[KongchangAI]
Paper

GHAgentCI

The first large-scale dataset targeting coding-agent invocations in CI, extracting coding-agent traces from GitHub repositories, including workflow histories, revisions and agent evidence records, accompanied by the open-source detection tool cai-miner.

Timeline (last 90 days)

Nov 13

GHAgentCI is the first large-scale dataset of CI coding agent invocations.

Unverified50%
Nov 13

GHAgentCI filtered more than 6K repositories containing traces of coding agents from over 196K GitHub repositories.

Unverified50%
Nov 13

GHAgentCI constructed over 15K workflow histories, covering more than 159K revisions and more than 205K agent evidence records.

Unverified50%
Nov 13

Each agent evidence record in GHAgentCI is linked to a workflow trigger and declared execution constraints.

Unverified50%
Nov 13

Exploratory analysis of the dataset shows that the integration methods of coding agents are shifting: command-line and GitHub Agentic Workflow integrations are gradually increasing, declared security safeguards vary, and detected agent workflows are frequently removed.

Unverified50%
Nov 13

GHAgentCI retains historical revisions and evidence records of workflows, enabling researchers to trace the full lifecycle of agent integration from introduction through adjustment to removal.

Unverified50%

All Facts (6)

Source Articles