diff --git a/.release/cli-boolean-negation.md b/.release/cli-boolean-negation.md
new file mode 100644
index 0000000..6934e8f
--- /dev/null
+++ b/.release/cli-boolean-negation.md
@@ -0,0 +1,4 @@
+---
+bump: minor
+---
+`parseArgs` generates a `--no-<field>` negation for every boolean input, so a `z.boolean().default(true)` can be turned off from the CLI without declaring a second override flag. The negation takes no value and consumes no token (`--no-loud=false` is rejected; `myverb --no-loud alice` still reads `alice` as a positional), and booleans stay scalar — `--loud --no-loud` is last-wins. The derived names live in a flag-only set: `no-<field>` is not an input key, so `positionals: ["no-loud"]` remains a spec error, and an input field that would shadow a generated negation is rejected as one. `toHelp` renders booleans as `--loud / --no-loud` and now prints each field's default; a defaulted field is no longer also labelled `(required)`, which `z.toJSONSchema` had made it look like.
diff --git a/README.md b/README.md
index 5813b72..e46566f 100644
--- a/README.md
+++ b/README.md
@@ -92,6 +92,33 @@ More generally: any argument that maps to nothing the verb declares — an unkno
 flag, an extra positional, or a positional naming no field — throws. Nothing
 falls through to a default.
 
+### Boolean flags and `--no-`
+
+Every boolean input gets both spellings: `--loud` sets it true, `--no-loud` sets
+it false. So a boolean that is *on by default* can be turned off from the command
+line with the one field that declares it:
+
+```ts
+input: z.object({ changedOnly: z.boolean().default(true) }),
+// `check`                     →  changedOnly: true   (the default)
+// `check --no-changedOnly`    →  changedOnly: false
+```
+
+- **The negation is a derived CLI name, not an input field.** `no-changedOnly` is
+  not a key of `input`, so it is not selectable as a positional —
+  `positionals: ["no-changedOnly"]` remains a spec error. An input field that
+  would shadow a generated negation (a boolean `loud` beside a field literally
+  named `no-loud`) is a spec error too, rather than a flag that quietly means
+  only one of the two things.
+- **It takes no value and consumes no token.** `--no-loud=false` is rejected, and
+  `myverb --no-loud alice` reads `alice` as a positional, not as the flag's value.
+- **Booleans stay scalar.** `--loud --no-loud` is last-wins, the same rule every
+  other scalar flag follows.
+- **Both defaults get one.** `--no-` is generated for every boolean, not only the
+  `default(true)` ones — which spelling an author needs follows from the default,
+  and a default is a value the verb may change. Generating it conditionally would
+  mean flipping a default silently deletes a flag from every script using it.
+
 ## Design
 
 - **One spec, four surfaces.** The CLI, MCP server, Anthropic tool schema, and
diff --git a/src/__tests__/index.test.ts b/src/__tests__/index.test.ts
index 77d3fae..f922bdc 100644
--- a/src/__tests__/index.test.ts
+++ b/src/__tests__/index.test.ts
@@ -1,7 +1,14 @@
 import { describe, expect, test } from "bun:test";
 import { z } from "zod";
 
-import { defineVerb, dispatch, parseArgs, toMcpTool, toOpenApiPaths } from "@bounded-systems/verbspec";
+import {
+  defineVerb,
+  dispatch,
+  parseArgs,
+  toHelp,
+  toMcpTool,
+  toOpenApiPaths,
+} from "@bounded-systems/verbspec";
 
 // parseArgs CLI-isms for array-typed input fields: repeated flags accumulate,
 // comma-separated values split, and the two forms compose. Scalars take the
@@ -313,3 +320,155 @@ describe("dispatch multi-word ids", () => {
     expect(out).toMatchObject({ kind: "ok", id: "plan", output: { kind: "plan" } });
   });
 });
+
+// ── boolean negation (#12) ───────────────────────────────────────────────────
+// `z.boolean().default(true)` had no way to be turned OFF from the CLI: only
+// presence-is-true worked, so "on by default, opt out on the CLI" forced authors
+// to declare a SECOND override flag and compute the effective value by hand —
+// two flags for one boolean dimension. `--no-<field>` is that missing spelling.
+const negVerb = defineVerb({
+  id: "neg-probe",
+  summary: "test-only verb with a default-true boolean",
+  actor: "work",
+  positionals: [],
+  input: z.object({
+    changedOnly: z.boolean().default(true),
+    dryRun: z.boolean().default(false),
+    slug: z.string().optional(),
+  }),
+  output: z.object({}),
+  run: () => ({}),
+});
+
+const negPos = defineVerb({
+  id: "neg-pos-probe",
+  summary: "test-only verb with a positional beside a boolean",
+  actor: "work",
+  positionals: ["name"],
+  input: z.object({ name: z.string(), loud: z.boolean().default(true) }),
+  output: z.object({}),
+  run: () => ({}),
+});
+
+describe("parseArgs boolean negation", () => {
+  test("--no-<field> turns a default(true) boolean off", () => {
+    expect(parseArgs(negVerb, ["--no-changedOnly"])).toMatchObject({ changedOnly: false });
+  });
+
+  test("the default still applies when the negation is absent", () => {
+    expect(parseArgs(negVerb, [])).toMatchObject({ changedOnly: true });
+  });
+
+  test("every boolean gets one, not only the default(true) ones", () => {
+    expect(parseArgs(negVerb, ["--dryRun", "--no-dryRun"])).toMatchObject({ dryRun: false });
+  });
+
+  test("a boolean stays scalar — the last spelling wins in either order", () => {
+    expect(parseArgs(negVerb, ["--no-changedOnly", "--changedOnly"])).toMatchObject({
+      changedOnly: true,
+    });
+    expect(parseArgs(negVerb, ["--changedOnly", "--no-changedOnly"])).toMatchObject({
+      changedOnly: false,
+    });
+  });
+
+  test("the negation carries its own value — it never eats the next token", () => {
+    expect(parseArgs(negVerb, ["--no-changedOnly", "--slug", "abc"])).toMatchObject({
+      changedOnly: false,
+      slug: "abc",
+    });
+    // The sharper case: a bare token after the negation is still a positional,
+    // not the flag's value.
+    expect(parseArgs(negPos, ["--no-loud", "alice"])).toMatchObject({
+      name: "alice",
+      loud: false,
+    });
+  });
+
+  test("--no-<field>=value is rejected rather than silently ignored", () => {
+    expect(() => parseArgs(negVerb, ["--no-changedOnly=false"])).toThrow(/takes no value/i);
+  });
+
+  test("only booleans get a negation — --no-<string field> is an unknown flag", () => {
+    expect(() => parseArgs(negVerb, ["--no-slug", "x"])).toThrow(/unknown flag|unrecognized/i);
+  });
+
+  test("the negation of an undeclared field is an unknown flag", () => {
+    expect(() => parseArgs(negVerb, ["--no-bogus"])).toThrow(/unknown flag|unrecognized/i);
+  });
+
+  test("the unknown-flag message lists the negations among the valid flags", () => {
+    expect(() => parseArgs(negVerb, ["--nope"])).toThrow(/--no-changedOnly/);
+  });
+
+  test("a negation reaches the same field through dispatch", async () => {
+    const out = await dispatch({ "neg-probe": negVerb }, ["neg-probe", "--no-changedOnly"]);
+    expect(out).toMatchObject({ kind: "ok", input: { changedOnly: false } });
+  });
+});
+
+// The negation names are DERIVED CLI spellings, not input fields. Widening the
+// one set that gated both questions would have made `positionals: ["no-loud"]`
+// legal — it binds a value `input.parse` then strips, the silent-drop this
+// package exists to refuse. The two sets stay separate; these hold that line.
+describe("parseArgs boolean negation — derived names are not input fields", () => {
+  test("a negation name in `positionals` is still a spec error", () => {
+    const negAsPositional = defineVerb({
+      id: "neg-as-positional-probe",
+      summary: "test-only verb naming a negation as a positional",
+      actor: "work",
+      positionals: ["no-loud"],
+      input: z.object({ loud: z.boolean().default(true) }),
+      output: z.object({}),
+      run: () => ({}),
+    });
+    expect(() => parseArgs(negAsPositional, [])).toThrow(/no input field/i);
+    expect(() => parseArgs(negAsPositional, [])).toThrow(/no-loud/);
+  });
+
+  test("an input field colliding with a generated negation is a spec error", () => {
+    const collide = defineVerb({
+      id: "neg-collision-probe",
+      summary: "test-only verb whose field shadows a generated negation",
+      actor: "work",
+      input: z.object({ loud: z.boolean().default(true), "no-loud": z.string().optional() }),
+      output: z.object({}),
+      run: () => ({}),
+    });
+    expect(() => parseArgs(collide, [])).toThrow(/collides/i);
+    expect(() => parseArgs(collide, [])).toThrow(/no-loud/);
+  });
+});
+
+describe("toHelp boolean negation", () => {
+  test("a boolean renders both spellings instead of a value placeholder", () => {
+    const help = toHelp(negVerb);
+    expect(help).toContain("--changedOnly / --no-changedOnly");
+    expect(help).not.toContain("--changedOnly <boolean>");
+  });
+
+  test("the default is shown, since it decides which spelling is useful", () => {
+    const help = toHelp(negVerb);
+    expect(help).toContain("--changedOnly / --no-changedOnly (default: true)");
+    expect(help).toContain("--dryRun / --no-dryRun (default: false)");
+  });
+
+  test("non-boolean flags are unchanged apart from their default", () => {
+    expect(toHelp(negVerb)).toContain("--slug <string>");
+  });
+});
+
+// Fallout from showing a boolean's default: `z.toJSONSchema` lists a DEFAULTED
+// field in `required` (the parsed output always carries it), so the old renderer
+// labelled `--limit` required — and beside a printed default that reads as a
+// flat contradiction. The default is the truthful half.
+describe("toHelp defaults", () => {
+  test("a defaulted field is not also labelled required", () => {
+    expect(toHelp(negVerb)).not.toContain("(required) (default:");
+    expect(toHelp(searchNotes)).toContain("--limit <integer> (default: 20)");
+  });
+
+  test("a field with no default keeps its required marker", () => {
+    expect(toHelp(searchNotes)).toContain("--q <string> (required)");
+  });
+});
diff --git a/src/index.ts b/src/index.ts
index 8fd8ee0..2523c3e 100644
--- a/src/index.ts
+++ b/src/index.ts
@@ -323,10 +323,50 @@ export async function dispatchNdjson(reg: Registry, line: string): Promise<strin
 // ── CLI projection: help + argv parser ───────────────────────────────────────
 
 type JsonProps = {
-  properties?: Record<string, { type?: string; description?: string }>;
+  properties?: Record<string, { type?: string; description?: string; default?: unknown }>;
   required?: string[];
 };
 
+/**
+ * The `--no-<field>` negations for a verb's boolean inputs, as a map from the
+ * derived flag name to the input field it turns off. `--no-changedOnly` is how
+ * a `z.boolean().default(true)` is opted out of on the command line, so one
+ * boolean input covers both directions and an author no longer has to declare a
+ * second override flag to express "on by default".
+ *
+ * These names are DERIVED, not declared: `no-changedOnly` is deliberately not an
+ * input key, so this set must never be folded into the set that gates positional
+ * names — `positionals: ["no-changedOnly"]` stays the spec error it is today.
+ *
+ * EVERY boolean gets one, not only the `default(true)` ones. Which spelling an
+ * author reaches for follows from the default, but the default is a value the
+ * verb may change: if `--no-x` existed only while `x` defaulted to true,
+ * flipping that default would silently delete a flag from every script using it.
+ */
+const negationsFor = (
+  props: Record<string, { type?: string } | undefined>,
+  id: string,
+): Map<string, string> => {
+  const negations = new Map<string, string>();
+  for (const [field, meta] of Object.entries(props)) {
+    if (meta?.type !== "boolean") continue;
+    const flag = `no-${field}`;
+    // An input field literally named `no-x` alongside a boolean `x` shadows the
+    // negation: `--no-x` binds the declared field and `x` becomes impossible to
+    // turn off, with no error. Halt on the spec itself — same reason as the
+    // undeclared-positional check, so it fails on every invocation rather than
+    // only the ones that happen to pass the flag.
+    if (props[flag] !== undefined) {
+      throw new Error(
+        `invalid spec for '${id}': input field '${flag}' collides with the generated ` +
+          `negation of boolean field '${field}'. Rename one — '--${flag}' cannot mean both.`,
+      );
+    }
+    negations.set(flag, field);
+  }
+  return negations;
+};
+
 /** Render a {@link VerbSpec} as CLI `--help` text (usage, positionals, flags) for `bin`. */
 export function toHelp(v: VerbSpec, bin = "prx"): string {
   const js = toInputJsonSchema(v) as JsonProps;
@@ -338,11 +378,25 @@ export function toHelp(v: VerbSpec, bin = "prx"): string {
   const lines = [`${bin} ${v.id} ${usagePos}`.trimEnd(), "", `  ${v.summary}`, ""];
   if (flags.length) {
     lines.push("Flags:");
+    const negationOf = new Map(
+      [...negationsFor(props, v.id)].map(([flag, field]) => [field, flag] as const),
+    );
     for (const f of flags) {
       const meta = props[f] ?? {};
-      const req = required.has(f) ? " (required)" : "";
+      // `z.toJSONSchema` lists a defaulted field in `required` (the parsed OUTPUT
+      // always carries it), but on a command line a field with a default is
+      // exactly the one you may omit. Printing "(required) (default: true)" says
+      // both at once; the default is the truthful half.
+      const req = required.has(f) && meta.default === undefined ? " (required)" : "";
       const desc = meta.description ? ` — ${meta.description}` : "";
-      lines.push(`  --${f} <${meta.type ?? "value"}>${req}${desc}`);
+      const neg = negationOf.get(f);
+      // Shown for every defaulted field, because dropping the (wrong) "required"
+      // marker above would otherwise leave them annotated with nothing at all —
+      // and for a boolean the default is what decides which of the two spellings
+      // the reader actually needs.
+      const dflt = meta.default === undefined ? "" : ` (default: ${JSON.stringify(meta.default)})`;
+      const name = neg ? `--${f} / --${neg}` : `--${f} <${meta.type ?? "value"}>`;
+      lines.push(`  ${name}${req}${desc}${dflt}`);
     }
   }
   return lines.join("\n");
@@ -350,9 +404,10 @@ export function toHelp(v: VerbSpec, bin = "prx"): string {
 
 /**
  * Parse argv into the verb's input, validated by its Zod schema. CLI-isms stay
- * here, not in the schemas: `--k v` / `--k=v` / boolean `--flag` / positionals,
- * and comma-split for array-typed fields (detected from the JSON Schema). The
- * Zod `parse` does coercion (`z.coerce.number`) and is the single validation.
+ * here, not in the schemas: `--k v` / `--k=v` / boolean `--flag` and its
+ * `--no-flag` negation / positionals, and comma-split for array-typed fields
+ * (detected from the JSON Schema). The Zod `parse` does coercion
+ * (`z.coerce.number`) and is the single validation.
  */
 export function parseArgs<I extends ZodType>(
   v: VerbSpec<I, ZodType>,
@@ -361,9 +416,14 @@ export function parseArgs<I extends ZodType>(
   const js = toInputJsonSchema(v) as { properties?: Record<string, { type?: string }> };
   const props = js.properties ?? {};
   const isArray = (key: string) => props[key]?.type === "array";
-  // Every declared input key is a valid flag; a `--flag` outside this set maps to
-  // nothing and must halt (see the strict-mapping block at the end).
+  // TWO sets, deliberately not one. `known` answers "is this a declared input
+  // field?" — it gates positional names and supplies the negation targets.
+  // `knownFlags` answers "is this an accepted flag name?" and is the only one
+  // carrying the derived `--no-` names, so widening the CLI vocabulary cannot
+  // quietly make `positionals: ["no-x"]` legal.
   const known = new Set(Object.keys(props));
+  const negations = negationsFor(props, v.id);
+  const knownFlags = new Set([...known, ...negations.keys()]);
   const unknownFlags: string[] = [];
 
   // `positionals` SELECTS input fields to read positionally — it does not declare
@@ -401,7 +461,21 @@ export function parseArgs<I extends ZodType>(
     if (a.startsWith("--")) {
       const eq = a.indexOf("=");
       const key = eq >= 0 ? a.slice(2, eq) : a.slice(2);
-      if (!known.has(key) && !unknownFlags.includes(key)) unknownFlags.push(key);
+      if (!knownFlags.has(key) && !unknownFlags.includes(key)) unknownFlags.push(key);
+      const negated = negations.get(key);
+      if (negated !== undefined) {
+        // `--no-x` IS the value; it never consumes the next token, and it never
+        // accumulates (a boolean is a scalar — `--x --no-x` is last-wins, the
+        // same rule every other scalar flag follows).
+        if (eq >= 0) {
+          throw new Error(
+            `--${key} takes no value: it sets '${negated}' to false. ` +
+              `Write '--${key}' on its own, or '--${negated}' to set it true.`,
+          );
+        }
+        raw[negated] = false;
+        continue;
+      }
       if (eq >= 0) {
         setRaw(key, a.slice(eq + 1));
       } else {
@@ -437,10 +511,15 @@ export function parseArgs<I extends ZodType>(
   // (filter is a `--slug` flag; the verb declares no positionals) became a live
   // deploy of EVERY card instead of one. Report both leak paths together so a
   // caller fixes the whole line at once.
+  const validFlags = [...knownFlags].map((k) => `--${k}`).join(", ") || "(none)";
+  // The positional message suggests somewhere to PUT a value, so it names input
+  // fields only — a `--no-x` accepts none.
   const valid = [...known].map((k) => `--${k}`).join(", ") || "(none)";
   if (unknownFlags.length) {
     const flags = unknownFlags.map((k) => `--${k}`).join(", ");
-    throw new Error(`unknown flag(s) for '${v.id}': ${flags}. Valid flags: ${valid}. See --help.`);
+    throw new Error(
+      `unknown flag(s) for '${v.id}': ${flags}. Valid flags: ${validFlags}. See --help.`,
+    );
   }
   // A trailing variadic (array-typed) positional absorbs every remaining value,
   // so it can never overflow; any other shape caps at the declared count.