A local harness for agent teams on Claude Code and Codex CLI: only what they prove counts as done. Waves and lanes split the work, reviewers from another model family judge it, councils keep the dissent that lost, typed decisions with Jev are measured before they are trusted, and approvals are bound to the bytes they approved. No model calls.
6-step wizard (check_python->check_git->check_deps->configure->validate->generate_and_summary) that scaffolds a new agent/skill. Non-interactive --interview/--answers + --demo; skip-exists with --force.
Handoff-v1.1: a session survives a stop/clear/crash without losing its next step. Schema with git-block + re_derive_cmd (every number carries the command that re-derives it live) + verify_first_cmd (check live before replaying any action t…
A squad of 12 roles available as slash commands and real subagents, with explicit tools and read-only QA, plus 3 token-safe parallel reading/consolidation skills. Ships no tasks/templates tree: the role commands that expect one (*create, *…
Standalone operational learning loop: a failure becomes knowledge. After every Bash command that fails, the postflight records the event classified by error family; when the same type recurs N times within a window it becomes a GOTCHA -- a…
Service health probe (http/cmd) driven by profile.yaml + a statusline segment with per-service detail (api:OK db:DOWN), cache-first (the statusline never touches the network). Embedded doctrine: SERVICE health != DATA health. +dashboard-bu…
N sessions without collisions. Lane board, cross-model maker!=checker, per-directory lock, git-guard and territory-guard. The checker_router detects Codex, Cursor and Gemini and picks a provider different from the maker's. Best-of-N: compe…
The harness's operational module: executable gates, governed loops, standalone pass@k/pass^k, audit/enforce policy, preflight, spec-driven planning, two read-only checkers and four review lenses answering in hpp.findings/v1.