PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteYoshihisa Kaino counted 10,261 of 26,251 commits across his own repositories as carrying a Claude co-author trailer—39% in the snapshot reported in his September 2026 article. That is a count of commits with declared co-authorship, not a measurement of how many lines an AI wrote or how much work it did. The result also changes sharply when private repositories are excluded: Kaino’s public-only count was 126 of 5,688 commits, or 2%.
What the commit count measures—and what it does not
Kaino’s question was personal: he knew his coding workflow had changed, but had only a feeling about how much AI was involved. He looked for Git commit-message trailers such as Co-Authored-By and counted commits that named Claude as a co-author. In one repository, he counted 960 such commits out of 1,107. Across all repositories he had touched, he reported 10,261 of 26,251 commits, or 39%.
As an Amazon Associate I earn from qualifying purchases.
A co-author trailer is a useful, searchable record of declared collaboration. It does not establish what share of a commit’s code was generated, accepted unchanged, or materially altered by an AI system. It also misses AI-assisted work when no such trailer was added. The numbers describe Kaino’s recorded Git history, not the share of code written with AI by developers generally.
“I knew my workflow had changed. I didn’t know it had changed that much, and I couldn’t have told you so with a number before this,” Kaino wrote.
#1 Best Overall
Why repository visibility changes the answer
Kaino compared his private-inclusive count with a query limited to public commits. The scopes and denominators differ:
| Scope in Kaino’s report | Claude-tagged commits | All counted commits | Share |
|---|---|---|---|
| Repositories visible to him, including private repositories | 10,261 | 26,251 | 39% |
| Public-only query | 126 | 5,688 | 2% |
These are Kaino’s own snapshot counts, not an apples-to-apples comparison of two estimates of the same complete history. A public-only search cannot include commits in private repositories that the querying identity cannot see. For someone trying to understand their own workflow, the public number can therefore be a poor proxy for their full count.
Kaino’s project, cocommit, is designed to run the calculation using the user’s own GitHub token, including in that user’s CI. He argues for this approach rather than a hosted service that pools users’ credentials. That is a design rationale, not a blanket security guarantee: users still need to review what a tool requests and how they configure their credentials.
How Kaino counted commits locally
His initial method searched commit-message bodies for the trailer. For a repository, the core command was:
git log --format='%b'That prints commit-message bodies, which can then be searched for a Co-Authored-By line. The repository-level count is straightforward to inspect, but it depends on the trailers actually present in that repository’s history. A trailer-based total is not a direct scan of source code or a reconstruction of every tool-assisted interaction.
Using GitHub commit search carefully
Kaino tested GitHub commit search with a co-authored-by: qualifier. In his tests, different email addresses produced different totals, an invented qualifier returned zero, and omitting the qualifier returned the unfiltered total. He inferred that the search parser recognized the qualifier, but says it was not documented in the commit-search documentation he checked. That observation is not a documented API contract, and behavior may change.
His practical safeguard was to check that a filtered result plausibly narrowed the unfiltered result. A zero count or a filtered count at least as large as the unfiltered total should be treated as unusable for this purpose: the filter may have matched nothing, or may not have been applied. This is a defensive check based on his observed behavior, not proof that a result is correct.
Free tools Windows power users keep installed
One-click scans. No signup required.
He also notes that his tool reads the API’s total_count, rather than enumerating repository names, commit messages, or diffs. In his account, commit-search enumeration is capped at 1,000 returned items even when the total count is higher. Anyone implementing a similar workflow should verify current GitHub documentation for the API behavior and limits before relying on them.
Why matching by email is safer than matching by name
Agent display names are not necessarily unique or stable. Kaino initially searched by display name and got 10,262 matches for Claude; searching by email returned 10,255. He traced the difference to other co-authors whose names matched the text, including a human collaborator. His conclusion was to identify known agents by the email address in the trailer rather than by display name alone.
Rank #4
He gives Aider as another reason to prefer email: its visible name can include the model and vary from commit to commit, while the email address remained stable in his examples. The addresses below are the values and public-commit totals Kaino reported from his own inspection; they are snapshots, not official identity specifications or current platform statistics.
| Agent named in Kaino’s article | Trailer email address(es) reported | Public commit total |
|---|---|---|
| Claude | [email protected] |
126 |
| Cursor | [email protected] |
Not stated in the article’s summary |
| Copilot | [email protected] |
Not stated in the article’s summary |
| Codex | [email protected] and [email protected] |
Not stated in the article’s summary |
| Devin | Not stated in the article’s summary | Not stated in the article’s summary |
| Gemini | Not stated in the article’s summary | Not stated in the article’s summary |
| Jules | Not stated in the article’s summary | Not stated in the article’s summary |
| Aider | Not stated in the article’s summary | Not stated in the article’s summary |
| Amp | Not stated in the article’s summary | Not stated in the article’s summary |
The article names these agents and reports an observed public-commit table, but the available figures here do not establish the other address-and-count pairs. Avoid treating a missing value as zero or filling it from current assumptions. For the same reason, even a known address should be checked against the actual trailers in the history being counted.
A total can hide when the workflow changed
A cumulative percentage answers how many counted commits carry a trailer across the selected history; it does not show when the pattern emerged. Kaino separately reported 42 commits in October 2025 and 2,067 in September 2026, which he described as 49 times higher. He also said recent months were above 80%. Those figures describe his personal timeline, not a general developer trend, and monthly commit volume is not a measure of code volume or effort.
Best Value
For an individual or team, a time series can make a history more informative: it can show whether trailer use increased, stayed steady, or varied across periods. It still tracks the recording convention and repositories included, so changes in how consistently trailers are added can affect the apparent trend.
What to take from the project
Kaino describes cocommit as MIT-licensed, dependency-free, and runnable with npx. Those project details are his description in the September 20, 2026 article and may change. The central takeaway is less about the tool than the measurement: Git metadata can turn a vague personal impression into a reproducible count, provided the count is described narrowly and its visibility, identity matching, and trailer-based limits are made explicit.
Read Yoshihisa Kaino’s original article on DEV Community.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
