
- * List every `*.spec.ts` file under `<devRoot>/__vibe_tests__/` with its
- * parsed header. Returns newest-first by mtime so the widget overlay shows
- * recently-saved specs at the top.
- */
-export declare function listSpecs(devRoot: string): Promise<SpecSummary[]>;
-//# sourceMappingURL=listSpecs.d.ts.map

package/dist/specs/listSpecs.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"listSpecs.d.ts","sourceRoot":"","sources":["../../src/specs/listSpecs.ts"],"names":[],"mappings":"AAoBA,OAAO,EAA0B,KAAK,sBAAsB,EAAE,MAAM,6BAA6B,CAAC;AAGlG,MAAM,WAAW,WAAW;IAC1B,yEAAyE;IACzE,IAAI,EAAE,MAAM,CAAC;IACb,0CAA0C;IAC1C,IAAI,EAAE,MAAM,CAAC;IACb;;qCAEiC;IACjC,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,kEAAkE;IAClE,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,4DAA4D;IAC5D,SAAS,EAAE,MAAM,CAAC;IAClB,qEAAqE;IACrE,OAAO,EAAE,MAAM,CAAC;IAChB;;mEAE+D;IAC/D,UAAU,EAAE,OAAO,CAAC;IACpB;;+DAE2D;IAC3D,gBAAgB,EAAE,MAAM,CAAC;IACzB;6DACyD;IACzD,YAAY,EAAE,sBAAsB,CAAC;CACtC;AAED,MAAM,WAAW,UAAU;IACzB,gEAAgE;IAChE,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,gCAAgC;IAChC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,oDAAoD;IACpD,KAAK,EAAE,MAAM,EAAE,CAAC;IAChB,kDAAkD;IAClD,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAED;;;;;GAKG;AACH,wBAAgB,eAAe,CAAC,MAAM,EAAE,MAAM,GAAG,UAAU,CAsB1D;AA4BD;;;;GAIG;AACH,wBAAsB,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC,CA2DvE"}

package/dist/specs/listSpecs.js

@@ -1,139 +0,0 @@
-/**
- * List + parse Hover-generated Playwright specs under `<devRoot>/__vibe_tests__/`.
- *
- * Used by:
- *   - The widget's "Specs" overlay tab (server pushes a SpecSummary[] list).
- *   - The CLI's `hover re-record <spec>` subcommand (parses one spec for its
- *     `Original prompt:` JSDoc header).
- *
- * Hand-authored specs (no Hover JSDoc header) are listed but reported with
- * `originalPrompt: null` — the UI / CLI surfaces that "this spec can't be
- * re-recorded automatically; the natural-language intent isn't recorded."
- *
- * Shares the SpecSummary row shape the widget's Specs tab renders.
- */
-import { readdir, readFile } from 'node:fs/promises';
-import { stat } from 'node:fs/promises';
-import { existsSync } from 'node:fs';
-import { join } from 'node:path';
-import { countOptimizableMarkers } from './writeSpec.js';
-import { readSeeds, relevantSeeds } from './seeds.js';
-import { optimizationSuggestion } from './optimizationSuggestion.js';
-/**
- * Parse the JSDoc header that `writeSpec.ts` emits. Tolerant of:
- *   - Specs without any JSDoc (returns all-null).
- *   - Hand-edited specs where users reordered or trimmed sections.
- *   - Long prompts that wrap across lines (we take only the first line).
- */
-export function parseSpecHeader(source) {
-    // JSDoc block right after the @playwright/test import (or at file top).
-    // We don't require it to be the very first JSDoc — there could be a
-    // banner comment from a linter. We DO require it to appear before the
-    // first `test(` / `test.describe(` so that long file footers can't
-    // confuse the parser.
-    const beforeFirstTest = source.split(/^\s*(?:test|test\.describe)\s*\(/m)[0] ?? source;
-    const blockMatch = beforeFirstTest.match(/\/\*\*([\s\S]*?)\*\//);
-    if (!blockMatch) {
-        return { originalPrompt: null, outcome: null, steps: [], expected: [] };
-    }
-    const block = blockMatch[1];
-    const originalPrompt = extractScalar(block, /^\s*\*\s*Original prompt:\s*(.+?)\s*$/m);
-    const outcome = extractScalar(block, /^\s*\*\s*Outcome:\s*(.+?)\s*$/m);
-    return {
-        originalPrompt,
-        outcome,
-        steps: extractList(block, /^\s*\*\s*Steps:\s*$/m),
-        expected: extractList(block, /^\s*\*\s*Expected:\s*$/m),
-    };
-}
-function extractScalar(block, re) {
-    const m = block.match(re);
-    return m ? m[1].trim() : null;
-}
-/**
- * Extract a JSDoc list-style block. Given a header regex matching "Steps:"
- * or "Expected:", read subsequent ` * <indented line>` lines until the next
- * top-level marker (blank ` *` line or another `Section:` header).
- */
-function extractList(block, headerRe) {
-    const match = block.match(headerRe);
-    if (!match)
-        return [];
-    const start = (match.index ?? 0) + match[0].length;
-    const tail = block.slice(start);
-    const lines = [];
-    for (const raw of tail.split('\n')) {
-        // Stop at a blank JSDoc line (` *` only) or another `Section:` header.
-        if (/^\s*\*\s*$/.test(raw))
-            break;
-        if (/^\s*\*\s*\w[\w ]*:\s*$/.test(raw) || /^\s*\*\s*\w[\w ]*:\s/.test(raw))
-            break;
-        const m = raw.match(/^\s*\*\s*(?:[•\-\*\d.]\s*)*(.+?)\s*$/);
-        if (m && m[1])
-            lines.push(m[1]);
-    }
-    return lines;
-}
-/**
- * List every `*.spec.ts` file under `<devRoot>/__vibe_tests__/` with its
- * parsed header. Returns newest-first by mtime so the widget overlay shows
- * recently-saved specs at the top.
- */
-export async function listSpecs(devRoot) {
-    const root = join(devRoot, '__vibe_tests__');
-    let entries;
-    try {
-        entries = await readdir(root);
-    }
-    catch {
-        return [];
-    }
-    // Seeds are devRoot-wide; read once and reuse for every spec's suggestion.
-    const seeds = await readSeeds(devRoot);
-    const summaries = [];
-    for (const entry of entries) {
-        if (!entry.endsWith('.spec.ts'))
-            continue;
-        const path = join(root, entry);
-        let content;
-        let mtimeMs = 0;
-        try {
-            content = await readFile(path, 'utf-8');
-            const st = await stat(path);
-            mtimeMs = st.mtimeMs;
-        }
-        catch {
-            continue;
-        }
-        const header = parseSpecHeader(content);
-        const slug = entry.replace(/\.spec\.ts$/, '');
-        const sidecarPath = join(root, '.hover', `${slug}.json`);
-        const hasSidecar = existsSync(sidecarPath);
-        const optimizableCount = countOptimizableMarkers(content);
-        // Which seeds could plausibly apply, from the sidecar's captured tools.
-        let relevantSeedNames = [];
-        if (hasSidecar && seeds.length > 0) {
-            try {
-                const sc = JSON.parse(await readFile(sidecarPath, 'utf-8'));
-                const tools = new Set((sc.steps ?? []).filter(s => s.kind === 'step' && s.tool).map(s => s.tool));
-                relevantSeedNames = relevantSeeds(seeds, tools).map(s => s.name);
-            }
-            catch {
-                /* malformed sidecar — treat as no relevant seeds */
-            }
-        }
-        summaries.push({
-            slug,
-            path,
-            originalPrompt: header.originalPrompt,
-            outcome: header.outcome,
-            stepCount: header.steps.length,
-            mtimeMs,
-            hasSidecar,
-            optimizableCount,
-            optimization: optimizationSuggestion({ hasSidecar, optimizableCount, relevantSeedNames }),
-        });
-    }
-    summaries.sort((a, b) => b.mtimeMs - a.mtimeMs);
-    return summaries;
-}

package/dist/specs/optimizationSuggestion.d.ts

@@ -1,26 +0,0 @@
-/**
- * The default-off "should we nudge the user to optimize this spec?" signal
- * (F7 / D10). Optimization never runs automatically; instead, when a saved spec
- * has a clear improvable shape, the widget surfaces a "review optimization?"
- * prompt. This computes that decision + human-readable reasons.
- *
- * Pure function so it's trivially testable; `listSpecs` gathers the inputs
- * (optimizable-marker count, sidecar presence, relevant seed names) and attaches
- * the result to each SpecSummary.
- */
-export interface OptimizationSuggestion {
-    /** Whether to nudge the user to run the optimization pass on this spec. */
-    suggested: boolean;
-    /** Human-readable reasons, for the widget tooltip / prompt. Empty when not
-     *  suggested. */
-    reasons: string[];
-}
-export declare function optimizationSuggestion(args: {
-    /** Whether a `.hover/<slug>.json` sidecar exists. */
-    hasSidecar: boolean;
-    /** Count of `// hover:optimizable` markers in the spec. */
-    optimizableCount: number;
-    /** Names of seeds whose signature is relevant to this spec's tools. */
-    relevantSeedNames: string[];
-}): OptimizationSuggestion;
-//# sourceMappingURL=optimizationSuggestion.d.ts.map

package/dist/specs/optimizationSuggestion.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"optimizationSuggestion.d.ts","sourceRoot":"","sources":["../../src/specs/optimizationSuggestion.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,MAAM,WAAW,sBAAsB;IACrC,2EAA2E;IAC3E,SAAS,EAAE,OAAO,CAAC;IACnB;qBACiB;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;CACnB;AAED,wBAAgB,sBAAsB,CAAC,IAAI,EAAE;IAC3C,qDAAqD;IACrD,UAAU,EAAE,OAAO,CAAC;IACpB,2DAA2D;IAC3D,gBAAgB,EAAE,MAAM,CAAC;IACzB,uEAAuE;IACvE,iBAAiB,EAAE,MAAM,EAAE,CAAC;CAC7B,GAAG,sBAAsB,CAqBzB"}

package/dist/specs/optimizationSuggestion.js

@@ -1,28 +0,0 @@
-/**
- * The default-off "should we nudge the user to optimize this spec?" signal
- * (F7 / D10). Optimization never runs automatically; instead, when a saved spec
- * has a clear improvable shape, the widget surfaces a "review optimization?"
- * prompt. This computes that decision + human-readable reasons.
- *
- * Pure function so it's trivially testable; `listSpecs` gathers the inputs
- * (optimizable-marker count, sidecar presence, relevant seed names) and attaches
- * the result to each SpecSummary.
- */
-export function optimizationSuggestion(args) {
-    const { hasSidecar, optimizableCount, relevantSeedNames } = args;
-    const reasons = [];
-    // The optimization pass reads the sidecar (observed feedback, captured steps);
-    // without one there's nothing to optimize from. Matches the widget's Optimize
-    // gate, so we never suggest what can't be acted on.
-    if (!hasSidecar)
-        return { suggested: false, reasons };
-    if (optimizableCount > 0) {
-        const n = optimizableCount;
-        reasons.push(`${n} interaction${n === 1 ? '' : 's'} couldn't be fully translated — the optimization pass can complete ${n === 1 ? 'it' : 'them'}`);
-    }
-    if (relevantSeedNames.length > 0) {
-        const k = relevantSeedNames.length;
-        reasons.push(`${k} seed${k === 1 ? '' : 's'} may apply: ${relevantSeedNames.join(', ')}`);
-    }
-    return { suggested: reasons.length > 0, reasons };
-}