Dev team applies strict public-reply guardrails to private DM bot responses

A development team building an AI agent that responds to Bluesky feedback extended their full set of public-reply safety controls to direct messages, rejecting the assumption that private channels require less oversight. The same tone review, batch caps, idempotency checks, and daily budget limits used for public replies were imported directly into the DM handler rather than rebuilt separately. Two public-reply mechanisms — liking competitor posts and thread re-fetching — had no DM equivalents, forcing the team to design replacements, including logging skipped competitor DMs with explicit reasons. The most notable addition was a "conversation-recency check" that blocks the bot from replying to an older message if a newer unanswered one already exists in the same conversation, preventing robotic out-of-order responses. Every possible send outcome, including a new "stale-superseded" state, was given a distinct label to make the system's behavior fully auditable.