Files
pdfcraft/crates/automation
jcl-bitandClaude Opus 5.5 70da0179a6 Security: --root confinement leaks and escapes (#136, #137) (#164)
* Automation: --root refuses every outside path the same way (#136)

With a root set, `resolve` canonicalized a path before checking it against
the root, so an existing outside path got "is outside the allowed
directory" while a missing one got the OS error ("No such file or
directory", "Not a directory", "invalid path"): a confined MCP client or
script could probe which files and folders exist anywhere on the disk.

Containment is now decided before anything about an outside path is
reported:
- `..` is resolved by name first (as Windows does), so a path can't pass
  through an outside folder on its way back in;
- on Windows another network share or device namespace (`\\host\share`,
  `\\?\UNC\...`, `\\.\...`) is refused by name, without contacting it;
- the deepest existing ancestor, links followed, must be inside the root
  before anything is said about the rest;
- a broken link is refused the same way, so whether its target exists
  stays hidden too.

Every outside path now gets `<path> is outside the allowed directory
<root>`, existing or not, file or folder, read or write. Filesystem errors
such as "not found" are still reported for paths inside the root.

Behaviour change: on Unix, `..` now means the parent by name, as on
Windows (`link/../x` is `x` beside `link`), and a write such as
`new/../x.pdf` that stays in the root now succeeds instead of failing with
"'..' is not allowed here".

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* CLI: script images stay inside --root (#137)

`pdfcraft-cli run --script steps.json --root DIR` wrote each step's
`"out"` PNG with `std::fs::write`, never through the root check the tools
use: an absolute `"out"` created or replaced any file the process could
write, and a relative one landed in the working directory instead of the
root.

The image is now saved through the new `Automation::write_output`, which
uses the same resolver and atomic write as a tool's own output: outside
paths get the usual "is outside the allowed directory" refusal (the step
fails, nothing is written), relative paths resolve inside the root, and
without --root the path is written as given, in place as before (so
`/dev/stdout` still works). `run <tool> --out` goes the same way.

Found while testing this, the same "." problem in two tools:
- a write path naming a folder, such as "." (the root itself), staged its
  temporary file beside that folder, i.e. outside the root for ".",
  overwriting and then deleting any file there called
  `.root.pdfcraft-tmp`. `write_atomic` now refuses a folder before
  writing anything;
- `image_save` adds an extension to a path without one, so "." became
  `root.png` beside the root. It now refuses a folder first.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Automation: files named after a document stay in the output folder

Found while reviewing the --root fixes for #136 and #137. Folder outputs
name their files after the document and joined that name straight onto the
resolved folder, without checking the result against the root:

- `doc_export_images` and `doc_export_all_images` used the whole document
  name, so `doc_create` with `name: "../../escape"` followed by
  `doc_export_images` wrote `escape_page_1.png` outside the root (an
  absolute name went anywhere);
- `page_extract` (separate files), `doc_split`, `ocr_recognize_files` and
  `action_run` took the last component, but on Windows a name such as
  `x/C:evil.pdf` gives `C:evil`, and `folder.join("C:evil-page1.pdf")`
  replaces the folder with a path relative to the current directory on C:.

Every such name now goes through `child()`, which turns separators,
colons and control characters into `_` (and prefixes a name of only dots),
so the file is always a direct child of the folder. `with_root` also
refuses a root that is not a folder.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-07 21:12:32 -04:00
..

pdfcraft-automation

Agent control for PdfCraft: a headless tool table over the engine, and an opt-in MCP server.

  • Layer: L7 (architecture §13), but headless. It depends on engine, render and organize, never on a UI toolkit.
  • Status: over a hundred tools; pdfcraft-cli tools lists the current set with their schemas (the count isn't repeated here, so it can't go stale). They cover inspection, rendering, text, page edits and organizing, page labels, bookmarks, comments and their review, form filling and authoring (fields, properties, scripts, data exchange, detection), redaction and sanitizing, password protection and signatures, adding text, images, links and stamps, optimizing, OCR, printing, export, metadata, undo/redo, save, combine, extract and split. Comment tools take geometry in the same top-left-origin points as everything else, and comment_add can mark text by searching for it (find). The MCP server also serves the open documents as resources, page images included. The running app's UI is driven separately, through the ui.* control channel in pdfcraft-ui-egui.

API

let mut a = Automation::new().with_root("/work")?;      // optional: confine all paths
let doc = a.call("doc_open", &json!({ "path": "in.pdf" }))?;   // Vec<Content>
a.write_output("page1.png", &png)?;                      // save a result yourself, under the same root rules
tools() -> Vec<ToolDef>                                  // name, title, description, input_schema, read_only, destructive, command
mcp::McpServer::new(a).serve(stdin, stdout)?             // newline-delimited JSON-RPC 2.0

Content is Json(Value) or Png { data, width, height }. Errors are UnknownTool, InvalidArgs (the call didn't match the schema) or Failed (a readable message for the agent).

Conventions

  • Tool names are snake_case (page_rotate): MCP clients reject dots. ToolDef::command links a tool to the registry id it automates (page.rotate), and command_list reports the link.
  • Pages and positions are 1-based. Rectangles are PDF points with the origin at the top-left of the displayed page.
  • Unknown arguments are rejected, so typos fail loudly.
  • Tools that change a document return its summary (doc, pages, dirty, undo, redo, …).
  • doc_close refuses to drop unsaved changes unless discard_changes: true. doc_save writes atomically, incrementally in place, and in full for a new path.
  • With a root set, every read and write path must resolve inside it (symlinks and .. included). Relative paths resolve inside the root, and .. is resolved by name before the check. Every path outside the root gets the same refusal (<path> is outside the allowed directory <root>), whether or not it exists, so a confined agent can't probe the rest of the disk; on Windows another network share or device path (\\host\share, \\?\UNC\…, \\.\…) is refused without being contacted. "Not found" and other filesystem errors are reported only for paths inside the root. Tools that write several files into a folder name them after the document or input file, with separators, colons and control characters replaced by _, so a name can't lead them out of the folder. The root itself must be a folder.

MCP server

Opt-in only. Nothing starts it automatically, and it opens no port. It runs while pdfcraft-cli mcp [--root DIR] runs, normally launched by an agent from its MCP configuration. It stops when stdin closes. The CLI's mcp Cargo feature (on by default) compiles it out entirely when disabled.

Implemented: initialize (protocol 2025-06-18, 2025-03-26, 2024-11-05), ping, tools/list (with readOnlyHint/destructiveHint annotations), tools/call (JSON results also returned as structuredContent; images as image/png), and resources/list, resources/templates/list and resources/read. The resources expose the open documents read-only: pdfcraft://doc/{doc}/info (JSON), …/text, …/page/{page}/text and …/page/{page}/image{?dpi} (PNG, 1–600 dpi). Tool failures come back as isError: true results, so the agent can read and recover from them.

Adding a tool

  1. Add the engine capability first, with its tests (the tool is a thin adapter).
  2. Add a ToolDef in src/tools.rs: a precise description, the schema, ro()/destructive(), and cmd() if a registry command exists.
  3. Handle it in Automation::call, validating pages and positions with the existing helpers.
  4. Add an end-to-end test in tests/automation.rs. tool_table_is_well_formed checks names, schemas and command links.