Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Workflow for deep project research with grounded references, parallel exploration, etc.
.claude/skills/griddynamics-research-flow/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -17% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 5% | 0% |
<research_flow>
<description_and_purpose> Orchestrates deep research via meta-prompting: craft an optimized research prompt, then execute it in a dedicated subagent. </description_and_purpose>
<workflow_phases>
<prerequisites phase="0", applies="ALL">
load-project-context, orchestration, hitlagents/TEMP/<FEATURE>/research-flow-state.md file./goal is set repeat phases 3-4 until goal is met.</prerequisites>
<context_load phase="1" subagent="researcher" role="Context gatherer for research scope" subagent_required_model="claude-sonnet-5, gpt-5.4-medium, gemini-3.1-pro, grok-4.5, gpt-5.6-terra">
research-flow-state.md.</context_load>
<prompt_craft phase="2" subagent="researcher" role="Research prompt architect" subagent_required_model="claude-opus-4-8, gpt-5.4-high, gpt-5.5-high, gemini-3.1-pro-high, gpt-5.6-sol">
research-prompt.md in FEATURE PLAN folder. Output ONLY the optimized prompt.research-prompt.md.reasoningresearch-flow-state.md.</prompt_craft>
<execute_research phase="3" subagent="researcher" role="Deep research executor" subagent_required_model="claude-sonnet-5, gpt-5.4-medium, gemini-3.1-pro, grok-4.5, gpt-5.6-terra">
research-prompt.md. Output: docs/<feature>-research.md.researchresearch-flow-state.md.</execute_research>
<finalize phase="4" subagent="researcher" role="Research finalizer" subagent_required_model="claude-sonnet-5, gpt-5.4-medium, gemini-3.1-pro, grok-4.5, gpt-5.6-terra">
docs/<feature>-research.md.research-flow-state.md and mark complete.</finalize>
</workflow_phases>
</research_flow>
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→fail | 4,465 | 7,152 | +60% | 1 | 1 | 0% | 249 | 1,272 | +411% | 0 | 0 | — |
case-01 | fail→fail | 2,999 | 9,832 | +228% | 1 | 1 | 0% | 224 | 1,392 | +521% | 0 | 0 | — |
case-02 | fail→fail | 10,264 | 7,569 | -26% | 1 | 1 | 0% | 1,520 | 1,247 | -18% | 0 | 0 | — |
case-03 | fail→fail | 28,978 | 7,709 | -73% | 1 | 1 | 0% | 4,499 | 1,339 | -70% | 0 | 0 | — |
case-05 | fail→fail | 14,219 | 11,956 | -16% | 1 | 1 | 0% | 2,184 | 2,629 | +20% | 0 | 0 | — |
case-06 | fail→fail | 31,865 | 4,632 | -85% | 1 | 1 | 0% | 4,957 | 1,126 | -77% | 0 | 0 | — |
case-07 | fail→fail | 9,917 | 3,874 | -61% | 1 | 1 | 0% | 1,552 | 1,414 | -9% | 0 | 0 | — |
case-08 | fail→fail | 4,118 | 2,581 | -37% | 1 | 1 | 0% | 700 | 1,154 | +65% | 0 | 0 | — |
case-13 | fail→pass | 8,620 | 1,657 | -81% | 1 | 1 | 0% | 1,312 | 1,013 | -23% | 0 | 0 | — |
case-09 | fail→pass | 8,859 | 8,001 | -10% | 1 | 1 | 0% | 1,497 | 2,350 | +57% | 0 | 0 | — |
case-10 | fail→pass | 7,869 | 2,406 | -69% | 1 | 1 | 0% | 1,227 | 1,137 | -7% | 0 | 0 | — |
case-11 | pass→pass | 9,750 | 3,228 | -67% | 1 | 1 | 0% | 1,465 | 1,289 | -12% | 0 | 0 | — |
case-12 | fail→pass | 8,159 | 1,739 | -79% | 1 | 1 | 0% | 1,201 | 991 | -17% | 0 | 0 | — |
case-14 | fail→fail | 4,459 | 2,447 | -45% | 1 | 1 | 0% | 629 | 1,151 | +83% | 0 | 0 | — |
case-15 | fail→pass | 10,522 | 4,788 | -54% | 1 | 1 | 0% | 1,487 | 1,565 | +5% | 0 | 0 | — |
case-16 | fail→pass | 10,024 | 2,800 | -72% | 1 | 1 | 0% | 1,439 | 1,199 | -17% | 0 | 0 | — |
case-17 | pass→pass | 7,616 | 1,862 | -76% | 1 | 1 | 0% | 1,119 | 1,034 | -8% | 0 | 0 | — |
case-18 | fail→pass | 9,134 | 3,315 | -64% | 1 | 1 | 0% | 1,512 | 1,232 | -19% | 0 | 0 | — |
case-19 | pass→fail | 13,991 | 5,814 | -58% | 1 | 1 | 0% | 2,789 | 1,136 | -59% | 0 | 0 | — |
case-20 | pass→fail | 6,621 | 16,275 | +146% | 1 | 1 | 0% | 1,232 | 3,530 | +187% | 0 | 0 | — |
case-21 | pass→fail | 19,637 | 10,310 | -47% | 1 | 1 | 0% | 2,817 | 2,241 | -20% | 0 | 0 | — |
case-22 | fail→fail | 9,992 | 8,546 | -14% | 1 | 1 | 0% | 926 | 1,583 | +71% | 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 17 counted toward the lift figure. The other 5 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 +18 percentage points is the difference between those two pass rates over the 17 comparable cases. 4 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.