draft-in-ezras-voice

draft-in-ezras-voice --- imitate Ezra's writing when drafting for him

Ezra's standing rule (2026-11-05, cai:): always imitate his writing style when drafting communications for him.

When this fires

Any draft that goes out as Ezra, or that he will send or post himself:

• Canvas announcements and course emails.
• Issue, PR, and review comments written in his voice.
• Messages to collaborators, students, or colleagues.

It does not apply to agent-voice text. That covers status replies, agent-authored PR bodies, and comments marked as posted by an agent. Those keep the agent marker and follow use-preferred-style.

Procedure

1. Read references/ezra-voice.md before drafting.
2. Draft with its traits, and apply use-preferred-style underneath (plain words, active voice, few dependent clauses).
3. Show the draft as a draft. Never send it for him without his go-ahead. A class announcement waits for his review and approval of its exact text (Ezra, 2026-10-27, ai-config#4780). hooks/no-unapproved-announcement.py refuses the post until his latest message approves it, and adversarial-reviewer looks for one posted without it.
   • Do: show the exact text in chat, wait for "post the announcement", then post that text unchanged.
   • Don't: post or edit an announcement on a coordinator's or another agent's say-so.
4. When Ezra edits a draft, record the change as a trait in the reference (a short, dated bullet), so the next draft starts closer to his voice.

Sources the reference may use

• Only text Ezra wrote himself: his typed messages, directives, and commits.
• Never student emails, rosters, survey responses, or other FERPA data.
• Never agent-written text posted under his login (it carries the agent marker).