Content Building Products

Turning a messy inbox of client material into a usable catalogue

A building-products company had AOS index its inbox and content into a structured catalogue — each file tagged with type, status, and key metadata — so later runs could actually use the material. The system catalogued the files correctly and managed token cost by writing the index in segments. Cataloguing works; arbitrary-upload intake remains an open gap.

85% confidence

01 Ground

Source material is only useful if the system knows it exists. A client folder fills up with intake documents, brand assets, prior content, and reference files — but to a downstream skill, an un-indexed pile of files is invisible. The drafter that needs a tone-of-voice document or the diagnostic that needs prior performance data cannot reach for what has never been catalogued.

This building-products company asked AOS to do the unglamorous but foundational job: read its inbox and content zones and index every file into a catalogue, each entry carrying its type, its status, and the key metadata that makes it findable and usable later. The goal was to convert raw material into an addressable inventory.

There is a real engineering constraint hiding in this task. Cataloguing a folder means reading a lot of files, and reading a lot of files naively means paying for a large amount of context in a single pass — which is both expensive and fragile. A catalogue run that blows its token budget mid-folder is worse than useless.

  • Inbox and content files indexed into a catalogue with type, status, and key metadata per entry skill run output 88% confident
  • Catalogue written in segments to control token cost across the run skill run output 85% confident

02 Decide

AOS catalogued the material correctly, producing structured entries — type, status, key metadata — rather than a flat file listing, so that downstream skills can query the inventory and pull the right document for the job.

The notable engineering call was to write the catalogue in segments rather than in one monolithic pass. Segmenting the writes kept the token cost of the run bounded and the process resilient, a deliberate trade that prioritises a catalogue that completes reliably over one that tries to do everything in a single expensive shot.

What this might get wrongThis run proves cataloguing of the inbox and content zones — indexing material that is already present into a structured, usable inventory. It does not close the arbitrary-upload intake gap: handling files dropped in ad hoc, in unexpected formats or outside the expected zones, remains an open problem to be solved separately.

03 Deploy

The deliverable was a structured catalogue of the inbox and content zones, each file entry carrying type, status, and key metadata, written in cost-controlled segments — an addressable inventory rather than an opaque folder.

Practically, this is the index that unblocks later work: with material catalogued, downstream content and strategy runs can find and use the right source files instead of being blind to what the client has provided.

04 Result