verikun 0.12.0 → 0.14.0

package/README.md

@@ -251,7 +251,8 @@ iteration by construction, which is what makes it a loop.
   the bundled per-1M price table if it drifts.
 - **`--model`** picks the model and its provider — Anthropic (`claude-haiku-4-5` ·
   `claude-sonnet-4-6` (default) · `claude-opus-4-8` · `claude-fable-5`), OpenAI
-  (`gpt-5.4-mini` · `gpt-5.4` · `gpt-5.5`), each read from its own key
+  (`gpt-5.4-mini` · `gpt-5.4` · `gpt-5.5` · `gpt-4.1`, the last a non-reasoning model
+  that undercuts the default sonnet at $2/$8 per 1M), each read from its own key
   (`ANTHROPIC_API_KEY` / `OPENAI_API_KEY`), or a CLI backend — **`codex-cli`** (the
   logged-in `codex` binary) or **`cursor-cli`** (`cursor-agent`) — which need no key at
   all: spend is on your subscription, so their cost line is `$0` and `--max-cost-usd` /
@@ -576,6 +577,20 @@ vk tap @tap_to_continue_label_id
 **Cost:** $0.45 · **Wall time:** ~4 min · **Model:** Claude Sonnet 4.6 with
 prompt-cache hits (1 M cache-read tokens kept cost low on a long conversation).
 
+## Feedback — help improve verikun
+
+verikun improves from the rough edges people hit while driving it. When verikun *itself* is
+the friction — a step that heals on every cached replay (an unstable compiled selector,
+often a label-only control with no resource-id), a repair "give-up", or a gotcha in its own
+operation — that's worth an issue at
+[github.com/ddikman/verikun/issues](https://github.com/ddikman/verikun/issues).
+
+Driving verikun with an AI agent + the [skill](.claude/skills/verikun/SKILL.md)? It hands
+off to the **`suggest-verikun-improvement`** skill, which drafts a short, TL;DR-first
+suggestion, **reviews it with you before anything is submitted**, and **redacts every
+app-under-test specific** (package, on-screen text, selector values, test prose, logs) so no
+client code or logic can leak.
+
 ## Build from source
 
 For local development, or to run an unreleased version, build from a clone:

package/dist/agent/cost.js

@@ -11,8 +11,9 @@ const errors_1 = require("../errors");
 // ALLOWED_MODELS and providerFor all derive from this, so the --model allowlist, its
 // price and its backend can never disagree. Prices WILL drift between releases
 // (Anthropic cached 2026-05-26; OpenAI 2026-07-02) — `--cost-override <input/output>`
-// is the escape hatch and is authoritative when supplied. Every model here bills cached
-// input at ~0.1x (Anthropic + OpenAI gpt-5.x alike), matching CACHE_READ_MULT below.
+// is the escape hatch and is authoritative when supplied. Nearly every model here bills
+// cached input at the ~0.1x CACHE_READ_MULT below (Anthropic + OpenAI gpt-5.x alike); the
+// exception carries an explicit `cacheReadMult`.
 const MODELS = {
     'claude-haiku-4-5': { input: 1, output: 5, provider: 'anthropic' },
     'claude-sonnet-4-6': { input: 3, output: 15, provider: 'anthropic' },
@@ -21,6 +22,11 @@ const MODELS = {
     'gpt-5.4-mini': { input: 0.75, output: 4.5, provider: 'openai' },
     'gpt-5.4': { input: 2.5, output: 15, provider: 'openai' },
     'gpt-5.5': { input: 5, output: 30, provider: 'openai' },
+    // The one NON-REASONING model in the registry, and cheaper than the default sonnet
+    // ($2/$8 vs $3/$15). Two consequences, both handled rather than papered over: it takes
+    // no reasoning_effort (openai.ts's REASONING_MODELS gate skips the param for it), and it
+    // bills cache reads at 0.25x rather than the 0.1x every other model here uses.
+    'gpt-4.1': { input: 2, output: 8, provider: 'openai', cacheReadMult: 0.25 },
     // CLI-agent backends: billed to the user's ChatGPT/Cursor subscription via an already-logged-in
     // CLI, not per token — so price is $0 and --max-cost-usd/--cost-override are inert no-ops (the
     // run is bounded by maxRepairs + --timeout instead). The `-cli` suffix reads clearly as "the
@@ -29,7 +35,9 @@ const MODELS = {
     'codex-cli': { input: 0, output: 0, provider: 'codex' },
     'cursor-cli': { input: 0, output: 0, provider: 'cursor' },
 };
-exports.MODEL_PRICES = Object.fromEntries(Object.entries(MODELS).map(([m, s]) => [m, { input: s.input, output: s.output }]));
+// Strip `provider` off each spec; what remains IS the Price (including any cacheReadMult,
+// so a per-model deviation reaches the tracker without being re-listed here).
+exports.MODEL_PRICES = Object.fromEntries(Object.entries(MODELS).map(([m, { provider: _provider, ...price }]) => [m, price]));
 exports.ALLOWED_MODELS = Object.keys(MODELS);
 exports.DEFAULT_MODEL = 'claude-sonnet-4-6';
 /** Resolve which provider backend serves a model (unknown → anthropic, the default). */
@@ -59,7 +67,8 @@ function parseCostOverride(raw) {
 function priceFor(model, override) {
     return override ?? exports.MODEL_PRICES[model] ?? exports.MODEL_PRICES[exports.DEFAULT_MODEL];
 }
-// Cache reads bill at ~0.1x input; cache writes (5-min TTL) at ~1.25x input.
+// Cache reads bill at ~0.1x input (unless the model overrides it via Price.cacheReadMult);
+// cache writes (5-min TTL) at ~1.25x input.
 const CACHE_READ_MULT = 0.1;
 const CACHE_WRITE_MULT = 1.25;
 const PER_M = 1_000_000;
@@ -72,7 +81,7 @@ function estimateCostUsd(usage, price) {
     return ((input * price.input +
         output * price.output +
         cacheWrite * price.input * CACHE_WRITE_MULT +
-        cacheRead * price.input * CACHE_READ_MULT) /
+        cacheRead * price.input * (price.cacheReadMult ?? CACHE_READ_MULT)) /
         PER_M);
 }
 /**

package/dist/agent/openai.js

@@ -27,9 +27,6 @@ const DEFAULT_REQUEST_TIMEOUT_MS = 120_000;
 // gpt-5.x are reasoning models: they take reasoning_effort, require max_completion_tokens
 // (not max_tokens), and reject a custom temperature. Map verikun's effort scale onto
 // OpenAI's (whose ceiling is 'high'), and only send it when the caller asked for one.
-// INVARIANT: every OpenAI model in cost.ts's registry is a reasoning model that accepts
-// reasoning_effort. If a non-reasoning model is ever added there, gate this send behind an
-// allowlist like claude.ts's EFFORT_MODELS (else it 400s → exit 2 with no retry).
 const EFFORT_MAP = {
     low: 'low',
     medium: 'medium',
@@ -37,6 +34,12 @@ const EFFORT_MAP = {
     xhigh: 'high',
     max: 'high',
 };
+// reasoning_effort is rejected outright by non-reasoning models (gpt-4.1), and a 400 is NOT
+// retried — it surfaces as exit 2 — so send the param only for models known to accept it,
+// mirroring claude.ts's EFFORT_MODELS. An ALLOWLIST on purpose: a model missing from here
+// merely forgoes effort, whereas a denylist would let the next non-reasoning model 400.
+// Keep in step with cost.ts's registry when adding an OpenAI model.
+const REASONING_MODELS = new Set(['gpt-5.4-mini', 'gpt-5.4', 'gpt-5.5']);
 const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
 const backoffMs = (attempt) => Math.min(1000 * 2 ** (attempt - 1), 15000);
 /** Map OpenAI's usage onto the normalized (Anthropic-shaped) Usage the CostTracker
@@ -162,7 +165,7 @@ class OpenAiProvider {
                 { role: 'user', content: user },
             ],
         };
-        if (this.opts.effort) {
+        if (this.opts.effort && REASONING_MODELS.has(this.opts.model)) {
             const mapped = EFFORT_MAP[this.opts.effort];
             if (mapped)
                 body.reasoning_effort = mapped;

package/dist/cli.js

@@ -1176,6 +1176,7 @@ async function runAiTest(file, opts, backend, platform, device) {
         (0, output_1.err)(`[ai] cost ceiling $${opts.maxCostUsd} reached during compile (${cost.summaryLine()}) — not running`);
         return {
             ok: false,
+            cached,
             costUsd: Number(cost.usd().toFixed(4)),
             costLine: cost.summaryLine(),
             modelRepairs: 0,
@@ -1267,6 +1268,7 @@ async function runAiTest(file, opts, backend, platform, device) {
     (0, output_1.err)(`[ai] estimated total cost: $${cost.usd().toFixed(4)}`);
     return {
         ok: result.ok,
+        cached,
         costUsd: Number(cost.usd().toFixed(4)),
         costLine,
         modelRepairs: result.modelRepairs,
@@ -1308,6 +1310,7 @@ async function cmdAi(positionals, flags) {
     if ((0, args_1.flagBool)(flags, 'json')) {
         (0, output_1.json)({
             ok: result.ok,
+            cached: result.cached,
             model: opts.model,
             cost: result.costLine,
             costUsd: result.costUsd,
@@ -1657,7 +1660,7 @@ AI (run a natural-language test — compile once, replay model-free, self-heal)
                                       path. The model is woken only to repair a step
                                       that fails to resolve; a green run persists the
                                       (repaired) plan so the next run is free. Needs
-                                      ANTHROPIC_API_KEY (Claude), OPENAI_API_KEY (gpt-5.x),
+                                      ANTHROPIC_API_KEY (Claude), OPENAI_API_KEY (gpt-*),
                                       or a logged-in agent CLI — no API key: --model
                                       codex-cli uses your 'codex login' ChatGPT
                                       subscription, cursor-cli your 'cursor-agent login'
@@ -1668,8 +1671,8 @@ AI (run a natural-language test — compile once, replay model-free, self-heal)
                                       running; --recompile ignores the cache.
                                       Models: claude-haiku-4-5 | claude-sonnet-4-6
                                       (default) | claude-opus-4-8 | claude-fable-5 |
-                                      gpt-5.4-mini | gpt-5.4 | gpt-5.5 | codex-cli |
-                                      cursor-cli.
+                                      gpt-5.4-mini | gpt-5.4 | gpt-5.5 | gpt-4.1 |
+                                      codex-cli | cursor-cli.
 
 SUITE (run a directory of natural-language tests as one gated suite)
   suite <dir> [--app <id>] [--name n] [--json]  (+ all \`ai\` flags, incl. --server)

package/dist/version.js

@@ -3,4 +3,4 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.VERSION = void 0;
 // GENERATED by scripts/gen-version.mjs from package.json's "version" at build time
 // (the `prebuild` script). Do NOT edit by hand; bump package.json instead.
-exports.VERSION = '0.12.0';
+exports.VERSION = '0.14.0';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "verikun",
-  "version": "0.12.0",
+  "version": "0.14.0",
   "description": "Drive Android emulators/devices and iOS simulators for AI agents: tap, type, swipe, screenshot, and inspect the UI hierarchy by semantic identifiers — like Puppeteer for native apps.",
   "keywords": [
     "android",