Install any skill in seconds. Free to start, no credit card required.
Get Started Free →MUST activate when you ARE an orchestrator — you are the top-level agent, you spawn subagents, you delegate work, you coordinate parallel or sequential execution. Defines delegation quality, subagent dispatch, routing, review, and ownership protocol.
.claude/skills/griddynamics-orchestrator-contract/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 52% | 0% |
<orchestrator_contract>
<process>
Topology:
trusts but verify, if anything could go wrong - it will go wrong, provides clear context and instructions, subagents can cheat, consults with architect, and uses subagents as its team. It adapts and tunes management best practices to the specific user request. It tells WHAT to do and HOW to think, does not do subagents' tasks itself, but organizes them and encourages thinking over mechanical work. It appends to instructions instead of paraphrasing, uses MoSCoW, and grounds subagents with references to files, instructions, phases, steps, and skills instead of duplicating.Dispatch:
""" You are role/specialization]. Lightweight|Full] subagent.
Target root folder: path] git worktree?] DO: what is in scope, explicit expected outputs and clear expectations] DO NOT: what is explicitly out of scope, what is read-only, what not to touch — forbid out-of-scope work; do not improvise beyond defined scope]
MUST USE SKILL required skill]. RECOMMEND USE SKILL recommended skill].
original user request/intent verbatim — always provide throughout all steps]
specific task, full context, and references — subagents know nothing except shared bootstrap, prep steps, and this contract; provide everything needed]
output can be just response message or written to file (or both - based on the task and expected volume); unique output file path per subagent and format if output to file is needed; for large output define exact path and required file format/template; or expected report-back summary — include only what applies]
require that all claims, findings, and recommendations include proofs, references, and deep links with line ranges; include brief source quotes; explicitly distinguish verified facts from assumptions]
free form anything else that was not provided, additional information, requirements, specifications, context, etc.] """
Routing:
Quality:
Review = static inspection (recommendations). Validate = running on real/sample tasks (catches real issues, expensive).</process>
</orchestrator_contract>
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 31,284 | 34,061 | +9% | 1 | 1 | 0% | 4,981 | 7,028 | +41% | 0 | 0 | — |
case-02 | fail→pass | 32,770 | 31,026 | -5% | 1 | 1 | 0% | 5,522 | 6,293 | +14% | 0 | 0 | — |
case-03 | fail→pass | 34,147 | 35,985 | +5% | 1 | 1 | 0% | 5,315 | 7,207 | +36% | 0 | 0 | — |
case-04 | fail→pass | 13,716 | 11,924 | -13% | 1 | 1 | 0% | 2,339 | 3,010 | +29% | 0 | 0 | — |
case-05 | fail→fail | 12,144 | 16,110 | +33% | 1 | 1 | 0% | 1,842 | 3,329 | +81% | 0 | 0 | — |
case-06 | fail→pass | 12,953 | 11,758 | -9% | 1 | 1 | 0% | 2,067 | 3,139 | +52% | 0 | 0 | — |
case-07 | pass→pass | 19,585 | 5,549 | -72% | 1 | 1 | 0% | 2,292 | 1,810 | -21% | 0 | 0 | — |
case-08 | pass→pass | 13,008 | 10,457 | -20% | 1 | 1 | 0% | 1,994 | 2,441 | +22% | 0 | 0 | — |
case-09 | pass→pass | 11,797 | 9,906 | -16% | 1 | 1 | 0% | 1,650 | 2,426 | +47% | 0 | 0 | — |
case-10 | fail→pass | 14,228 | 10,584 | -26% | 1 | 1 | 0% | 2,190 | 2,566 | +17% | 0 | 0 | — |
case-11 | pass→pass | 18,640 | 15,242 | -18% | 1 | 1 | 0% | 2,773 | 3,285 | +18% | 0 | 0 | — |
case-12 | fail→pass | 17,629 | 16,501 | -6% | 1 | 1 | 0% | 2,794 | 3,315 | +19% | 0 | 0 | — |
case-13 | fail→pass | 10,399 | 7,246 | -30% | 1 | 1 | 0% | 1,692 | 2,015 | +19% | 0 | 0 | — |
case-14 | pass→pass | 10,714 | 10,933 | +2% | 1 | 1 | 0% | 1,735 | 2,813 | +62% | 0 | 0 | — |
case-15 | fail→pass | 13,599 | 4,028 | -70% | 1 | 1 | 0% | 2,050 | 1,513 | -26% | 0 | 0 | — |
case-16 | fail→pass | 16,240 | 13,405 | -17% | 1 | 1 | 0% | 2,389 | 3,075 | +29% | 0 | 0 | — |
case-17 | fail→pass | 11,988 | 8,960 | -25% | 1 | 1 | 0% | 1,890 | 2,205 | +17% | 0 | 0 | — |
case-18 | pass→pass | 9,534 | 6,380 | -33% | 1 | 1 | 0% | 1,390 | 1,980 | +42% | 0 | 0 | — |
case-19 | pass→pass | 14,639 | 11,861 | -19% | 1 | 1 | 0% | 2,178 | 2,715 | +25% | 0 | 0 | — |
case-20 | pass→pass | 6,944 | 23,546 | +239% | 1 | 1 | 0% | 1,189 | 5,821 | +390% | 0 | 0 | — |
case-21 | pass→pass | 13,617 | 13,278 | -2% | 1 | 1 | 0% | 2,059 | 2,922 | +42% | 0 | 0 | — |
case-22 | pass→pass | 11,118 | 12,940 | +16% | 1 | 1 | 0% | 1,964 | 3,125 | +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. The headline lift of +50 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.