picocode-core 0.9.125 → 0.9.127

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "picocode-core",
-  "version": "0.9.125",
+  "version": "0.9.127",
   "description": "The agent runtime behind pico: sessions, tools, subagents, MCP, memory, and model access",
   "type": "module",
   "license": "MIT",
@@ -28,6 +28,7 @@
     "@prsm/ai": "^1.6.1",
     "diff": "^7.0.0",
     "fast-glob": "^3.3.0",
-    "proper-lockfile": "^4.1.2"
+    "proper-lockfile": "^4.1.2",
+    "unpdf": "^1.8.1"
   }
 }

package/src/tools/read.js

@@ -1,10 +1,44 @@
 import { describeParam } from './recorder.js'
 import { readFile } from 'node:fs/promises'
 import { resolve } from 'node:path'
+import { execFile } from 'node:child_process'
+import { extractText } from 'unpdf'
 
 const MAX_LINES = 2000
 const MAX_LINE_LENGTH = 2000
 
+const isPdf = (buffer) => buffer.subarray(0, 5).toString('latin1') === '%PDF-'
+
+// a pdf reads as its extracted text, one page after another, so the same
+// offset and limit window applies. poppler's pdftotext keeps the page
+// layout (labels beside their values, columns in order) so it is preferred
+// when installed; unpdf is the bundled fallback
+function pdftotext(full) {
+  return new Promise((resolve) => {
+    execFile('pdftotext', ['-layout', full, '-'], { maxBuffer: 64 * 1024 * 1024 }, (error, stdout) => {
+      if (error) return resolve(null)
+      const pages = stdout.replace(/\f$/, '').split('\f')
+      resolve(pages.map((page) => page.replace(/\s+$/, '').split('\n')))
+    })
+  })
+}
+
+async function unpdfPages(buffer) {
+  const { text } = await extractText(new Uint8Array(buffer), { mergePages: false })
+  return (Array.isArray(text) ? text : [text]).map((page) => String(page).split('\n'))
+}
+
+async function pdfLines(full, buffer) {
+  const pages = (await pdftotext(full)) ?? (await unpdfPages(buffer))
+  const lines = []
+  pages.forEach((page, i) => {
+    if (i) lines.push('')
+    lines.push(`[page ${i + 1} of ${pages.length}]`)
+    lines.push(...page)
+  })
+  return lines
+}
+
 function isBinary(buffer) {
   const len = Math.min(buffer.length, 8000)
   for (let i = 0; i < len; i++) {
@@ -27,20 +61,21 @@ export function createRead({ cwd, recorder, tracker }) {
       const full = resolve(cwd, path)
       recorder.extra({ title: path })
       const buf = await readFile(full)
-      if (isBinary(buf)) throw new Error(`${path} is a binary file`)
-      const lines = buf.toString('utf-8').split('\n')
+      const lines = isPdf(buf) ? await pdfLines(full, buf) : null
+      if (!lines && isBinary(buf)) throw new Error(`${path} is a binary file`)
+      const source = lines ?? buf.toString('utf-8').split('\n')
 
       const start = Math.max(0, offset - 1)
       const count = Math.min(limit, MAX_LINES)
-      const sliced = lines.slice(start, start + count)
+      const sliced = source.slice(start, start + count)
       const numbered = sliced
         .map((line, i) => `${start + i + 1}\t${line.length > MAX_LINE_LENGTH ? line.slice(0, MAX_LINE_LENGTH) + '…' : line}`)
         .join('\n')
 
       recorder.extra({ fullOutput: sliced.join('\n') })
-      const result = { content: numbered, totalLines: lines.length }
-      if (start + count < lines.length) {
-        result.note = `showing lines ${start + 1}-${start + sliced.length} of ${lines.length}`
+      const result = { content: numbered, totalLines: source.length }
+      if (start + count < source.length) {
+        result.note = `showing lines ${start + 1}-${start + sliced.length} of ${source.length}`
       }
       const context = tracker.check(full)
       if (context.length) result.context_from_agents_md = context