File-naming convention automation: why Claude Code drifts, and how to enforce it
A naming convention only works if it's followed exactly, every time. That's precisely where a regenerating model struggles. Here's how to make it airtight.
A file-naming convention has to be applied byte-for-byte identically every time, and that's exactly where Claude Code struggles, because it regenerates output rather than applying a fixed rule. Caddi encodes your convention as deterministic code so every file matches the pattern exactly.
You can absolutely build this with Claude Code. The question is which version you need. An individual, judgment-heavy helper suits Claude Code well. A high-volume workflow that has to connect to your real systems, run identically every time, and stay maintained is where most of these projects stall, and that version runs better in Caddi.
Why naming conventions drift in a coding agent
- Regenerated, not templated. The model produces a name each run, so date formats, separators, and abbreviations vary even for the same document type.
- Conventions are rules, not suggestions. 'YYYY-MM-DD_Client_Matter_DocType' must be enforced exactly; described in a prompt it's only approximated.
- Filing context gets lost. Naming usually pairs with filing to the right workspace and matter, which needs a real DMS connection, not just a string.
What slows it down in production
Inconsistent names quietly break retrieval, conflict checks, and audits, and a person ends up renaming files by hand to fix it, which is the work you were trying to automate in the first place.
| Claude Code | Caddi | |
|---|---|---|
| Prototype the workflow | Yes, with prompting | Yes, show it |
| Connect your real tools | Custom code & credentials | Connected for you |
| Run it identically every time | Output can vary | Deterministic code at runtime |
| Run unattended at volume | Needs supervision | Hundreds of runs a week |
| Maintenance | Yours | Done for you |
| Best for | Individual, judgment-heavy work | High-volume, high-accuracy work |
| Monitoring | Per script, set up by you | Caddi runs and the skills it hands Claude |
How Caddi runs the high-volume version end to end
Caddi enforces your naming and filing conventions as deterministic code: it reads each document's key attributes, builds the name to your exact pattern, and files it into the correct iManage, NetDocuments, or SharePoint workspace with the right metadata, identically every time, with an audit trail. You define the convention by showing Caddi how you do it.
Connect iManage to Caddi
Allow1Sign in to your tools
One click each for iManage, NetDocuments and Microsoft Outlook. No API keys.
Automate document filing.
ScreenshareType it2Show or tell it
Screenshare the job, or type it in chat. Change it any time.
- Running
14 done1 for review
3Watch it run
Caddi does the work and flags anything unsure for review.
Keep reading
- Document management automation
- Is Claude Code reliable for law firms?
- Caddi + NetDocuments
- Caddi + iManage
See deterministic automation in action
Caddi builds reliable automations from a screen recording and runs them across 150+ tools as deterministic code. Explore real workflows for law firms and RIAs & financial advisors, or book a demo to see your own workflow built live.
Caddi
See how Caddi AI Agents can rename files to your convention and file them in the right folder
Frequently asked questions
Why doesn't Claude Code follow my naming convention exactly?
Because it regenerates the file name each run instead of applying a fixed rule, so separators, date formats, and abbreviations drift. A convention needs deterministic enforcement.
How does Caddi guarantee exact file names?
It compiles your convention into deterministic code that builds every name to the same pattern and files it to the right workspace, identically on every run.
Does it work with my DMS?
Yes, Caddi connects to iManage, NetDocuments, SharePoint, and others to name, profile, and file documents automatically.
Do I have to choose between Claude Code and Caddi?
No. Caddi Discover finds and ranks the repetitive work across your systems, with evidence, so you know which of these to build first. High-volume, high-accuracy versions run in Caddi as deterministic code, the same way every run. Individual, judgment-heavy ones can be exported to Claude as a skill. Caddi Govern monitors every Caddi run, and the skills Caddi hands to Claude report each decision and API call back to it, in one firm-wide view.