Document Ingest

Document ingest is a part of the Discovery process that turns an unstructured set of customer test documentation into clean, consistent test plans that you can use to generate Cycle test cases. Source documents can be given in almost any format: .docx, .pdf, .xlsx, and more. Document ingest converts them into Markdown test plan files, checks their health, and repairs common problems along the way. The result is a tidy set of source material that streamlines your progression to Formulation.

Document ingest runs through your AI coding agent using the Cycle MCP Server, so the agent does the heavy lifting while you review and steer. It is part of the local Cycle Intelligence experience, not the Intelligent Discovery web app.

Blue Yonder focus

Document ingest is presently tuned for Blue Yonder Warehouse Management System (WMS) documentation.

Why use document ingest?

Customer test documentation is rarely ready for test automation development. It arrives as a mix of file types written by different people in different styles. It can have duplicated scenarios, typos, and inconsistent structure. Document ingest refines the collection of test information into a clear, consistent set of test plans. Instead of manually triaging dozens of files, you get:

• A consistent format — every source document is converted into markdown test specifications with the same shape, so they are easy to scan, compare, and reuse.
• A health assessment — a quick read on the utility of the source material, so you can tell early whether a corpus is a solid foundation or a rough starting point.
• Duplicate detection — a first-pass analysis that flags overlapping test specifications, so you don't build the same test twice.
• Automatic repairs — simple issues such as spelling are corrected for you, and you are walked through the judgment calls that need a human.

The payoff is speed and confidence: you reach Formulation with source material you understand and trust, rather than a folder of documents you are afraid to open.

Prerequisites

• A supported AI coding agent wired up to the Cycle MCP Server (see the Cycle MCP setup guide).
• The customer's source documents on hand, in whatever formats you received them — including archives such as .zip.
• Knowledge of which WMS the documentation describes, so you can pass it to the ingest prompt.

How does it work?

Document ingest is a guided, prompt-driven flow. You create a project, drop your documents into it, and hand off to the agent, which walks through extraction, conversion, analysis, and repair one stage at a time — pausing for your input on the decisions that matter.

Behind a single prompt, the agent:

1. Extracts any archive files (for example, .zip) so every source document is available.
2. Converts each input file into atomic test specifications in markdown — one coherent scenario per specification, rather than a wall of prose.
3. Analyzes the health of the resulting specifications, summarizing issues across the whole corpus.
4. Cures simple problems automatically, such as spelling.
5. Walks you through the more complex repair options that need a human decision.

How do I run it?

Run the following from your coding agent, in order.

1. Create a project

Start a new project with the Cycle MCP prompt:

/cycle:create-project {project-name} {parent-directory} {username}

This scaffolds the project structure document ingest expects.

2. Add your source documents

Copy the customer's test specifications into the project. The default suggestion is a folder named Customer-Docs, but you can point the ingest step at any folder you choose.

3. Invoke document ingest

Call the ingest prompt, passing the folder that holds your documents and the WMS they describe:

/cycle:document-ingest {documents-folder} {wms-system}

The agent then runs the extract → convert → analyze → cure flow described above, prompting you whenever a repair needs your judgment.

Review as you go

The agent does the typing, but the results are only as good as the source material. Read the summaries it produces and confirm its repair suggestions, especially where it flags duplicates or ambiguous specifications.

Long-running work

The agent may take a long time to process the documents, especially if the documents are large in size or number.

What you get

When the process completes, document ingest leaves two folders in your project:

• Conversion-Helpers — the scripts used to process your input files. Keep these if you want to re-run or adjust the conversion later.
• Test-Specs — your ingested output and its analysis:
  • The test specifications themselves, in consistent markdown format.
  • Input-Sources.txt — an index of every file that was ingested, so you can trace each specification back to its source.
  • Executive-Summary.md — a high-level read on the health of the input documents; start here.
  • Verbose-Detail.md — a detailed, per-file breakdown of the issues found, for when you need to dig in.
  • Duplicate-Detection.txt — a first-pass analysis of test specifications that appear to overlap.
  • Repair-Log.md — a record of every repair that was made, so nothing changes silently.

With these in hand, you can judge the quality of the source material at a glance and move confidently into Formulation.