brieflisted

# Brief: what happened while you were away An agent can work for hours without you. When you come back, the question is not what is configured, it is what happened, what is stuck, and what needs you. This skill answers that in one screen. It is the mirror image of the checkpoint. The checkpoint writes durable context **for the model**, so reasoning survives a compaction. This writes a briefing **for the person**, once, on screen, and it dies when you have read it. That is why it stores nothing. ## Rules that are not negotiable - **It writes no file.** No report, no cache, no marker. A briefing is re-derivable, and a stored one only goes stale. The window comes from the transcript, which already records every earlier run of this command. - **It changes nothing else either.** No commits, no flushing the ledger, no fixing what it finds. It names the next steps and stops. Self starting maintenance was measured as noise in this kit before, and it is not coming back through a status command. - **It reports in the user's language**, matching the language they write in, and in that language's own characters. Never transliterate an accent or a diacritic away: a project may ban them in file and code names for portability, and that rule stops at the filename. ## The process ### 1. Get the facts Run the digest, from the project root, exactly like this: ```bash T=$(ls -d .claude/tools tools "${CLAUDE_PLUGIN_ROOT:-/nonexistent}/tools" 2>/dev/null | head -1); bash "$T/br