Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Setup Maestro in a project using or adopting Maestro: use for init/install/sync/doctor, global skills, hooks, harness setup, or agent integration diagnosis/repair.
.claude/skills/reinamaccredy-maestro-setup/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -3% | 0% |
Tune a Maestro-enabled repository harness from current repository evidence.
Activate with a known session id: maestro hook record --event skill_activation --skill maestro-setup --session <session_id>
Recipe checkpoint: Maestro's main workflow is the loop. Use maestro status for current state and maestro loop next as the read-only router when the next lifecycle is not obvious. Setup writes still use the existing Maestro verbs named by this skill (init, install, sync, doctor, upgrade, uninstall, and shell-init), not hidden setup state. Rule: loop next recommends; outcome/proof/memory verbs write. maestro loop next --chain explains current chain position without writing; maestro loop outcome appends structured outcomes and transition receipts after native work; maestro loop trace <card> audits card-scoped receipts; maestro loop improve is a read-only proposer. Do not use hidden stores, hidden schedulers, silent recipe mutation, or proof/QA bypass.
maestro init.maestro install.running binary with minimal repeated token cost.
maestro doctor reports setup or local agent integration problems.pwd, then maestro status.setup, take the cheap rerun path before any broader read-in:
maestro sync --dry-run.maestro sync.maestro sync, then report only changed resources,backup paths, and blockers.
maestro sync --global-skills only when the user asks for globalskills or maestro doctor reports global skill drift.
maestro doctor, then maestro status.maestro init --dry-run.operating on <path> and that path is not the intendedcurrent project root, stop and ask the user to pick or create the root before any write.
maestro init --yes.maestro doctor.maestro install --agent codexunless the user asked for another agent.
repo root, and under each folder matched by the projects: globs in .maestro/harness/harness.yml, read AGENTS.md, CLAUDE.md, README.md, and top-level docs/*.md. Stay shallow (one level per location, no deep crawl) and skip outsized files (roughly 64 KB and up) so a vendored doc dump cannot flood context. With no projects: declared, this is the repo root alone.
per project (a single section when nothing is declared). This is read-in only: never write maestro-managed guidance into a sub-project's own AGENTS.md/CLAUDE.md -- maestro install/sync write managed blocks at the repo root alone. Cite the inspected files; do not tune from guesses.
maestro doctor, then maestro status.maestro init --dry-run writes nothing; use it before init writes.maestro install --dry-run beforeagent mirror writes, maestro sync --dry-run before setup refresh writes, maestro capability to explain optional provider state, and maestro maturity to read context, proof gaps, friction, maturity level, and next owner before claiming setup readiness. Generated CLI references prove command shape; Harness and targeted skills teach the workflow.
maestro init --yes keeps existing files and creates missing files.maestro sync --dry-run is the cheap already-initialized update check; do notrun broader setup read-in when it proves the setup is current and no doctor blocker remains.
maestro init --force only for deliberate refresh; it backs up managedfiles first.
memory.
maestro-managed guidance into a sub-project's AGENTS.md/CLAUDE.md.
maestro status.current-state read, and every changed managed resource is named with its backup path or blocker.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 4,363 | 6,582 | +51% | 1 | 1 | 0% | 176 | 1,484 | +743% | 0 | 0 | — |
case-01 | fail→fail | 5,475 | 6,393 | +17% | 1 | 1 | 0% | 177 | 1,452 | +720% | 0 | 0 | — |
case-03 | fail→fail | 5,039 | 7,464 | +48% | 1 | 1 | 0% | 198 | 1,498 | +657% | 0 | 0 | — |
case-04 | pass→pass | 4,524 | 3,791 | -16% | 1 | 1 | 0% | 770 | 1,914 | +149% | 0 | 0 | — |
case-05 | pass→pass | 8,055 | 5,567 | -31% | 1 | 1 | 0% | 1,253 | 2,104 | +68% | 0 | 0 | — |
case-06 | fail→pass | 13,517 | 4,170 | -69% | 1 | 1 | 0% | 2,022 | 1,955 | -3% | 0 | 0 | — |
case-07 | pass→pass | 16,899 | 5,689 | -66% | 1 | 1 | 0% | 1,806 | 1,868 | +3% | 0 | 0 | — |
case-08 | fail→pass | 13,785 | 4,608 | -67% | 1 | 1 | 0% | 1,920 | 1,823 | -5% | 0 | 0 | — |
case-09 | fail→pass | 13,548 | 10,175 | -25% | 1 | 1 | 0% | 1,849 | 1,995 | +8% | 0 | 0 | — |
case-10 | fail→pass | 18,064 | 5,466 | -70% | 1 | 1 | 0% | 2,124 | 1,743 | -18% | 0 | 0 | — |
case-11 | fail→fail | 11,488 | 2,833 | -75% | 1 | 1 | 0% | 1,784 | 1,534 | -14% | 0 | 0 | — |
case-12 | fail→pass | 10,152 | 2,020 | -80% | 1 | 1 | 0% | 1,620 | 1,566 | -3% | 0 | 0 | — |
case-13 | fail→pass | 20,871 | 3,105 | -85% | 1 | 1 | 0% | 2,869 | 1,569 | -45% | 0 | 0 | — |
case-14 | fail→pass | 18,386 | 2,582 | -86% | 1 | 1 | 0% | 2,467 | 1,613 | -35% | 0 | 0 | — |
case-15 | fail→pass | 13,183 | 2,044 | -84% | 1 | 1 | 0% | 1,624 | 1,489 | -8% | 0 | 0 | — |
case-16 | fail→pass | 15,321 | 2,206 | -86% | 1 | 1 | 0% | 1,393 | 1,588 | +14% | 0 | 0 | — |
case-17 | fail→pass | 13,714 | 2,570 | -81% | 1 | 1 | 0% | 2,055 | 1,586 | -23% | 0 | 0 | — |
case-18 | fail→pass | 70,816 | 2,004 | -97% | 1 | 1 | 0% | 3,001 | 1,517 | -49% | 0 | 0 | — |
case-19 | pass→pass | 10,731 | 3,416 | -68% | 1 | 1 | 0% | 1,626 | 1,664 | +2% | 0 | 0 | — |
case-20 | pass→pass | 12,763 | 14,538 | +14% | 1 | 1 | 0% | 1,910 | 3,601 | +89% | 0 | 0 | — |
case-21 | pass→fail | 15,810 | 6,299 | -60% | 1 | 1 | 0% | 2,305 | 1,409 | -39% | 0 | 0 | — |
case-22 | pass→pass | 10,497 | 14,886 | +42% | 1 | 1 | 0% | 2,054 | 3,266 | +59% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +45 percentage points is the difference between those two pass rates over the 18 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.