← ClaudeAtlas

maestro-token-efficiencylisted

Use when improving Maestro token efficiency, Codex fit, benchmark context hygiene, or agent orientation without weakening verification quality.
mbanderas/maestro · ★ 3 · AI & Automation · score 74
Install: claude install-skill mbanderas/maestro
# Maestro Token Efficiency Use this skill for Maestro repo changes that affect agent context, instruction loading, benchmark search hygiene, or Codex behavior. 1. Read `docs/agent-map.md` first for navigation. 2. Prefer benchmark summary files over raw streams unless doing forensic scoring. Use `rg --no-ignore` for raw stream audits. 3. Preserve Maestro's quality bar: no weaker verification status, no hidden-oracle exposure, and no default behavior change without benchmark evidence. 4. For Codex-specific behavior, check `docs/codex.md` and current official Codex docs before changing durable guidance. 5. Keep always-loaded instruction files stable. Put volatile run notes, dated scratch state, machine-local paths, and benchmark deltas in summaries or checkpoints, not `AGENTS.md`/adapters. Before completion after edits, run: ```bash npm test npm run lint npm run bench-verify ```