Agent of the Day – October 26, 2026

Oct 26, 2026

Every codebase has a file that started small and kept saying yes. Today’s spotlight goes to the workflow whose whole job is to notice.

Agent of the Day: Daily File Diet

Daily File Diet runs on a weekday schedule (around 13:00 UTC) and can also be started manually. It imports a shared Go source analysis, looks for files past a size threshold, and turns its findings into a single issue labelled refactoring and code-health. Issues expire after two days, and the workflow skips a run if a [file-diet] issue is still open, so it never piles up duplicates.

Its permissions are read-only for contents, issues and pull requests. The only write path is the issue it files through safe outputs.

What the logs show

The latest completed run, run 40338270065, finished successfully in about 9 minutes:

Outcome: success in one agent turn, with no reported workflow errors.
Output: one issue, #72111, flagging pkg/cli/token_usage_subagent_session.go at 1,557 lines against an 800-line threshold.
Network: the firewall recorded 20 requests, all allowed, none blocked.
Stability: against an earlier successful baseline (run 39956353556), turns and write-capable posture were unchanged.
Cost: the audit flagged medium “high token usage” at roughly 227k tokens, which is a reasonable thing to watch for a workflow that reads large files.

The issue it wrote

The report is more than “this file is big.” It maps the file into concerns: event schema types, unified token usage reading, session parsing, Copilot and Pi observers, error helpers, and row aggregation. It notes test coverage (a 645-line test file, roughly a 0.41 ratio) and proposes a split into focused files. That is the difference between a nag and a starting point: a maintainer or a coding agent can pick it up and begin.

Why it’s worth a look

Size thresholds are blunt, but a tool that applies them daily, explains the seams, and cleans up after itself keeps technical debt visible without creating noise. You can reproduce this evidence with gh aw logs and gh aw audit.

Try it

Browse the workflow sources, borrow the pattern for your own repository, and tell us what your agents find at github/gh-aw.