terse-modelisted

# Terse Mode ## Identity When `/terse on` is active, CEO and non-VETO sub-agents produce minimally-verbose outputs. Target: -40-60% output tokens in exploratory research, bullet-list summaries, short factual answers. ## Rules - **Tool first. Result first.** State the action/outcome in line 1. - **Fragments OK** in: exploratory research, bullet lists, summaries, status updates, sanity checks. - **Full sentences REQUIRED** in: code-review findings, security findings, debate consensus.md, verdict.md artifacts, audit reports, compliance write-ups, adopter-facing documentation. - **Never truncate code** — code blocks remain verbatim. - **Never drop numbers** — quantitative claims carry units + source. - **Never use ellipsis to hide content** — if a finding matters, name it. ## VETO auto-off Terse mode AUTO-DISABLES for these roles (hardcoded in check_agent_spawn.py P0-TERSE-VETO guard): - code-reviewer (merge VETO) - security-engineer (auth/crypto VETO) - qa-architect (when writing verdict.md) - compliance-specialist (LGPD/GDPR findings) - CEO writing consensus.md or verdict.md For these roles, spawn prompt receives injected marker: `## TERSE-MODE-DISABLED — VETO role requires full-prose rationale`. ## Toggle - `/terse on` — activate for session - `/terse off` — deactivate - `/terse status` — print current state ## Measurement Session cost delta is tracked via ceo-cost.py. Terse-mode sessions emit `terse_mode_session` audit event with `mode_start_ts`, `mode_end