textile

Writing Textile markup works well for the agent because the prefix syntax, such as h2. for headings and pipe characters for tables, is direct and simple. The agent can construct complex document hierarchies, inline styling, code blocks, and tables using plain text. This keeps the draft phase straightforward and syntactically uncomplicated.

The tool performs adequately when handling links and embedded images, which use specific tag structures like exclamation marks for image references. For workflows that require embedded figures, the agent executes an external Python script to generate the image, then references its path in the Textile source. Resolving these dependencies requires multi-step coordination, but it works reliably because file paths are resolved locally.

The compiler performs poorly at alerting the agent to formatting mistakes. Pandoc silently compiles incorrect Textile markup as plain text rather than failing or returning explicit error messages. Because the compiler does not raise exceptions for syntax errors, the agent cannot rely on standard error streams to debug format regressions and must manually inspect the rendered XML or HTML outputs to confirm structural integrity.

This tool is a good fit for agents designed to generate structured files such as scientific reports, technical specifications, or documents with heavy tabular data. It suits workflows where output layout is straightforward and the operating environment allows direct control over local compilation commands.

It is a poor fit for agents that must produce precise visual layouts without external verification capabilities. Because the Pandoc parser does not output compilation errors when encountering invalid markup, an unmonitored agent can easily render broken tables or unparsed tags without detecting the defects.

This diagram shows the multi-step compile and verification loop used to generate a document with embedded assets.