fix(agents): validate document formats and trim HTML URLs

docs/development/agent-workflow.md

@@ -57,6 +57,7 @@ Only headings present outside inert HTML containers contribute slugs or duplicat
 Explicit HTML anchors use `parse5`, excluding comments, scripts, styles and template contents.
 Rendered HTML anchor and image-map area hrefs and image sources use the same local-reference checks
 as Markdown links, including decoded attributes and fragment validation.
+URL attributes discard surrounding ASCII control/space characters before resolution.
 Iframe sources are document references and retain Markdown-target anchor checks.
 Image references check file existence without interpreting image fragments as
 Markdown headings; document links keep anchor checks even when sharing a target.
@@ -70,7 +71,8 @@ Declared scoped dependencies, scope wildcards and matching TypeScript path alias
 are recognized as package/alias mentions. Traversal and document-file imports are
 rejected before those exemptions, including document paths with fragments or queries.
 All recognized Markdown extensions share the document-import guard; reStructuredText
-and AsciiDoc documents are also excluded from package exemptions. URL-encoded
+and AsciiDoc, PDF, Word, OpenDocument, RTF, Org and TeX documents are also excluded
+from package exemptions. URL-encoded
 paths do not receive package exemptions. TypeScript configuration is parsed as JSONC.
 Declared packages also permit safe subpaths; exact aliases stay exact.
 Declared package mentions may include a version (including semver comparators and wildcard ranges) or dist-tag qualifier.

tools/skills/agent-guidance-markdown.mjs

@@ -83,7 +83,10 @@ function htmlNavigation(html, inspect = () => {}) {
                 (node.tagName === 'video' && attribute.name === 'poster')
             )
                 references.push({
-                    target: attribute.value,
+                    target: attribute.value.replace(
+                        /^[\u0000-\u0020]+|[\u0000-\u0020]+$/gu,
+                        ''
+                    ),
                     image: !['a', 'area', 'iframe'].includes(node.tagName),
                 });
             if (

tools/skills/validate-agent-guidance.mjs

@@ -130,7 +130,9 @@ async function packageMentions(rootDir) {
         if (
             /%[\da-f]{2}/iu.test(token) ||
             MARKDOWN_EXTENSION.test(path) ||
-            /\.(?:txt|json|ya?ml|html?|rst|rest|adoc|asciidoc)$/iu.test(path)
+            /\.(?:txt|json|ya?ml|html?|rst|rest|adoc|asciidoc|pdf|docx?|odt|rtf|org|tex)$/iu.test(
+                path
+            )
         )
             return false;
         if (packages.includes(token)) return true;

tools/skills/validate-agent-guidance.test.mjs

@@ -1266,3 +1266,29 @@ for (const extension of ['rst', 'rest', 'adoc', 'asciidoc']) {
         );
     });
 }
+
+for (const extension of ['pdf', 'doc', 'docx', 'odt', 'rtf', 'org', 'tex']) {
+    test(`document format is not a package exemption: ${extension}`, async (t) => {
+        assert.ok(
+            (
+                await diagnostics(t, {
+                    'package.json': JSON.stringify({
+                        dependencies: { '@angular/core': '*' },
+                    }),
+                    'CLAUDE.md':
+                        '@AGENTS.md\n\nRead @angular/core/docs/guide.' +
+                        extension,
+                })
+            ).some((message) => message.includes('additional or inline'))
+        );
+    });
+}
+for (const html of [
+    '<a href=" docs/example.md#repeat ">Guide</a>',
+    '<iframe src="&#9;docs/example.md&#10;"></iframe>',
+    '<img src=" docs/example.md ">',
+]) {
+    test(`HTML URL edge whitespace is ignored: ${html}`, async (t) => {
+        assert.deepEqual(await diagnostics(t, { 'AGENTS.md': html }), []);
+    });
+}