Agents and memory, checks and conflicts, profiles, slug renames, custom domains
Agents: every agent run is a record (kind, pull request, model, status, current step and a timeline of steps reported from the sandbox, cost), with Stop and Message. Projects get an Agents page (At work, Sessions, Memory), workspaces an Agent fleet and Workspace memory, and pull requests an Agent panel and a working badge in the list. Memory has two levels, project and workspace; agents write it with remember and read it with recall (MCP and REST), every g1t run starts with it, members curate it, and anything that looks like a secret is refused. Pull requests: the merge box lists every check and workflow job, failed ones open with their output (colours, line numbers, exit code), with re-run and history. Mergeability is worked out on every push: changed paths first, then a metered sandbox probe with git merge-tree when both sides touched a file. Conflicts are listed up front, block the merge, and are handed to the agent with the files named. Profiles at /u/<username>: name, picture, bio, location, website, joined; pull requests and issues with filters, only on repositories the viewer may read; workspaces shown only where membership is already shared or public. Authors link to their profiles; profiles get social cards. Workspace slugs can be renamed by owners. Every service moves its rows on workspace.renamed, idempotently and in any order; deployed apps are rebuilt under their new names with the old addresses redirecting; old links and git remotes redirect for 90 days. Repositories keep a store key that never changes, so a freed name can never adopt old git data. Custom domains for production: add a hostname (and its www or apex twin), get the DNS records to set, certificates by Cloudflare for SaaS, status kept current by the sweep; three included with Deployments, then $0.12 a month. The wildcard route waits for SaaS to be enabled on g1t.page. The sidebar no longer mistakes /u/<name> for a project.
No changes