read-filelisted

# Read any file There is no dedicated document-reading tool — you extract the content yourself with the shell (the same approach AgentWorks uses: write a little Python and run it via `execute_shell_command`). Always start by identifying the file, then use the tool that is **already installed** for that format. Do NOT reach for a tool that isn't there (e.g. `pdftotext`/poppler and `libreoffice`/`soffice` are NOT installed on this machine — don't use them). ``` file "<path>" # what is it, really? (extension can lie) ``` ## By format - **Plain text / Markdown / CSV / JSON / code** — just `cat "<path>"` (or `head`/`jq` for large/structured files). - **PDF (digital, has a real text layer)** — use `pypdf` (already installed), the canonical fast path: ``` python3 -c "import sys,pypdf; r=pypdf.PdfReader(sys.argv[1]); print('\n\n'.join((p.extract_text() or '') for p in r.pages))" "<path>" ``` - **PDF (scanned / image-only), garbled or empty pypdf output, OR any complex layout** (tables, multi-column, mixed text + figures) — parse it **locally** with **`liteparse`** (LlamaIndex's local parser — this app's OCR/parse path). Install on demand if missing, then run in **local mode** so nothing leaves the machine: ``` pip3 install --break-system-packages liteparse # or: uv pip install liteparse ``` `liteparse` handles scanned pages, OCR, tables, and layout far better than plain extraction. For a single scanned page you may instead call the `read_image` tool (visi