Nano PDF Tool

Problem context

PDFs often appear in contracts, reports, and document cleanup. The slow part is rarely opening the file; it is repeatedly locating text, fixing titles, correcting typos, and adjusting content. Manual edits can miss pages, while script-based workflows still require maintaining parsing and write-back logic. nano-pdf targets small, describable, verifiable PDF editing tasks and collapses natural-language PDF editing into a CLI-style invocation.

How it works

The skill is organized around natural-language instructions. Core inputs typically include:
- instruction: required, describing the editing goal
- context: optional, adding context such as page scope, terminology, or constraints

The execution flow is roughly: parse the task configuration and dependencies, build an execution DAG, execute nodes in topological order, handle conditional branches and retries, and finally aggregate node outputs, status, and logs into a reviewable result. The documented capability examples lean toward structured edits on specific pages or content, such as changing a title, fixing typos, or organizing page text.

Boundaries and caveats

It is intended for Agent environments that support SKILL.md, relies on a built-in LLM, and supports Windows, macOS, and Linux. Clearer instructions produce more stable results. Complex layout changes, cross-page dependencies, graphics edits, or compliance-critical documents should still be reviewed manually. API keys should be configured via environment variables and should not be hard-coded.