Skip to main content

research-flow

Workflow for deep project research with grounded references, parallel exploration, etc.

Quellinformationen

Repository
griddynamics/rosetta
Letzte Quellaktivität
19. August 2026 um 22:06
Erkannte Sprache von SKILL.md
Englisch
Sterne
360
Forks
80

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
research-flow
description
Workflow for deep project research with grounded references, parallel exploration, etc.
<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"> 1. All Rosetta prep steps MUST be FULLY completed 2. USE SKILL `load-project-context`, `orchestration`, `hitl` 3. MUST ALWAYS use todo tasks ledger, ASAP. Phases are sequential. Independent tasks can run in parallel. 4. Orchestrator trusts the system and skills; coordinates sequence, artifacts, state, and approvals only. 5. Workflow state MUST be saved to `agents/TEMP/<FEATURE>/research-flow-state.md` file. 6. If `/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="inherit"> 1. Read all lines from CONTEXT.md, ARCHITECTURE.md, and IMPLEMENTATION.md. 2. Input: user research request. Output: loaded project context. 3. Update `research-flow-state.md`. </context_load> <prompt_craft phase="2" subagent="researcher" role="Research prompt architect" subagent_required_model="inherit"> 1. Create an optimized research prompt for the user request. 2. Save as `research-prompt.md` in FEATURE PLAN folder. Output ONLY the optimized prompt. 3. Input: user request + project context. Output: `research-prompt.md`. 4. Required skills: `reasoning` 5. Update `research-flow-state.md`. 6. HITL approval of research prompt before execution. </prompt_craft> <execute_research phase="3" subagent="researcher" role="Deep research executor" subagent_required_model="inherit"> 1. Execute the approved research prompt as a separate subagent. 2. Input: approved `research-prompt.md`. Output: `docs/<feature>-research.md`. 3. Required skills: `research` 4. Update `research-flow-state.md`. </execute_research> <finalize phase="4" subagent="researcher" role="Research finalizer" subagent_required_model="inherit"> 1. Finalize `docs/<feature>-research.md`. 2. Input: completed research document. Output: finalized research document. 3. Update `research-flow-state.md` and mark complete. </finalize> </workflow_phases> </research_flow>
Auf GitHub ansehen