The resident

A resident is the repo-aware coworker behind the remote door. Claude Code or Codex is the medium for a particular run; the resident is the continuity that survives when that process exits and another one starts.

Each repo gets:

• working memory for the thread it is carrying;
• project knowledge and recent activity;
• a playbook for how work should be done there;
• the live run facts needed to deliver a result safely.

What arrives in a wake

Before the task, brnrd assembles a compact orientation layer:

• the repo contract and current run facts;
• the resident’s working memory and playbook;
• recent project activity and pitfalls relevant to the task;
• queue, quota, delivery, environment, and branch posture;
• the request and the conversation that led to it.

The longer tail stays pull-based. Project knowledge may live in a private account home, a repo-owned knowledge base, or ordinary project docs. brnrd injects a useful slice and points the resident to the rest when needed. It does not assume that a repo-local kb/ directory is the only or default shape.

That split gives a wake enough continuity to begin somewhere without making every run reread the project’s entire history.

One resident, different models

Changing the Shell or Core does not create a new coworker. A cheaper Core can handle routine work, while the same resident can hand a hard pass to a stronger local Core with its context intact. See Models & quota.