feat(compile): make the ## Sources section optional

CHANGELOG.md

@@ -19,6 +19,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
   An index written before llmwiki recorded the endpoint carries only its model name. It is preserved while you run without `LLMWIKI_EMBEDDING_PROVIDER` or an endpoint override, so upgrading does not re-embed an existing project; under either override the model name cannot establish where the vectors came from, so the next compile rebuilds the index once and records the full configuration from then on.
 
+- **Optional `## Sources` section** — `LLMWIKI_SOURCES_SECTION=off`, or `--no-sources-section` on `llmwiki compile`, stops page generation from asking the model for a trailing `## Sources` section. Unset preserves the prompt byte-for-byte.
+
+  This is for projects that render source attribution themselves. A page already carries its provenance twice — the `sources:` frontmatter, which the compiler builds from the source files it actually read rather than from anything the model writes, and the inline `^[file.md:1-5]` citation markers — so a consumer that displays either one shows the same list a third time in the prose. Nothing downstream reads the section: it is a prompt instruction only, and no linter, exporter, or citation rule parses it.
+
+  Suppressing the request is the only reliable way to not have the section, because it is not a stable string to strip. Under `--lang` the model localizes that heading along with the rest of the page, so a downstream matcher keyed on `## Sources` silently stops matching the moment a project sets an output language.
+
+  `PROMPT_VERSION` is unchanged. The constant was introduced with the optional language directive already in the page prompt, so `v1` already denotes a contract with user-selected prompt modifiers in it, and the default path here is byte-identical. Say the word if you would rather it move.
+
 ### Fixed
 
 - **Windows: profile path validation rejected every declared directory** — on win32, `llmwiki template init` failed for every template with `entity directory must be under 'wiki/'`, any profile declaring a workflow `projectionFile` failed to load, and an entity directory declared as `wiki/` was wrongly accepted despite containing every reserved subtree — on win32 it was the only entity directory that loaded at all. Declared directories canonicalize to `/`-joined repo-relative paths, but the containment check built its prefix with the platform separator (`\` on Windows), so no nested path ever matched. The lexical profile-path checks now compare POSIX paths directly; native path confinement is unchanged. Reported and diagnosed by @squ1ddy (#163).

docs/cli/compile.mdx

@@ -22,6 +22,7 @@ Run from your project root. If `sources/` is empty or doesn't exist yet, compile
 |------|-------------|
 | `--review` | Write generated pages to `.llmwiki/candidates/` instead of `wiki/`. Pages are held for human review and don't appear in the live wiki until you approve them with `llmwiki review approve <id>`. |
 | `--lang <code>` | Override the output language for this compile run (e.g. `zh-CN`, `Chinese`, `ja`, `Japanese`). Wins over the `LLMWIKI_OUTPUT_LANG` environment variable. Unset preserves the model's default behaviour. |
+| `--no-sources-section` | Stop asking the model for a trailing `## Sources` section in generated pages. Provenance is unaffected: the `sources:` frontmatter and the inline `^[file.md:1-5]` citation markers are produced by the compiler and stay exactly as they are. Equivalent to setting `LLMWIKI_SOURCES_SECTION=off`. |
 | `--concurrency <n>` | Maximum LLM calls run in parallel during this compile, across both extraction and page generation. Wins over the `LLMWIKI_COMPILE_CONCURRENCY` environment variable; defaults to `5`. Values above `50` are clamped. |
 | `--verbose` | Print detailed per-step progress: source sizes, concept merge details, page char counts, embedding batch summary, and total compile time. Equivalent to setting `LLMWIKI_VERBOSE=1`. Quiet mode (`--json`) always wins over verbose. |
 

docs/configuration/environment-variables.mdx

@@ -154,6 +154,7 @@ new hosts.
 | Variable | Default | Description |
 |---|---|---|
 | `LLMWIKI_OUTPUT_LANG` | *(model default, typically English)* | Language for generated wiki content. Applies to every prompt in the compile and query pipelines. Examples: `zh-CN`, `Chinese`, `ja`, `Japanese`. The `--lang` CLI flag on `llmwiki compile` and `llmwiki query` overrides this for a single invocation |
+| `LLMWIKI_SOURCES_SECTION` | *(enabled)* | Set to `off`, `false`, `0`, or `no` to stop asking the model for a trailing `## Sources` section in generated pages. Useful when your own renderer already displays the `sources:` frontmatter, which would otherwise appear twice. Does not change provenance: the frontmatter and the inline `^[file.md:1-5]` citation markers come from the compiler, not from this instruction. The `--no-sources-section` flag on `llmwiki compile` sets it for a single invocation |
 | `LLMWIKI_PROMPT_BUDGET_CHARS` | `200000` | Character ceiling for combined per-concept source content sent to the LLM. Raise for larger-context models; lower for small-context local models. A stderr warning prints when the cap fires |
 
 ---

src/cli.ts

@@ -27,6 +27,7 @@ import quickstartCommand from "./commands/quickstart.js";
 import contextCommand, { type ContextCommandOptions } from "./commands/context.js";
 import { startMCPServer } from "./mcp/server.js";
 import { applyLanguageOption } from "./utils/output-language.js";
+import { applySourcesSectionOption } from "./utils/sources-section.js";
 import { ensureProviderAvailable } from "./utils/provider-guard.js";
 import { setVerbose } from "./utils/output.js";
 import { parseConcurrencyFlag } from "./compiler/concurrency.js";
@@ -116,15 +117,26 @@ program
     "--lang <code>",
     "Target language for generated wiki content (e.g. \"Chinese\", \"ja\", \"zh-CN\"). Equivalent to setting LLMWIKI_OUTPUT_LANG.",
   )
+  .option(
+    "--no-sources-section",
+    "Omit the trailing ## Sources section from generated pages. Source attribution stays in the sources: frontmatter and the inline ^[...] citation markers. Equivalent to setting LLMWIKI_SOURCES_SECTION=off.",
+  )
   .option(
     "--concurrency <n>",
     "Max concurrent LLM calls during compile (or set LLMWIKI_COMPILE_CONCURRENCY; default 5)",
   )
   .option("--verbose", "Print detailed progress (or set LLMWIKI_VERBOSE=1)")
-  .action(async (options: { review?: boolean; lang?: string; concurrency?: string; verbose?: boolean }) => {
+  .action(async (options: {
+    review?: boolean;
+    lang?: string;
+    sourcesSection?: boolean;
+    concurrency?: string;
+    verbose?: boolean;
+  }) => {
     try {
       setVerbose(verboseEnabled(options.verbose));
       applyLanguageOption(options.lang);
+      applySourcesSectionOption(options.sourcesSection);
       requireProvider();
       await compileCommand({ review: options.review, concurrency: parseConcurrencyFlag(options.concurrency) });
     } catch (err) {

src/compiler/prompts.ts

@@ -12,6 +12,7 @@ import type {
 } from "../utils/types.js";
 import type { PageKindRule, SeedPage } from "../schema/index.js";
 import { languageDirective } from "../utils/output-language.js";
+import { sourcesSectionEnabled } from "../utils/sources-section.js";
 
 /**
  * Build a list of optional prompt lines, omitting empty entries so the
@@ -24,6 +25,16 @@ function withLangLine(...lines: string[]): string[] {
   return lang ? [...lines, lang] : lines;
 }
 
+/**
+ * The page-generation instruction for the trailing `## Sources` section, or
+ * nothing when the project opted out. Spreadable so the default prompt keeps
+ * its exact wording instead of gaining a blank line where the request was.
+ */
+function sourcesSectionLines(): string[] {
+  if (!sourcesSectionEnabled()) return [];
+  return ["Include a ## Sources section at the end listing the source document."];
+}
+
 /**
  * Named version of the extraction + page-generation prompt contract.
  *
@@ -174,7 +185,7 @@ export function buildPagePrompt(
     ...withLangLine(
       `You are a wiki author. Write a clear, well-structured markdown page about "${concept}".`,
       "Draw facts only from the provided source material.",
-      "Include a ## Sources section at the end listing the source document.",
+      ...sourcesSectionLines(),
       "Suggest [[wikilinks]] to related concepts where appropriate.",
       "Write in a neutral, informative tone. Be concise but thorough.",
     ),

src/utils/sources-section.ts

@@ -0,0 +1,50 @@
+/**
+ * Controls whether page generation asks the model for a trailing
+ * `## Sources` section.
+ *
+ * A compiled page already carries its provenance twice: the `sources:`
+ * frontmatter list, which the compiler builds from the source files it
+ * actually read rather than from anything the model writes, and the inline
+ * `^[file.md:1-5]` citation markers. A project that renders either of those
+ * itself ends up showing a third, redundant copy in the prose.
+ *
+ * Stripping the section downstream is awkward because it is not a stable
+ * string: under `--lang` the model localizes that heading along with the rest
+ * of the page, so a consumer has no fixed text to match on. Suppressing the
+ * instruction is the only reliable way to not have it.
+ *
+ * Opting out is a rendering choice, not a provenance one — the frontmatter and
+ * the citation markers are untouched. Unset preserves the historical prompt
+ * byte-for-byte.
+ */
+
+const SOURCES_SECTION_ENV_VAR = "LLMWIKI_SOURCES_SECTION";
+
+/**
+ * Values that switch the section off, compared case-insensitively after
+ * trimming. Users reach for different spellings of "no" and silently ignoring
+ * three of the four would be worse than accepting all of them.
+ */
+const DISABLED_VALUES: ReadonlySet<string> = new Set(["0", "false", "off", "no"]);
+
+/** True when generated pages should still include a `## Sources` section. */
+export function sourcesSectionEnabled(): boolean {
+  const raw = process.env[SOURCES_SECTION_ENV_VAR];
+  if (raw === undefined) return true;
+  return !DISABLED_VALUES.has(raw.trim().toLowerCase());
+}
+
+/**
+ * Apply the CLI `--no-sources-section` flag into the shared env slot so the
+ * prompt builder picks it up downstream.
+ *
+ * Commander defaults a `--no-x` option to `true`, so only an explicit `false`
+ * carries the user's intent; leaving the variable alone otherwise keeps it
+ * authoritative for setups that configure the project rather than a single
+ * invocation. Mirrors `applyLanguageOption` in output-language.ts.
+ */
+export function applySourcesSectionOption(enabled: boolean | undefined): void {
+  if (enabled === false) {
+    process.env[SOURCES_SECTION_ENV_VAR] = "off";
+  }
+}

test/sources-section.test.ts

@@ -0,0 +1,87 @@
+/**
+ * Unit tests for the `## Sources` section toggle and its effect on the page
+ * prompt.
+ *
+ * Default behaviour (no env, no flag) must keep the page prompt byte-identical
+ * to the previous implementation. Opting out must remove only the section
+ * request — the inline citation contract and the `sources:` frontmatter that
+ * carry provenance are not part of this switch.
+ */
+
+import { describe, it, expect, afterEach } from "vitest";
+import {
+  applySourcesSectionOption,
+  sourcesSectionEnabled,
+} from "../src/utils/sources-section.js";
+import { buildPagePrompt } from "../src/compiler/prompts.js";
+
+const ENV_KEY = "LLMWIKI_SOURCES_SECTION";
+const SECTION_REQUEST = "Include a ## Sources section";
+
+afterEach(() => {
+  delete process.env[ENV_KEY];
+});
+
+describe("sourcesSectionEnabled", () => {
+  it("defaults to enabled when the env var is unset", () => {
+    expect(sourcesSectionEnabled()).toBe(true);
+  });
+
+  it.each(["0", "false", "off", "no"])("treats %s as disabled", value => {
+    process.env[ENV_KEY] = value;
+    expect(sourcesSectionEnabled()).toBe(false);
+  });
+
+  it("ignores surrounding whitespace and case", () => {
+    process.env[ENV_KEY] = "  OFF  ";
+    expect(sourcesSectionEnabled()).toBe(false);
+  });
+
+  it("stays enabled for any other value", () => {
+    process.env[ENV_KEY] = "on";
+    expect(sourcesSectionEnabled()).toBe(true);
+  });
+});
+
+describe("applySourcesSectionOption", () => {
+  it("leaves the env var untouched when the flag was not passed", () => {
+    applySourcesSectionOption(undefined);
+    expect(process.env[ENV_KEY]).toBeUndefined();
+  });
+
+  it("leaves an existing opt-out in place on commander's default true", () => {
+    process.env[ENV_KEY] = "off";
+    applySourcesSectionOption(true);
+    expect(sourcesSectionEnabled()).toBe(false);
+  });
+
+  it("disables the section when --no-sources-section was passed", () => {
+    applySourcesSectionOption(false);
+    expect(sourcesSectionEnabled()).toBe(false);
+  });
+});
+
+describe("buildPagePrompt honours the toggle", () => {
+  it("requests the section by default", () => {
+    expect(buildPagePrompt("Concept", "src", "", "")).toContain(SECTION_REQUEST);
+  });
+
+  it("omits the request when disabled", () => {
+    process.env[ENV_KEY] = "off";
+    expect(buildPagePrompt("Concept", "src", "", "")).not.toContain(SECTION_REQUEST);
+  });
+
+  it("keeps the inline citation contract when disabled", () => {
+    process.env[ENV_KEY] = "off";
+    const out = buildPagePrompt("Concept", "src", "", "");
+    expect(out).toContain("^[filename.md:START-END]");
+    expect(out).toContain("Draw facts only from the provided source material.");
+  });
+
+  it("drops the line rather than blanking it", () => {
+    const enabled = buildPagePrompt("Concept", "src", "", "");
+    process.env[ENV_KEY] = "off";
+    const disabled = buildPagePrompt("Concept", "src", "", "");
+    expect(disabled.split("\n").length).toBe(enabled.split("\n").length - 1);
+  });
+});