@aibridge/cli 0.1.1 → 0.3.0

package/README.md

@@ -15,7 +15,7 @@
 
 ---
 
-Your agent is one model, from one provider. Your machine probably has several more sitting behind CLIs you already use — `grok`, `agy` (Antigravity), `codex`, `claude`. **aibridge** turns them into seats your agent can drive: a planner that studies your repo, an implementer that edits it and runs your real tests, a reviewer from a *different* model family that cross-checks the diff against the plan, concurrent one-shot delegates — and capabilities your agent's own provider may not offer at all, like real image generation (gpt-image-2, Grok Imagine).
+Your agent is one model, from one provider. Your machine probably has several more sitting behind CLIs you already use — `grok`, `agy` (Antigravity), `codex`, `claude`. **aibridge** turns them into seats your agent can drive: a planner that studies your repo, an implementer that edits it and runs your real tests, a reviewer from a *different* model family that cross-checks the diff against the plan, concurrent one-shot delegates — and capabilities your agent's own provider may not offer at all, like real image generation (Codex, Antigravity, or Grok seats).
 
 ## Install
 
@@ -28,7 +28,7 @@ npx skills add ycmjason/aibridge
 That's it. The skill runs the CLI on demand via `npx -y @aibridge/cli` — nothing else to install. Ask your agent to "use aibridge", or try it yourself:
 
 ```bash
-npx -y @aibridge/cli subagent "summarize the architecture of this repo"
+npx -y @aibridge/cli subagent --model xai-grok/grok-4.5 "summarize the architecture of this repo"
 ```
 
 <sup>Want the `aibridge` command on your PATH for manual use? `npm i -g @aibridge/cli` (optional).</sup>
@@ -37,21 +37,21 @@ npx -y @aibridge/cli subagent "summarize the architecture of this repo"
 
 | Command | Use when |
 |---|---|
-| `aibridge plan "<task>"` | You want a delegate model to study the repo and expand a task into a detailed, reviewable **plan file** before any code is written |
-| `aibridge implement <plan.md>` | You have an approved plan file and want it executed in place — with your project's **real typecheck and tests** run until green |
-| `aibridge review [--plan <plan.md>]` | You want a **different model** to pressure-test the working-tree diff against the plan contract (over-reach is a finding) — or to review the plan itself before implementing |
-| `aibridge subagent "<task>"` | A self-contained task deserves a concurrent delegate, a cross-model second opinion, or a red-team pass |
-| `aibridge image-gen "<prompt>"` | You need a real raster image — gpt-image-2 (default) or Grok Imagine, with render verification |
+| `aibridge plan --model xai-grok/grok-4.5 --out plan.md "<task>"` | You want a delegate model to study the repo and expand a task into a detailed, reviewable **plan file** before any code is written |
+| `aibridge implement --model google-antigravity/gemini-3.6-flash <plan.md>` | You have an approved plan file and want it executed in place — with your project's **real typecheck and tests** run until green |
+| `aibridge review --model xai-grok/grok-4.5 --out review.md [--plan <plan.md>]` | You want a **different model** to pressure-test the working-tree diff against the plan contract (over-reach is a finding) — or to review the plan itself before implementing |
+| `aibridge subagent --model xai-grok/grok-4.5 "<task>"` | A self-contained task deserves a concurrent delegate, a cross-model second opinion, or a red-team pass |
+| `aibridge image-gen --model openai-codex/gpt-5.6-sol --out out.png "<prompt>"` | You need a real raster image — on a Codex, Antigravity, or Grok seat, with render verification |
 | `aibridge quota` | Two-second check of every backend's remaining quota before you pipeline work |
 | `aibridge runs` | Inspect or watch past delegation runs (`~/.aibridge/runs`) |
 
 The three verbs compose into an orchestrator-driven loop your agent stays in charge of:
 
 ```
-aibridge plan "add rate limiting to the API"   # delegate writes plan.md
+aibridge plan --model xai-grok/grok-4.5 --out plan.md "add rate limiting to the API"   # delegate writes plan.md
 # → your agent reads, edits, approves the plan
-aibridge implement plan.md                     # another model executes it, runs your gates
-aibridge review --plan plan.md                 # a third seat cross-checks the diff
+aibridge implement --model google-antigravity/gemini-3.6-flash plan.md                 # another model executes it, runs your gates
+aibridge review --model xai-grok/grok-4.5 --out review.md --plan plan.md               # a third seat cross-checks the diff
 ```
 
 Plan files — not their contents — travel between stages, so the loop is nearly free on your agent's context.
@@ -61,7 +61,7 @@ Plan files — not their contents — travel between stages, so the loop is near
 - **The skill carries judgment; the CLI owns execution.** The skill teaches your agent prompt-craft, seat selection, and when to gate; the CLI deterministically drives the backing CLIs, captures their output, verifies results (a "generated image" under 100 KB is a code-drawn fake, an empty answer is a quota death), and logs every run.
 - **Seats stay cross-model by default.** Grok plans and reviews, Gemini implements — a model never reviews its own diff, and independent eyes catch what shared blind spots miss.
 - **No API keys.** Delegation runs on the backing CLIs' existing logins, each spending its own quota. (The skill treats a backend that shares your agent's own quota pool as a last resort.)
-- **Models are canonical slugs**: `<vendor>-<cli>/<model>[-<effort>]` — e.g. `xai-grok/grok-4.5`, `google-antigravity/gemini-3.6-flash`, `openai-codex/gpt-5.6-sol-high`, `anthropic-claude/opus`. No aliases; `aibridge <command> --help` lists every seat.
+- **Models are canonical slugs**: `<vendor>-<cli>/<model>[-<effort>]` — e.g. `xai-grok/grok-4.5`, `google-antigravity/gemini-3.6-flash`, `openai-codex/gpt-5.6-sol-high`, `anthropic-claude/opus-5`. No aliases — not short ones, and not moving vendor aliases like `opus`: every seat pins an exact model version. `aibridge <command> --help` lists every seat.
 
 ## Requirements
 
@@ -72,6 +72,10 @@ Plan files — not their contents — travel between stages, so the loop is near
 
 Everything is published under the [`@aibridge`](https://www.npmjs.com/org/aibridge) scope: [`@aibridge/cli`](https://www.npmjs.com/package/@aibridge/cli) (the command), `@aibridge/proc` (spawn/capture), and one driver per backing CLI — `@aibridge/driver-agy`, `@aibridge/driver-grok`, `@aibridge/driver-codex`, `@aibridge/driver-claude` — reusable if you want to drive a single CLI from your own code.
 
+## Security
+
+aibridge executes real delegation — that's the product, and security scanners rightly notice: backing CLIs read/write files and run shell in tools mode, **at the same trust level as the agent you already run**. Nothing gains more access than you granted your agent and those CLIs when you installed them. Task content goes to the delegate's provider (use `--no-tools` for untrusted input — reasoning only, no file/shell access). All packages publish from this public repo via OIDC with [SLSA provenance](https://www.npmjs.com/package/@aibridge/cli), with no install-time scripts.
+
 ## Contributing & development
 
 Dev docs, architecture, and the working agreements for coding agents live in [`AGENTS.md`](AGENTS.md); design history in [`docs/`](docs/). Quick loop:

package/dist/cli.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { r as runCli, t as buildContext } from "./context-8JjDzZBG.mjs";
+import { r as runCli, t as buildContext } from "./context-B6I4QI9z.mjs";
 //#region src/cli.ts
 await runCli(buildContext(process), process.argv.slice(2));
 //#endregion

package/dist/{context-8JjDzZBG.mjs → context-B6I4QI9z.mjs}

@@ -3,7 +3,6 @@ import { ExitCode, buildApplication, buildCommand, buildRouteMap, run } from "@s
 import { appendFileSync, closeSync, copyFileSync, existsSync, mkdirSync, mkdtempSync, openSync, readFileSync, readSync, readdirSync, rmSync, statSync, writeFileSync } from "node:fs";
 import { homedir, tmpdir } from "node:os";
 import { isAbsolute, join, resolve } from "node:path";
-import { isNotFound, runCaptured } from "@aibridge/proc";
 import * as agy from "@aibridge/driver-agy";
 import { fetchAgyQuota, findModelQuota } from "@aibridge/driver-agy";
 import * as claude from "@aibridge/driver-claude";
@@ -11,6 +10,7 @@ import { fetchClaudeQuota } from "@aibridge/driver-claude";
 import * as codex from "@aibridge/driver-codex";
 import { fetchCodexQuota } from "@aibridge/driver-codex";
 import * as grok from "@aibridge/driver-grok";
+import { runCaptured } from "@aibridge/proc";
 import { randomBytes } from "node:crypto";
 //#region src/models.ts
 const MODELS = {
@@ -35,7 +35,7 @@ const MODELS = {
 			"high"
 		],
 		defaultEffort: "high",
-		brief: "Google Gemini 3.6 Flash via agy — default for implement; own Antigravity login"
+		brief: "Google Gemini 3.6 Flash via agy — default for implement, also image-gen; own Antigravity login"
 	},
 	"google-antigravity/claude-sonnet-4-6": {
 		slug: "google-antigravity/claude-sonnet-4-6",
@@ -70,10 +70,10 @@ const MODELS = {
 		],
 		brief: "OpenAI Codex gpt-5.6-sol via codex CLI"
 	},
-	"anthropic-claude/sonnet": {
-		slug: "anthropic-claude/sonnet",
+	"anthropic-claude/sonnet-5": {
+		slug: "anthropic-claude/sonnet-5",
 		backend: "claude",
-		backendModel: "sonnet",
+		backendModel: "claude-sonnet-5",
 		efforts: [
 			"low",
 			"medium",
@@ -81,12 +81,12 @@ const MODELS = {
 			"xhigh",
 			"max"
 		],
-		brief: "Claude Sonnet via claude CLI — bills your Claude subscription"
+		brief: "Claude Sonnet 5 via claude CLI — bills your Claude subscription"
 	},
-	"anthropic-claude/opus": {
-		slug: "anthropic-claude/opus",
+	"anthropic-claude/opus-5": {
+		slug: "anthropic-claude/opus-5",
 		backend: "claude",
-		backendModel: "opus",
+		backendModel: "claude-opus-5",
 		efforts: [
 			"low",
 			"medium",
@@ -95,15 +95,33 @@ const MODELS = {
 			"max"
 		],
 		defaultEffort: "high",
-		brief: "Claude Opus via claude CLI (default effort: high) — bills subscription"
+		brief: "Claude Opus 5 via claude CLI (default effort: high) — bills subscription"
+	},
+	"anthropic-claude/opus-5-1m": {
+		slug: "anthropic-claude/opus-5-1m",
+		backend: "claude",
+		backendModel: "claude-opus-5[1m]",
+		efforts: [
+			"low",
+			"medium",
+			"high",
+			"xhigh",
+			"max"
+		],
+		defaultEffort: "high",
+		brief: "Claude Opus 5, 1M context via claude CLI — long-context work; bills subscription"
 	}
 };
-const DEFAULT_MODEL = "xai-grok/grok-4.5";
-const DEFAULT_IMPLEMENTER = "google-antigravity/gemini-3.6-flash";
-const DEFAULT_IMAGE_GEN = "openai-codex/gpt-5.6-sol";
-const IMAGE_GEN_BACKENDS = /* @__PURE__ */ new Set(["codex", "grok"]);
+const IMAGE_GEN_FORMATS = /* @__PURE__ */ new Map([
+	["agy", "jpg"],
+	["codex", "png"],
+	["grok", "jpg"]
+]);
 function supportsImageGen(resolved) {
-	return IMAGE_GEN_BACKENDS.has(resolved.spec.backend);
+	return IMAGE_GEN_FORMATS.has(resolved.spec.backend);
+}
+function imageFormatFor(resolved) {
+	return IMAGE_GEN_FORMATS.get(resolved.spec.backend);
 }
 const EFFORTS_SET = /* @__PURE__ */ new Set([
 	"low",
@@ -132,7 +150,6 @@ function resolveModel(input) {
 	}
 }
 function backendModelId(resolved) {
-	if (!resolved.spec.backendModel) return;
 	if (resolved.spec.backend === "agy") {
 		const effort = resolved.effort ?? resolved.spec.defaultEffort;
 		if (effort) return `${resolved.spec.backendModel}-${effort}`;
@@ -142,7 +159,7 @@ function backendModelId(resolved) {
 function listModelHelpLines(opts = {}) {
 	const lines = [];
 	for (const [slug, spec] of Object.entries(MODELS)) {
-		if (opts.imageOnly && !IMAGE_GEN_BACKENDS.has(spec.backend)) continue;
+		if (opts.imageOnly && !IMAGE_GEN_FORMATS.has(spec.backend)) continue;
 		lines.push(`  ${slug}`);
 		lines.push(`    ${spec.brief}`);
 	}
@@ -186,7 +203,8 @@ const DRIVERS = {
 	agy: {
 		probe: () => agy.probe(),
 		run: (task) => agy.run(task),
-		quota: () => agy.fetchAgyQuota()
+		quota: () => agy.fetchAgyQuota(),
+		generateImage: (req) => agy.generateImage(req)
 	},
 	grok: {
 		probe: () => grok.probe(),
@@ -213,38 +231,29 @@ function getDriver(backend) {
 //#endregion
 //#region src/commands/image-gen/impl.ts
 const MIN_REAL_BYTES_CODEX = 1e5;
-const MIN_REAL_BYTES_GROK = 1e4;
+const MIN_REAL_BYTES_TOOL = 1e4;
 async function imageGen$1(flags, prompt) {
 	const fail = (msg) => {
 		this.process.stderr.write(`aibridge image-gen: ${msg}\n`);
 		this.process.exitCode = 1;
 	};
-	const inputSlug = flags.model ?? "openai-codex/gpt-5.6-sol";
+	const inputSlug = flags.model;
 	const model = resolveModel(inputSlug);
 	if (!model) return fail(formatUnknownModelError(inputSlug));
 	if (!supportsImageGen(model)) return fail(formatImageGenModelError(inputSlug, model));
-	if (model.spec.backend === "codex" && model.effort) return fail(`effort "-${model.effort}" has no effect on image-gen (gpt-image-2 renders, not the seat model); use ${DEFAULT_IMAGE_GEN}.`);
-	const quality = (flags.quality ?? "high").toLowerCase();
-	if (![
-		"low",
-		"medium",
-		"high"
-	].includes(quality)) return fail(`invalid --quality "${flags.quality}" (use low | medium | high)`);
-	let size;
-	if (flags.size !== void 0) {
-		const m = flags.size.match(/^(\d+)\s*x\s*(\d+)$/i);
-		if (!m) return fail(`invalid --size "${flags.size}" (expected e.g. 1024x1024)`);
-		size = {
-			w: Number(m[1]),
-			h: Number(m[2])
-		};
-		if (model.spec.backend === "codex") {
-			const constraint = sizeConstraintError(size.w, size.h);
-			if (constraint) return fail(`invalid --size ${size.w}x${size.h}: ${constraint}`);
-		} else if (size.w < 1 || size.h < 1) return fail(`invalid --size ${size.w}x${size.h}: dimensions must be positive`);
+	if (model.spec.backend === "codex" && model.effort) return fail(`effort "-${model.effort}" has no effect on image-gen (the image tool renders, not the seat model); pass the un-suffixed slug "${model.spec.slug}" instead.`);
+	const expected = imageFormatFor(model);
+	if (expected === void 0) return fail(formatImageGenModelError(inputSlug, model));
+	const label = expected === "png" ? "PNG" : "JPEG";
+	if (!(expected === "png" ? /\.png$/i.test(flags.out) : /\.jpe?g$/i.test(flags.out))) return fail(`--out "${flags.out}" must end in ${expected === "png" ? ".png" : ".jpg or .jpeg"} — the ${model.spec.slug} seat renders ${label} and aibridge does not convert.`);
+	let aspectRatio;
+	if (flags.aspectRatio !== void 0) {
+		const m = flags.aspectRatio.match(/^(\d+)\s*:\s*(\d+)$/);
+		if (!m || Number(m[1]) < 1 || Number(m[2]) < 1) return fail(`invalid --aspect-ratio "${flags.aspectRatio}" (expected e.g. 16:9)`);
+		aspectRatio = `${Number(m[1])}:${Number(m[2])}`;
 	}
 	const timeoutSec = flags.timeout ?? 600;
-	const outPath = resolve(this.process.cwd(), flags.out ?? "./aibridge-image.png");
+	const outPath = resolve(this.process.cwd(), flags.out);
 	const imagePaths = [];
 	if (flags.image !== void 0) for (const raw of flags.image.split(",").map((s) => s.trim()).filter(Boolean)) {
 		const abs = resolve(this.process.cwd(), raw);
@@ -253,16 +262,15 @@ async function imageGen$1(flags, prompt) {
 	}
 	const driver = getDriver(model.spec.backend);
 	if (!driver.generateImage) return fail(formatImageGenModelError(inputSlug, model));
-	const minBytes = model.spec.backend === "codex" ? MIN_REAL_BYTES_CODEX : MIN_REAL_BYTES_GROK;
+	const minBytes = model.spec.backend === "codex" ? MIN_REAL_BYTES_CODEX : MIN_REAL_BYTES_TOOL;
 	const work = mkdtempSync(join(tmpdir(), "aibridge-imagegen-"));
 	try {
 		let outcome = await driver.generateImage({
 			prompt,
 			workDir: work,
-			backendModel: model.spec.backendModel,
+			backendModel: backendModelId(model),
 			effort: model.effort,
-			quality,
-			size,
+			aspectRatio,
 			imagePaths,
 			timeoutSec,
 			forceful: false,
@@ -271,10 +279,9 @@ async function imageGen$1(flags, prompt) {
 		if (model.spec.backend === "codex" && outcome.kind === "suspect") outcome = await driver.generateImage({
 			prompt,
 			workDir: work,
-			backendModel: model.spec.backendModel,
+			backendModel: backendModelId(model),
 			effort: model.effort,
-			quality,
-			size,
+			aspectRatio,
 			imagePaths,
 			timeoutSec,
 			forceful: true,
@@ -282,34 +289,20 @@ async function imageGen$1(flags, prompt) {
 		});
 		if (outcome.kind === "ok" && outcome.bytes < minBytes) outcome = { kind: "suspect" };
 		if (outcome.kind === "error") return fail(outcome.reason);
-		if (outcome.kind === "suspect") return fail(model.spec.backend === "grok" ? "grok produced no usable image. Check SuperGrok image access and re-run with a simpler prompt." : "codex produced only a tiny/code-drawn image, not a real gpt-image-2 render. Try --quality high or a clearer, simpler prompt.");
+		if (outcome.kind === "suspect") return fail(model.spec.backend === "agy" ? "agy produced no usable image. Re-run with a simpler prompt, or check Antigravity image access." : model.spec.backend === "grok" ? "grok produced no usable image. Check SuperGrok image access and re-run with a simpler prompt." : "codex produced only a tiny/code-drawn image, not a real render. Try a clearer, simpler prompt.");
 		const local = join(work, "result.bin");
 		copyFileSync(outcome.path, local);
-		let dims = imageSize(local);
-		if (size && dims && (dims.width !== size.w || dims.height !== size.h)) if (await magick([
-			local,
-			"-resize",
-			`${size.w}x${size.h}!`,
-			local
-		])) dims = imageSize(local) ?? dims;
-		else this.process.stderr.write(`aibridge image-gen: rendered ${dims.width}x${dims.height}, wanted ${size.w}x${size.h}, and ImageMagick (magick/convert) is unavailable to resize.
-`);
-		const outExt = /\.png$/i.test(outPath) ? "png" : /\.jpe?g$/i.test(outPath) ? "jpg" : null;
-		const actualFmt = pngSize(local) ? "png" : jpegSize(local) ? "jpg" : null;
-		const needsConvert = outExt !== null && actualFmt !== null && outExt !== actualFmt;
-		if (!needsConvert || !await magick([local, outPath])) {
-			if (needsConvert) this.process.stderr.write(`aibridge image-gen: render is ${actualFmt.toUpperCase()} but out path wants ${outExt.toUpperCase()}, and ImageMagick (magick/convert) is unavailable to convert; writing the raw bytes as-is.
-`);
-			copyFileSync(local, outPath);
-		}
+		const dims = imageSize(local);
+		const actual = pngSize(local) ? "png" : jpegSize(local) ? "jpg" : null;
+		if (actual !== null && actual !== expected) this.process.stderr.write(`aibridge image-gen: expected a ${label} render from this seat but got ${actual === "png" ? "PNG" : "JPEG"}; wrote the raw bytes to ${outPath} anyway — the extension does not match the contents.\n`);
+		copyFileSync(local, outPath);
 		const bytes = statSync(outPath).size;
 		if (flags.json) this.process.stdout.write(`${JSON.stringify({
 			out: outPath,
 			bytes,
 			width: dims?.width ?? null,
 			height: dims?.height ?? null,
-			sizeRequested: flags.size ?? null,
-			quality: model.spec.backend === "codex" ? quality : null,
+			aspectRatio: flags.aspectRatio ?? null,
 			model: model.spec.slug,
 			backend: model.spec.backend,
 			real: true
@@ -317,8 +310,7 @@ async function imageGen$1(flags, prompt) {
 		else {
 			const kb = Math.round(bytes / 1024);
 			const dimStr = dims ? `${dims.width}x${dims.height}, ` : "";
-			const qualityStr = model.spec.backend === "codex" ? `, ${quality} quality` : "";
-			this.process.stdout.write(`✓ Wrote ${outPath} (${dimStr}${kb} KB${qualityStr}, ${model.spec.slug})\n`);
+			this.process.stdout.write(`✓ Wrote ${outPath} (${dimStr}${kb} KB, ${model.spec.slug})\n`);
 		}
 	} finally {
 		rmSync(work, {
@@ -327,15 +319,6 @@ async function imageGen$1(flags, prompt) {
 		});
 	}
 }
-function sizeConstraintError(w, h) {
-	if (w % 16 !== 0 || h % 16 !== 0) return "each edge must be divisible by 16";
-	const long = Math.max(w, h);
-	if (long / Math.min(w, h) > 3) return "aspect ratio must be within 1:3–3:1";
-	if (long > 3840) return "longest edge must be <= 3840px";
-	const px = w * h;
-	if (px < 655360 || px > 8294400) return "total pixels must be 655,360–8,294,400";
-	return null;
-}
 function imageSize(path) {
 	return pngSize(path) ?? jpegSize(path);
 }
@@ -384,24 +367,15 @@ function jpegSize(path) {
 		return null;
 	}
 }
-async function magick(args) {
-	for (const tool of ["magick", "convert"]) try {
-		const r = await runCaptured(tool, [...args], { timeoutMs: 6e4 });
-		if (!r.timedOut && r.code === 0) return true;
-	} catch (err) {
-		if (!isNotFound(err)) throw err;
-	}
-	return false;
-}
 //#endregion
 //#region src/commands/image-gen/command.ts
 const fullDescription$6 = [
-	"Renders an image by driving the seat's CLI (codex → gpt-image-2, grok →",
-	"Imagine), then verifies the result is a real render before returning it.",
+	"Renders an image by driving the seat's own CLI, then verifies the result is",
+	"a real render before returning it.",
 	"",
 	"Image-gen seats (canonical slug):",
 	...listModelHelpLines({ imageOnly: true }),
-	`Default: ${DEFAULT_IMAGE_GEN} (gpt-image-2 via codex; historical default).`
+	"Recommended seat: openai-codex/gpt-5.6-sol."
 ].join("\n");
 const imageGen = buildCommand({
 	func: imageGen$1,
@@ -410,20 +384,18 @@ const imageGen = buildCommand({
 			model: {
 				kind: "parsed",
 				parse: String,
-				optional: true,
-				brief: `Model slug (default: ${DEFAULT_IMAGE_GEN})`
+				brief: "Model slug (required) — see the seat list above"
 			},
 			out: {
 				kind: "parsed",
 				parse: String,
-				optional: true,
-				brief: "Path to write the image (default: ./aibridge-image.png)"
+				brief: "Path to write the image — extension must match the seat format (.png for codex, .jpg for agy/grok)"
 			},
-			size: {
+			aspectRatio: {
 				kind: "parsed",
 				parse: String,
 				optional: true,
-				brief: "WIDTHxHEIGHT (codex: each edge ÷16; grok: mapped to aspect_ratio, then optionally resized)"
+				brief: "Aspect ratio N:M, e.g. 16:9 (agy/grok: real tool param; codex: prompt hint)"
 			},
 			image: {
 				kind: "parsed",
@@ -431,12 +403,6 @@ const imageGen = buildCommand({
 				optional: true,
 				brief: "Reference image path(s), comma-separated — visual reference"
 			},
-			quality: {
-				kind: "parsed",
-				parse: String,
-				optional: true,
-				brief: "low | medium | high (codex/gpt-image-2; default high)"
-			},
 			timeout: {
 				kind: "parsed",
 				parse: positiveIntSeconds,
@@ -473,7 +439,7 @@ async function delegate(opts, driver = getDriver(opts.model.spec.backend)) {
 		tools: opts.tools,
 		timeoutSec: opts.timeoutSec,
 		cwd: opts.cwd,
-		backendModel: backendModelId(opts.model) ?? opts.model.spec.backendModel,
+		backendModel: backendModelId(opts.model),
 		effort: opts.model.effort,
 		onStdout: (c) => opts.run.stdout(c),
 		onStderr: (c) => opts.run.stderr(c),
@@ -526,7 +492,7 @@ async function preflightModel(resolved) {
 	if (resolved.spec.backend === "codex") return preflightCodex();
 	if (resolved.spec.backend !== "agy") return { ok: true };
 	try {
-		return evaluateAgyPreflight(await fetchAgyQuota(), backendModelId(resolved) ?? "");
+		return evaluateAgyPreflight(await fetchAgyQuota(), backendModelId(resolved));
 	} catch (err) {
 		return {
 			ok: true,
@@ -679,7 +645,7 @@ function readRunLogs(id) {
 //#endregion
 //#region src/commands/implement/impl.ts
 async function implement$1(flags, planFile) {
-	const inputSlug = flags.model ?? "google-antigravity/gemini-3.6-flash";
+	const inputSlug = flags.model;
 	const model = resolveModel(inputSlug);
 	if (!model) {
 		this.process.stderr.write(`${formatUnknownModelError(inputSlug)}\n`);
@@ -749,8 +715,7 @@ const implement = buildCommand({
 			model: {
 				kind: "parsed",
 				parse: String,
-				optional: true,
-				brief: `Model slug (default: ${DEFAULT_IMPLEMENTER})`
+				brief: "Model slug (required) — see the seat list above"
 			},
 			timeout: {
 				kind: "parsed",
@@ -811,7 +776,7 @@ function extractPathFromPorcelainLine(line) {
 	return content;
 }
 async function plan$1(flags, taskPrompt) {
-	const inputSlug = flags.model ?? "xai-grok/grok-4.5";
+	const inputSlug = flags.model;
 	const model = resolveModel(inputSlug);
 	if (!model) {
 		this.process.stderr.write(`${formatUnknownModelError(inputSlug)}\n`);
@@ -831,7 +796,7 @@ async function plan$1(flags, taskPrompt) {
 	const cwd = this.process.cwd();
 	const promptSnippet = taskPrompt.replace(/\r?\n/g, " ").slice(0, 80);
 	const run = startRun("plan", `${model.spec.slug}: ${promptSnippet}`);
-	const absOutPath = flags.out ? isAbsolute(flags.out) ? flags.out : resolve(cwd, flags.out) : resolve(run.dir, "plan.md");
+	const absOutPath = isAbsolute(flags.out) ? flags.out : resolve(cwd, flags.out);
 	const beforePorcelain = await getPorcelainStatus(cwd);
 	const outcome = await delegate({
 		model,
@@ -896,14 +861,12 @@ const plan = buildCommand({
 			model: {
 				kind: "parsed",
 				parse: String,
-				optional: true,
-				brief: `Model slug (default: ${DEFAULT_MODEL})`
+				brief: "Model slug (required) — see the seat list above"
 			},
 			out: {
 				kind: "parsed",
 				parse: String,
-				optional: true,
-				brief: "Where to write the plan (default: <run.dir>/plan.md)"
+				brief: "Where to write the plan file (required)"
 			},
 			timeout: {
 				kind: "parsed",
@@ -1062,7 +1025,7 @@ function parseReviewVerdict(response) {
 	};
 }
 async function review$1(flags) {
-	const inputSlug = flags.model ?? "xai-grok/grok-4.5";
+	const inputSlug = flags.model;
 	const model = resolveModel(inputSlug);
 	if (!model) {
 		this.process.stderr.write(`${formatUnknownModelError(inputSlug)}\n`);
@@ -1112,7 +1075,7 @@ async function review$1(flags) {
 	const timeoutSec = flags.timeout ?? 1200;
 	const modeDetail = isDirty ? absPlanPath ? `diff + plan (${absPlanPath})` : `diff (${baseRef})` : `plan-only (${absPlanPath})`;
 	const run = startRun("review", `${model.spec.slug}: ${modeDetail}`);
-	const absOutPath = flags.out ? isAbsolute(flags.out) ? flags.out : resolve(cwd, flags.out) : resolve(run.dir, "review.md");
+	const absOutPath = isAbsolute(flags.out) ? flags.out : resolve(cwd, flags.out);
 	let reviewPrompt;
 	if (isDirty) reviewPrompt = `You are an expert code reviewer. Inspect the working tree diff against base '${baseRef}' and untracked files at ${cwd}.\n` + (absPlanPath ? `Compare the implementation against the plan contract at ${absPlanPath}. Any file modified or feature added outside the plan contract counts as over-reach (severity: major unless harmful, then critical).\n` : "") + `Write your detailed review report to the file ${absOutPath}. For each finding, include file:line, severity (critical|major|minor), and rationale.\nYour final answer (last message) must consist of EXACTLY ONE VERDICT LINE:\nEither: "PASS"\nOr: "FINDINGS: <c> critical, <m> major, <n> minor"`;
 	else reviewPrompt = `You are an expert architecture reviewer. Inspect the plan contract file at ${absPlanPath}.\nReview the plan for soundness, missing edge cases, safety, and feasibility.\nWrite your detailed review report to the file ${absOutPath}. For each finding, include severity (critical|major|minor) and rationale.\nYour final answer (last message) must consist of EXACTLY ONE VERDICT LINE:\nEither: "PASS"\nOr: "FINDINGS: <c> critical, <m> major, <n> minor"`;
@@ -1164,8 +1127,7 @@ const review = buildCommand({
 		model: {
 			kind: "parsed",
 			parse: String,
-			optional: true,
-			brief: `Model slug (default: ${DEFAULT_MODEL})`
+			brief: "Model slug (required) — see the seat list above"
 		},
 		plan: {
 			kind: "parsed",
@@ -1182,8 +1144,7 @@ const review = buildCommand({
 		out: {
 			kind: "parsed",
 			parse: String,
-			optional: true,
-			brief: "Where to write the review report (default: <run.dir>/review.md)"
+			brief: "Where to write the review report (required)"
 		},
 		timeout: {
 			kind: "parsed",
@@ -1375,7 +1336,7 @@ const runs = buildCommand({
 //#endregion
 //#region src/commands/subagent/impl.ts
 async function subagent$1(flags, prompt) {
-	const inputSlug = flags.model ?? "xai-grok/grok-4.5";
+	const inputSlug = flags.model;
 	const model = resolveModel(inputSlug);
 	if (!model) {
 		this.process.stderr.write(`${formatUnknownModelError(inputSlug)}\n`);
@@ -1414,15 +1375,13 @@ async function subagent$1(flags, prompt) {
 		this.process.exitCode = 1;
 		return;
 	}
-	if (flags.json) {
-		const modelId = backendModelId(model) ?? null;
-		this.process.stdout.write(`${JSON.stringify({
-			model: modelId,
-			slug: model.spec.slug,
-			response: outcome.response,
-			exitCode: outcome.exitCode
-		})}\n`);
-	} else this.process.stdout.write(`${outcome.response}\n`);
+	if (flags.json) this.process.stdout.write(`${JSON.stringify({
+		model: backendModelId(model),
+		slug: model.spec.slug,
+		response: outcome.response,
+		exitCode: outcome.exitCode
+	})}\n`);
+	else this.process.stdout.write(`${outcome.response}\n`);
 }
 //#endregion
 //#region src/commands/subagent/command.ts
@@ -1431,7 +1390,7 @@ const fullDescription = [
 	"",
 	"Available models (canonical slug):",
 	...listModelHelpLines(),
-	`Default: ${DEFAULT_MODEL} (off-budget). The claude-backend slugs are FALLBACKS for`,
+	"Recommended first choice: xai-grok/grok-4.5 (off-budget). The claude-backend slugs are FALLBACKS for",
 	"when the off-budget CLIs are quota-exhausted — they bill your Claude subscription."
 ].join("\n");
 const subagent = buildCommand({
@@ -1441,8 +1400,7 @@ const subagent = buildCommand({
 			model: {
 				kind: "parsed",
 				parse: String,
-				optional: true,
-				brief: `Model slug to delegate to (default: ${DEFAULT_MODEL})`
+				brief: "Model slug (required) — see the seat list above"
 			},
 			timeout: {
 				kind: "parsed",
@@ -1509,7 +1467,7 @@ const app = buildApplication(buildRouteMap({
 		runs,
 		quota
 	},
-	docs: { brief: "Bridge tasks to non-Claude AI CLIs — a plan → implement → review workflow, task delegation, and image generation (codex gpt-image-2 / grok Imagine)." }
+	docs: { brief: "Bridge tasks to non-Claude AI CLIs — a plan → implement → review workflow, task delegation, and image generation (codex / agy / grok seats)." }
 }), {
 	name: "aibridge",
 	versionInfo: { currentVersion: version },
@@ -1526,4 +1484,4 @@ function buildContext(process) {
 	return { process };
 }
 //#endregion
-export { listModelHelpLines as C, formatUnknownModelError as S, supportsImageGen as T, DEFAULT_IMPLEMENTER as _, readRunLogs as a, backendModelId as b, evaluateCodexPreflight as c, renderPreflightRefusal as d, delegate as f, DEFAULT_IMAGE_GEN as g, positiveIntSeconds as h, listRuns as i, preflightCodex as l, nonEmptyPrompt as m, app as n, startRun as o, getDriver as p, runCli as r, evaluateAgyPreflight as s, buildContext as t, preflightModel as u, DEFAULT_MODEL as v, resolveModel as w, formatImageGenModelError as x, MODELS as y };
+export { supportsImageGen as S, backendModelId as _, readRunLogs as a, listModelHelpLines as b, evaluateCodexPreflight as c, renderPreflightRefusal as d, delegate as f, MODELS as g, positiveIntSeconds as h, listRuns as i, preflightCodex as l, nonEmptyPrompt as m, app as n, startRun as o, getDriver as p, runCli as r, evaluateAgyPreflight as s, buildContext as t, preflightModel as u, formatImageGenModelError as v, resolveModel as x, formatUnknownModelError as y };

package/dist/index.d.mts

@@ -20,14 +20,16 @@ declare function runCli(ctx: LocalContext, argv: readonly string[]): Promise<voi
  *
  * Models are registered by canonical, provider-qualified slug —
  * `<vendor>-<cli>/<model>[-<effort>]`, e.g. `openai-codex/gpt-5.6-sol-high`.
- * Canonical slugs only — no short aliases, by design.
+ * Canonical slugs only — no short aliases, by design. That holds on both sides:
+ * `backendModel` is a pinned model id (`claude-opus-5`), never a moving vendor
+ * alias (`opus`), so a seat never silently changes model under you.
  */
 type Backend = 'agy' | 'claude' | 'codex' | 'grok';
 type Effort = 'low' | 'medium' | 'high' | 'xhigh' | 'max';
 interface ModelSpec {
   readonly slug: string;
   readonly backend: Backend;
-  readonly backendModel: string | undefined;
+  readonly backendModel: string;
   readonly efforts: readonly Effort[] | null;
   readonly defaultEffort?: Effort;
   readonly brief: string;
@@ -37,12 +39,9 @@ interface ResolvedModel {
   readonly effort: Effort | undefined;
 }
 declare const MODELS: Record<string, ModelSpec>;
-declare const DEFAULT_MODEL = "xai-grok/grok-4.5";
-declare const DEFAULT_IMPLEMENTER = "google-antigravity/gemini-3.6-flash";
-declare const DEFAULT_IMAGE_GEN = "openai-codex/gpt-5.6-sol";
 declare function supportsImageGen(resolved: ResolvedModel): boolean;
 declare function resolveModel(input: string): ResolvedModel | undefined;
-declare function backendModelId(resolved: ResolvedModel): string | undefined;
+declare function backendModelId(resolved: ResolvedModel): string;
 declare function listModelHelpLines(opts?: {
   readonly imageOnly?: boolean;
 }): string[];
@@ -62,7 +61,7 @@ interface DelegationTask {
   readonly tools: boolean;
   readonly timeoutSec: number;
   readonly cwd: string;
-  readonly backendModel: string | undefined;
+  readonly backendModel: string;
   readonly effort: Effort | undefined;
   readonly onStdout?: (chunk: string) => void;
   readonly onStderr?: (chunk: string) => void;
@@ -82,13 +81,9 @@ type QuotaSnapshot = AgyQuotaSnapshot | CodexQuotaSnapshot | ClaudeQuotaSnapshot
 interface ImageGenRequest {
   readonly prompt: string;
   readonly workDir: string;
-  readonly backendModel: string | undefined;
+  readonly backendModel: string;
   readonly effort: Effort | undefined;
-  readonly quality: string;
-  readonly size: {
-    readonly w: number;
-    readonly h: number;
-  } | undefined;
+  readonly aspectRatio: string | undefined;
   readonly imagePaths: readonly string[];
   readonly timeoutSec: number;
   readonly forceful: boolean;
@@ -181,4 +176,4 @@ declare function renderPreflightRefusal(cmd: string, verdict: {
   resetAt: string | undefined;
 }): string;
 //#endregion
-export { type AgentCliDriver, type Backend, DEFAULT_IMAGE_GEN, DEFAULT_IMPLEMENTER, DEFAULT_MODEL, type DelegateOptions, type DelegateOutcome, type DelegationResult, type DelegationTask, type Effort, type LocalContext, MODELS, type ModelSpec, type PreflightVerdict, type ResolvedModel, type RunLog, type RunMeta, app, backendModelId, buildContext, delegate, evaluateAgyPreflight, evaluateCodexPreflight, formatImageGenModelError, formatUnknownModelError, getDriver, listModelHelpLines, listRuns, nonEmptyPrompt, positiveIntSeconds, preflightCodex, preflightModel, readRunLogs, renderPreflightRefusal, resolveModel, runCli, startRun, supportsImageGen };
+export { type AgentCliDriver, type Backend, type DelegateOptions, type DelegateOutcome, type DelegationResult, type DelegationTask, type Effort, type LocalContext, MODELS, type ModelSpec, type PreflightVerdict, type ResolvedModel, type RunLog, type RunMeta, app, backendModelId, buildContext, delegate, evaluateAgyPreflight, evaluateCodexPreflight, formatImageGenModelError, formatUnknownModelError, getDriver, listModelHelpLines, listRuns, nonEmptyPrompt, positiveIntSeconds, preflightCodex, preflightModel, readRunLogs, renderPreflightRefusal, resolveModel, runCli, startRun, supportsImageGen };