Claudeagent and Claudeagentg1t keeps the actions workflows use from GitHub. When a job asks where an action is and g1t doesn't host it, g1t reads which commit its ref names from GitHub's advertisement (a tag before a branch, looked up again at most every five minutes, and the last commit it named while GitHub doesn't answer, which the job warns of), fetches that commit's archive once into R2, and the job downloads it from g1t; an archive over 32 MB, or one g1t can't fetch, is downloaded from GitHub by the job itself, now at the resolved commit. Jobs download every action their steps use before the first step, as GitHub's "Set up job" does, naming each commit in the log, and fail there when one can't be had. Archives unused for 30 days are swept hourly, and count toward no workspace's storage. The Actions guide says how.a4c727bHistory
- about.rs
- addresses.rs
- alerts.rs
- audit.rs
- billing.rs
- blobs.rs
- checks.rs
- deploy_keys.rs
- deployments.rs
- folios.rs
- lib.rs
- limits.rs
- logs.rs
- mcp.rs
- mirrors.rs
- notifications.rs
- oauth.rs
- oidc.rs
- openapi.rs
- operations.rs
- packages.rs
- people.rs
- pins.rs
- projects.rs
- protection.rs
- reference.json
- renamed.rs
- responses.rs
- rest.rs
- rules.rs
- run_artifacts.rs
- runners.rs
- security.rs
- token_policy.rs
- toolkit.rs
- tools.rs