fix(read): normalized recovery paths

Used the resolved cwd-relative path in summary recovery selectors, PDF image handles, and notebook diagnostics.

Fixes #7788

packages/coding-agent/CHANGELOG.md

@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Normalized resolved file paths in `read` summary recovery selectors, PDF image handles, and notebook errors so suffix-matched input does not teach agents malformed follow-up paths ([#7788](https://github.com/can1357/oh-my-pi/issues/7788)).
+
 ## [17.2.9] - 2026-08-05
 
 ### Breaking Changes

packages/coding-agent/src/tools/read.ts

@@ -2486,6 +2486,7 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 		const imageMetadata = await readImageMetadata(absolutePath);
 		const mimeType = imageMetadata?.mimeType;
 		const ext = path.extname(absolutePath).toLowerCase();
+		const resolvedDisplayPath = formatPathRelativeToCwd(absolutePath, this.session.cwd);
 		const shouldConvertWithMarkit = CONVERTIBLE_EXTENSIONS.has(ext);
 
 		// Profiler reports (macOS `sample` call trees, V8 `.cpuprofile` JSON):
@@ -2530,7 +2531,7 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 				fileSize,
 			}));
 		} else if (isNotebookPath(absolutePath) && !isRawSelector(parsed)) {
-			const notebookText = await readEditableNotebookText(absolutePath, localReadPath);
+			const notebookText = await readEditableNotebookText(absolutePath, resolvedDisplayPath);
 			if (isMultiRange(parsed) && parsed.kind === "lines") {
 				return this.#buildInMemoryMultiRangeResult(notebookText, parsed.ranges, {
 					details: { resolvedPath: absolutePath },
@@ -2549,7 +2550,7 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 			const result = await convertFileWithMarkit(absolutePath, signal);
 			if (result.ok) {
 				const renderedContent =
-					ext === ".pdf" ? rewritePdfImagePlaceholders(result.content, localReadPath) : result.content;
+					ext === ".pdf" ? rewritePdfImagePlaceholders(result.content, resolvedDisplayPath) : result.content;
 				// Route the converted markdown through the in-memory text builder
 				// so line-range selectors (`file.pdf:50-100`, `:5-16,40-80`) and
 				// raw mode apply against the converted output. Without this,
@@ -2592,7 +2593,7 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 				return toolResult<ReadToolDetails>({ resolvedPath: absolutePath, suffixResolution })
 					.text(
 						prependSuffixResolutionNotice(
-							`[Cannot read binary file '${formatPathRelativeToCwd(absolutePath, this.session.cwd)}' (${formatBytes(fileSize)}); not valid UTF-8 text. Use ':raw' to read bytes verbatim.]`,
+							`[Cannot read binary file '${resolvedDisplayPath}' (${formatBytes(fileSize)}); not valid UTF-8 text. Use ':raw' to read bytes verbatim.]`,
 							suffixResolution,
 						),
 					)
@@ -2609,7 +2610,7 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 				if (summary?.parsed && summary.elided) {
 					const renderedSummary = this.#renderSummary(summary);
 					const footer = formatSummaryElisionFooter(
-						localReadPath,
+						resolvedDisplayPath,
 						renderedSummary.elidedRanges,
 						renderedSummary.elidedLines,
 					);

packages/coding-agent/test/read-summary.test.ts

@@ -82,6 +82,24 @@ describe("read summary", () => {
 		expect(result.details?.summary?.elidedSpans).toBe(2);
 	});
 
+	it("uses the resolved path in elision recovery selectors after suffix matching", async () => {
+		const fixture = path.join(tmpDir, "project", "src", "fixture.ts");
+		await fs.mkdir(path.dirname(fixture), { recursive: true });
+		await fs.writeFile(
+			fixture,
+			"export function alpha(value: string): string {\n\tconst clean = value.trim();\n\tconst label = clean || 'alpha';\n\treturn label.toUpperCase();\n}\n\nexport function beta(): number {\n\tconst one = 1;\n\tconst two = 2;\n\treturn one + two;\n}\n",
+		);
+		const malformed = "src/fixture.ts";
+
+		const tool = new ReadTool(createSession(tmpDir));
+		const result = await tool.execute("read-summary-suffix-path", { path: malformed });
+		const text = textOutput(result);
+
+		expect(result.details?.suffixResolution?.to).toBe("project/src/fixture.ts");
+		expect(text).toContain("with project/src/fixture.ts:1-5,7-11]");
+		expect(text).not.toContain(`with ${malformed}:`);
+	});
+
 	it("summarizes Markdown only when prose summaries are enabled", async () => {
 		const fixture = path.join(tmpDir, "fixture.md");
 		await fs.writeFile(
@@ -316,9 +334,9 @@ describe("read summary", () => {
 		expect(result.details?.summary?.elidedSpans).toBe(2);
 		expect(result.details?.summary?.elidedLines).toBeGreaterThan(0);
 		expect(text).toContain("ln elided");
-		expect(text).toContain(`${fixture}:1-5,7-11`);
-		expect(text).not.toContain(`${fixture}:raw`);
-		expect(text).not.toContain(`${fixture}:1-9999`);
+		expect(text).toContain("footer.ts:1-5,7-11");
+		expect(text).not.toContain("footer.ts:raw");
+		expect(text).not.toContain("footer.ts:1-9999");
 		// Footer must be the LAST block of output so the recovery hint sits
 		// next to the structural summary it describes.
 		expect(text.trimEnd().endsWith("]")).toBe(true);

packages/coding-agent/test/tools/read-pdf-images.test.ts

@@ -125,8 +125,8 @@ describe("read PDF image extraction", () => {
 			.join("\n");
 
 		expect(text).not.toContain("<!-- image:");
-		expect(text).toContain(`read \`${pdfPath}:p11-img0.png\``);
-		expect(text).toContain(`read \`${pdfPath}:p11-img1.png\``);
+		expect(text).toContain("read `doc.pdf:p11-img0.png`");
+		expect(text).toContain("read `doc.pdf:p11-img1.png`");
 		// Page/size metadata is preserved in the handle text.
 		expect(text).toContain("page 11, 199x124pt");
 	});
@@ -144,7 +144,7 @@ describe("read PDF image extraction", () => {
 			.join("\n");
 
 		expect(text).not.toContain("<!-- image:");
-		expect(text).toContain(`read \`${pdfPath}:p3-img0.png\``);
+		expect(text).toContain("read `doc.pdf:p3-img0.png`");
 	});
 
 	it("extracts a PDF image member as an inline image block", async () => {