Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Intent-driven knowledge-store and Run knowledge lifecycle management — audit/prune, stage candidates (with signal recording), review/resolve/promote candidates, harvest artifacts, or manage wiki/domain knowledge.
.claude/skills/catlog22-maestro-knowledge/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 91% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 100% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 161% | 0% |
<purpose> Intent-driven knowledge-store management. No fixed grammar — state your intent; the command classifies it and runs the matching workflow or direct lifecycle command. Explicit keywords still work as deterministic shortcuts.
| Operation | Keywords | Step | |-----------|----------|------| | audit | audit / 审计 / 清理 / prune / 检查知识库 | knowledge-audit | | review | review / 审查 / 证据 / 下一步 / 匹配 / 去重 / 冲突检测 / 裁决 / 候选 / backlog | maestro knowledge review <session-id> [--refresh] [--resolve <id> --as <choice> --reason "..."] | | stage | stage / 暂存 / candidate / 沉淀候选 / cited / validated / contradicted / 记录命中关系 | maestro knowledge stage ... [--signal <signal> --signal-ids <ids>] | | promote | promote / 晋升 / 发布候选 | maestro knowledge promote ... [--all] | | harvest | harvest / 提取 / 收割 / 从工件 | harvest | | wiki | wiki / 知识图谱 / 连接 / 摘要 / 健康 | wiki-manage / wiki-connect / wiki-digest | | extractors | extractors / 抽取器 / 生成抽取规则 | extractors | | domain | domain / 领域术语 / 注册术语 / term | domain-add | </purpose>
<dispatch> Classify the intent in $ARGUMENTS into one operation, then execute the chosen step (or the direct maestro knowledge CLI) and follow it completely. Step-based operations run inside a v3 Session: open one with maestro session open "<objective>" --id <slug> --chain <step> --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 <step> [args...] --session {session_id} ... --json); read context read-only with maestro session status / maestro session resume-view, and maestro run complete {run_id} ... --advance publishes outputs and auto-stages knowledge candidates. (v2's run skill dispatcher is removed from the v3 surface.)
review / stage / promote map directly to the corresponding maestro knowledge CLI. review --refresh includes reconciliation; review --resolve includes disposition resolution; stage --signal --signal-ids includes signal recording. Preserve stable knowledge IDs, graph aliases, Run ID, Session ID, signal, candidate ID, disposition, target, and reason exactly; do not translate these operations into direct spec/knowhow writes.connect/连接 → wiki-connect; digest/摘要 → wiki-digest; health/search/cleanup/stats/健康/检查/_(none)_ → wiki-manage.maestro knowledge stage knowhow ...; project knowhow is written only by explicit promotion. Outside a Run, direct /maestro-knowhow capture remains available.maestro knowledge stage <target> "<title>" --content-file <path|->. Inline positional content containing spaces, quotes, unicode (e.g. …), newlines, or leading dashes is misparsed and shifts later arguments.--signal-ids takes comma-separated IDs (--signal-ids spec:project:a,knowhow:b); space-separated values leak into positional arguments and corrupt the stage call.maestro knowledge review <session-id> as the human review surface. It shows fresh/missing/stale receipts, diversified evidence-backed matches, and copyable promote commands. --refresh reconciles all candidate source Runs. --resolve <candidate-id> --as <choice> --reason "..." resolves a candidate inline before displaying the refreshed view.review --resolve.promote --all promotes all eligible pending candidates (observed-only emits a warning); --include-observed has been removed.audit --prune --apply may only perform backed-up soft lifecycle transitions. Never physically delete knowledge or prune solely because it has low usage.</dispatch>
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | fail→fail | 8,404 | 6,518 | -22% | 1 | 1 | 0% | 1,082 | 1,462 | +35% | 0 | 0 | — |
case-01 | pass→fail | 10,542 | 10,072 | -4% | 1 | 1 | 0% | 1,670 | 1,853 | +11% | 0 | 0 | — |
case-02 | fail→pass | 6,289 | 2,404 | -62% | 1 | 1 | 0% | 1,005 | 1,542 | +53% | 0 | 0 | — |
case-03 | fail→fail | 5,184 | 7,549 | +46% | 1 | 1 | 0% | 706 | 1,702 | +141% | 0 | 0 | — |
case-04 | fail→fail | 5,826 | 7,159 | +23% | 1 | 1 | 0% | 959 | 1,608 | +68% | 0 | 0 | — |
case-05 | fail→fail | 4,673 | 7,557 | +62% | 1 | 1 | 0% | 722 | 1,746 | +142% | 0 | 0 | — |
case-06 | fail→pass | 11,066 | 2,558 | -77% | 1 | 1 | 0% | 818 | 1,562 | +91% | 0 | 0 | — |
case-07 | fail→fail | 10,368 | 5,296 | -49% | 1 | 1 | 0% | 614 | 1,492 | +143% | 0 | 0 | — |
case-08 | fail→pass | 8,702 | 4,733 | -46% | 1 | 1 | 0% | 1,262 | 1,406 | +11% | 0 | 0 | — |
case-09 | fail→fail | 8,521 | 9,364 | +10% | 1 | 1 | 0% | 1,443 | 1,688 | +17% | 0 | 0 | — |
case-10 | fail→pass | 12,288 | 14,553 | +18% | 1 | 1 | 0% | 1,810 | 3,616 | +100% | 0 | 0 | — |
case-11 | fail→fail | 9,377 | 10,203 | +9% | 1 | 1 | 0% | 1,371 | 2,956 | +116% | 0 | 0 | — |
case-12 | fail→pass | 7,191 | 8,634 | +20% | 1 | 1 | 0% | 1,057 | 2,754 | +161% | 0 | 0 | — |
case-13 | pass→pass | 18,991 | 6,204 | -67% | 1 | 1 | 0% | 2,782 | 1,619 | -42% | 0 | 0 | — |
case-14 | fail→pass | 2,671 | 7,125 | +167% | 1 | 1 | 0% | 319 | 2,431 | +662% | 0 | 0 | — |
case-15 | fail→pass | 5,178 | 17,203 | +232% | 1 | 1 | 0% | 678 | 3,256 | +380% | 0 | 0 | — |
case-16 | fail→fail | 4,379 | 3,065 | -30% | 1 | 1 | 0% | 636 | 1,611 | +153% | 0 | 0 | — |
case-17 | fail→pass | 13,553 | 8,568 | -37% | 1 | 1 | 0% | 1,924 | 2,483 | +29% | 0 | 0 | — |
case-19 | pass→pass | 6,887 | 7,169 | +4% | 1 | 1 | 0% | 943 | 2,374 | +152% | 0 | 0 | — |
case-20 | pass→pass | 2,705 | 2,781 | +3% | 1 | 1 | 0% | 396 | 1,619 | +309% | 0 | 0 | — |
case-21 | pass→pass | 3,202 | 2,902 | -9% | 1 | 1 | 0% | 426 | 1,568 | +268% | 0 | 0 | — |
case-22 | pass→pass | 2,695 | 3,433 | +27% | 1 | 1 | 0% | 411 | 1,705 | +315% | 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 14 counted toward the lift figure. The other 8 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 +32 percentage points is the difference between those two pass rates over the 14 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.