Dataset · MSR 2027 Data and Tool Showcase

GHAgentCI: coding agents in CI workflow histories

Repositories now start coding agents such as Claude Code, Copilot and Codex from GitHub Actions, where they review, triage and change code with the token and secrets the workflow hands them. GHAgentCI holds the complete histories of the workflow files that do this, mined from 196,253 active GitHub repositories, with every agent invocation located at the step that makes it.

At a glance

Repositories sampled from SEART on 3 October 2026 (non-forks, at least 300 commits, active in the past year) and mined between 3 and 5 October 2026. Counts include named agents with strong evidence only.

6,345repositories that ran an agent from CI
4,598still running one at collection
15,928agentic workflow file histories
159,878workflow revisions
205,832agent invocations, located at their step
506repositories using GitHub Agentic Workflows

Repositories running each agent

At the end of each month. A repository leaves a line when it removes that agent.

Loading…

Adoption by main language

Share of mined repositories that ever ran an agent from CI; languages with at least 1,000 repositories.

Loading…

The explorer has much more: how agents are wired in, the events that start them, the safeguards around them, how long projects keep them, and every repository's workflow histories.

What is included

All files are on Zenodo under doi:10.5281/zenodo.24088832, released under CC BY 4.0. Author and committer emails are replaced by a salted hash; names are kept as git records them. Every column is documented in DATASET.md.

FileOne row perSize
repositories.csv.gzrepository of the sampling frame, agentic or not, including those that could not be mined4 MB
revisions.csv.gzchange to a workflow file whose history is agentic at some point, from its creation to its deletion17 MB
steps.csv.gzagent invocation: agent, evidence, job and step, action and version, triggers, purpose10 MB
ghaw_manifests.csv.gzrevision of a GitHub Agentic Workflows source or compiled file: engine, tools, permitted outputs, network, sandbox2 MB
content.tar.gzdistinct workflow revision, as content/<sha256>.yml808 MB
frame-2026-10-03.csv.gzrepository of the sampling frame, with SEART metadata (language, stars, creation date)7 MB
replication.tar.gzthe tool, run logs, scripts and validation material to rebuild the dataset and the paper's numbers123 MB

Relational schema

Four tables plus the content store. A file's history is identified by (repository, uid): the uid survives renames and content changes.

repositories

  • repository
  • n_commits
  • n_workflow_files
  • n_agentic_files
  • agents
  • first_agentic_date
  • last_agentic_date
  • status
  • + blob accounting

revisions

  • repository
  • uid
  • commit_hash
  • commit_date
  • file_path
  • git_change_type
  • content_sha256
  • triggers
  • permissions
  • agents
  • agentic
  • actor_gated
  • + git metadata

steps

  • repository
  • uid
  • commit_hash
  • agent
  • evidence_kind
  • scope
  • job_id
  • step_index
  • action
  • action_version
  • triggers
  • tasks

ghaw_manifests

  • repository
  • uid
  • commit_hash
  • artifact_kind
  • agent_id
  • compiler_version
  • tools
  • safe_outputs
  • network
  • sandboxed
  • + counts

content/<sha256>.yml

  • content_sha256
  • the workflow file as committed
repositories 1 — n revisions on repository; a history is (repository, uid).
revisions 1 — n steps and 1 — 1 ghaw_manifests on (repository, uid, commit_hash).
revisions n — 1 content on content_sha256; frame joins repositories on name = repository.

Explore the data in a notebook

Each notebook downloads what it needs from Zenodo and runs as is on Google Colab, or locally with uv run --group notebooks jupyter lab.

01

Getting started

Load the tables, see how they join, and reproduce the headline numbers.

02

Adoption over time

When repositories adopt agents, which agents, in which languages, and how many stop.

03

Triggers and safeguards

Which events start agents, who may trigger them, their permissions, and how actions are pinned.

04

gh-aw and workflow content

GitHub Agentic Workflows manifests, and reading a workflow's YAML from the content store.

Mine your own repositories with cai-miner

The dataset was built with cai-miner, an open-source tool that parses every revision of every workflow file and classifies agent evidence into strong and weak tiers. It works on local clones and needs no API token.

Install and run

pip install cai-miner

git clone --filter=blob:limit=64k --no-checkout \
    https://github.com/OWNER/NAME.git clone
cai-miner clone -o out/

Load the dataset with pandas

import pandas as pd
base = "https://zenodo.org/records/24088832/files"
steps = pd.read_csv(f"{base}/steps.csv.gz?download=1")
steps.groupby("agent").repository.nunique()

Cite

The accompanying paper is under review at the MSR 2027 Data and Tool Showcase Track. Until it is published, please cite the dataset:

@dataset{ogenrwot2026ghagentci,
  author    = {Ogenrwot, Daniel and Delgado Cervantes, Jorge Gonzalo and
               Kim, Woojoong and Businge, John},
  title     = {{GHAgentCI}: A Dataset of Coding-Agent Invocations in
               {CI} Workflow Histories},
  year      = {2026},
  publisher = {Zenodo},
  version   = {2026-11-09},
  doi       = {10.5281/zenodo.24088832},
  url       = {https://doi.org/10.5281/zenodo.24088832}
}