Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Intent-driven spec precipitation — state a constraint in natural language (加一条规范：禁止用 any / 记录架构约束：服务间走 gRPC / 质量规则：覆盖率≥80%) and the workflow infers the category and records a <spec-entry>. Spec = 项目约束规则（编码规范、架构约束、质量标准）；可复用知识文档走 /maestro-knowhow capture。Triggers on "maestro-spec add", "记录规范", "添加约束", "添加规则", "加一条规范", "spec add". Terminology：spec = project constraints/rules (<spec-entry>). Reusable knowledge documents use /maestro-knowhow capture. Learning discoveries from /maestro-learn use <lear
.claude/skills/catlog22-maestro-spec/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 141% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 150% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 301% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -2% | 0% |
<purpose> Intent-driven spec precipitation path (沉淀路径) — records project constraint rules. No fixed grammar — state the constraint; the specs-add step infers the category and scope and formats the <spec-entry>. Explicit form still works as a shortcut: [--scope <scope>] <category> <content>.
Categories: coding · arch · quality · debug · test · review · learning · ui Scopes: project (default) · global · team · personal </purpose>
<dispatch> Execute the specs-add step inside a v3 Session: open one with maestro session open "<objective>" --id <slug> --chain specs-add --participant {p} --actor {a} --request-id {r} --reason "<reason>" --json and dispatch with fenced maestro run next --session {session_id} ... --expected-orchestration-revision {rev} --json (or self-start with maestro run create specs-add [args...] --session {session_id} ... --json), passing the full $ARGUMENTS as the step input (repeatable --arg / positional passthrough per the command contract; the add keyword is implied). Read context read-only with maestro session status / maestro session resume-view; maestro run complete {run_id} ... --advance publishes outputs and auto-stages knowledge candidates. (v2's run skill dispatcher is removed from the v3 surface.)
The step infers category + scope + content and appends the entry.
<category> <content> and --scope/--uid flags still work and override inference.</dispatch>
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 12,048 | 9,600 | -20% | 1 | 1 | 0% | 1,900 | 1,202 | -37% | 0 | 0 | — |
case-02 | fail→pass | 6,911 | 10,662 | +54% | 1 | 1 | 0% | 911 | 2,197 | +141% | 0 | 0 | — |
case-03 | fail→fail | 4,488 | 6,483 | +44% | 1 | 1 | 0% | 633 | 738 | +17% | 0 | 0 | — |
case-04 | fail→pass | 3,275 | 3,748 | +14% | 1 | 1 | 0% | 425 | 1,062 | +150% | 0 | 0 | — |
case-05 | fail→pass | 2,945 | 6,506 | +121% | 1 | 1 | 0% | 350 | 1,402 | +301% | 0 | 0 | — |
case-06 | fail→pass | 12,251 | 6,733 | -45% | 1 | 1 | 0% | 1,942 | 1,537 | -21% | 0 | 0 | — |
case-07 | fail→pass | 11,943 | 9,241 | -23% | 1 | 1 | 0% | 1,997 | 1,956 | -2% | 0 | 0 | — |
case-08 | fail→pass | 3,019 | 6,463 | +114% | 1 | 1 | 0% | 361 | 1,549 | +329% | 0 | 0 | — |
case-09 | fail→pass | 10,490 | 8,130 | -22% | 1 | 1 | 0% | 1,699 | 1,930 | +14% | 0 | 0 | — |
case-10 | fail→fail | 8,086 | 3,730 | -54% | 1 | 1 | 0% | 1,176 | 1,138 | -3% | 0 | 0 | — |
case-11 | fail→pass | 11,568 | 8,204 | -29% | 1 | 1 | 0% | 1,599 | 1,771 | +11% | 0 | 0 | — |
case-12 | fail→pass | 5,113 | 10,418 | +104% | 1 | 1 | 0% | 694 | 2,224 | +220% | 0 | 0 | — |
case-13 | pass→fail | 6,396 | 3,215 | -50% | 1 | 1 | 0% | 815 | 913 | +12% | 0 | 0 | — |
case-14 | fail→pass | 5,689 | 10,932 | +92% | 1 | 1 | 0% | 792 | 1,789 | +126% | 0 | 0 | — |
case-15 | fail→fail | 11,261 | 7,789 | -31% | 1 | 1 | 0% | 1,633 | 864 | -47% | 0 | 0 | — |
case-16 | fail→pass | 13,164 | 12,669 | -4% | 1 | 1 | 0% | 2,005 | 1,709 | -15% | 0 | 0 | — |
case-17 | fail→pass | 15,346 | 7,208 | -53% | 1 | 1 | 0% | 2,442 | 1,765 | -28% | 0 | 0 | — |
case-18 | pass→pass | 10,231 | 13,961 | +36% | 1 | 1 | 0% | 1,651 | 2,102 | +27% | 0 | 0 | — |
case-19 | fail→pass | 11,763 | 8,358 | -29% | 1 | 1 | 0% | 1,520 | 1,106 | -27% | 0 | 0 | — |
case-20 | pass→pass | 2,936 | 7,813 | +166% | 1 | 1 | 0% | 369 | 1,140 | +209% | 0 | 0 | — |
case-21 | pass→pass | 6,893 | 13,558 | +97% | 1 | 1 | 0% | 1,149 | 2,708 | +136% | 0 | 0 | — |
case-22 | fail→fail | 11,940 | 6,740 | -44% | 1 | 1 | 0% | 1,736 | 779 | -55% | 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 19 counted toward the lift figure. The other 3 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 +55 percentage points is the difference between those two pass rates over the 19 comparable cases. 2 cases got worse with the skill loaded, and they are 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.