Have an agent work through a large document batch

A user asked Dot to handle 80 documents in a GitHub-based workflow, but it halted around document 60 and kept running into permission blocks.

Done with Dots· Intermediate· Oct 2026

What they did
The poster set Dot loose on coding tasks connected to a GitHub repository they controlled, and also experimented with running it inside its own VM. They instructed it to move through 80 documents, use the browser, and handle PRs, while Dot repeatedly paused to ask for approval. They also tried to broaden its access with authorizations, memory, custom rules, and direct git and gh CLI access.
How it went
Dot got through about 60 documents, then stopped after its Sentry safety check judged the remaining batch too risky. The breakdown was mostly operational: it would not commit or merge, and browser use, file access, git and gh commands, and PR merging were denied or hung up.
Tips
For a job this long, expect to re-issue permissions and keep an eye on whether safety checks are carrying over into later steps. It helps to ask for a clear progress log and a final list of unfinished items before any commit or merge.

Try it yourself with Dots

Take on [number] documents in [folder/location]. For each one, [summarize/extract/classify] the needed information and track progress in a checklist with one line per document. Keep working until every item is complete, and if anything blocks you, stop and name the exact documents left plus the reason. Before you make, commit, merge, delete, or submit any change, show me the exact diff or action plan and wait for my approval. After I approve, [write/save] the results to [output file].

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.