원클릭으로 Manus에서 모든 스킬 실행

research-orchestration

스타31

포크11

업데이트2026년 5월 18일 13:55

Agents should invoke this skill for broad multi-claim research projects needing planning, parallel investigation, source merging, gap closure, citation audit, and final synthesis when narrower research skills are insufficient.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

Firstp1ck

Firstp1ck/npm-packages

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

Research orchestration

Single workflow entry point for end-to-end investigations. Detailed rules live in workspace-researcher/AGENTS.md (task class, parallel investigation, Deep Research Protocol, citation audit); this skill is the ordered checklist for executing them.

When to use

Broad question with several independent sub-questions or claim clusters.
Mixed sources (web + docs + papers) and one final report or contract output.
User or delegating agent asked for deep research without pointing at a single comparison or single paper.

Prefer tech-deep-dive, competitor-analysis, or paper-summarizer when the task clearly matches one of those shapes.

Pipeline

Classify depth — Quick / standard / deep (AGENTS.md task-class table). Honor explicit quick / standard / deep or /deep-research quick|standard|max when the user names them. Set expected tool budget.
Plan sub-questions — List facets and claims to verify; note which are high-stakes (citation audit mandatory). Prefer internal → official → web (AGENTS.md data source order) when gathering evidence.
Parallel pass — Batch web_search / web_fetch per independent facet; optional subagents with tight scopes and structured handoff.
Merge — One synthesis pass; resolve contradictions; prefer primary sources.
Gap closure — List unresolved material gaps; targeted follow-up searches or explicit “blocked” notes.
Citation audit — For mandatory cases: every key finding → source row or unsupported / inferential (AGENTS.md).
Deliver — Default report shape, contract JSON, or consumer-specific packet from AGENTS.md; include Limitations and Research trace when depth is standard/deep or requested (APPEND_SYSTEM.md).
Log — Append a row to workspace-researcher/MEMORY.md → Research History (topic, task class, tool bucket, major gaps, notable sources).

Scout scripts → pipeline step

Run from the installed skill directory, or adjust paths to absolute locations. Policy defaults to ./scripts/policy.json next to the bundled scripts.

Step	Script	When
After depth + topic fixed	`./scripts/scout_query_plan.py`	Emit ordered queries (`Q001`…) and `plan_hash` for the research trace
After URL collection	`./scripts/scout_normalize_sources.py`	Dedupe and sort sources before the final table
Long / multi-session runs	`./scripts/scout_evidence_bundle.py`	Freeze fetch records (JSONL → `evidence_bundle.json` + `content_sha256`)
Before high-stakes delivery	`./scripts/scout_citation_audit.py`	Gate: every `key_findings[]` row has `source_ids` or `unsupported` / `inferential`

Details and exit codes: ./scripts/README.md.

Handoff schema (optional, for subagents)

## Scope
[One sentence]

## Findings
- [Bullet + URL]

## Gaps
- [What could not be verified]

## Sources touched
- [Title](URL) — quality flag

Scout skill — Orchestrated research pipeline · 2026-04-10

이 저장소의 다른 Skills

같은 저장소

html-report

Firstp1ck/npm-packages

Creates polished self-contained HTML explanation reports for complex, multi-step topics using overview tables, evidence cards, adaptive graphs or diagrams, meaningful SVG/media, and accessible tabs for long pages. Use when the user requests a browser-readable diagnostic report, technical guide, decision analysis, implementation plan, investigation, or research synthesis as HTML.

2026-07-1831

patch-md

Firstp1ck/npm-packages

Agents should invoke this skill to create, migrate, inspect, apply, verify, or roll back versioned PATCH.md v2 lifecycle packages with machine-readable manifests, deterministic plans, drift detection, and transactional safety.

2026-07-1831

docx-editor

Firstp1ck/npm-packages

Safely inspect, search, render, edit, diff, validate, and commit Microsoft Word DOCX files with the dedicated docx_* tools. Use for DOCX/DOTX/DOCM/DOTM document work; enforces dry-run-first edits, active-content refusal, preservation checks, and save-as by default.

2026-07-1431

deep-research

Firstp1ck/npm-packages

Agents should invoke this skill for high-stakes or complex research needing multi-source evidence, scientific/technical fact-checking, decision traces, or rigorous verification. Runs deterministic two-phase research with schema/policy validation.

2026-07-1431

workbook-editor

Firstp1ck/npm-packages

Use when inspecting, reading, rendering, editing, formatting, diffing, or validating local .xlsx and .xlsm Excel workbooks. Provides fail-closed workbook tools with dry-run planning, SHA-256 conflict protection, PNG previews, and byte-identity checks for VBA and other protected OOXML parts. Do not use for CSV-only work or for executing/editing VBA.

2026-07-1431

writing-great-skills

Firstp1ck/npm-packages

Reference for writing and editing skills well — the vocabulary and principles that make a skill predictable.

2026-06-3031

name	research-orchestration
description	Agents should invoke this skill for broad multi-claim research projects needing planning, parallel investigation, source merging, gap closure, citation audit, and final synthesis when narrower research skills are insufficient.

Research orchestration

When to use

Broad question with several independent sub-questions or claim clusters.
Mixed sources (web + docs + papers) and one final report or contract output.
User or delegating agent asked for deep research without pointing at a single comparison or single paper.

Prefer tech-deep-dive, competitor-analysis, or paper-summarizer when the task clearly matches one of those shapes.

Pipeline

Classify depth — Quick / standard / deep (AGENTS.md task-class table). Honor explicit quick / standard / deep or /deep-research quick|standard|max when the user names them. Set expected tool budget.
Plan sub-questions — List facets and claims to verify; note which are high-stakes (citation audit mandatory). Prefer internal → official → web (AGENTS.md data source order) when gathering evidence.
Parallel pass — Batch web_search / web_fetch per independent facet; optional subagents with tight scopes and structured handoff.
Merge — One synthesis pass; resolve contradictions; prefer primary sources.
Gap closure — List unresolved material gaps; targeted follow-up searches or explicit “blocked” notes.
Citation audit — For mandatory cases: every key finding → source row or unsupported / inferential (AGENTS.md).
Deliver — Default report shape, contract JSON, or consumer-specific packet from AGENTS.md; include Limitations and Research trace when depth is standard/deep or requested (APPEND_SYSTEM.md).
Log — Append a row to workspace-researcher/MEMORY.md → Research History (topic, task class, tool bucket, major gaps, notable sources).

Scout scripts → pipeline step

Run from the installed skill directory, or adjust paths to absolute locations. Policy defaults to ./scripts/policy.json next to the bundled scripts.

Step	Script	When
After depth + topic fixed	`./scripts/scout_query_plan.py`	Emit ordered queries (`Q001`…) and `plan_hash` for the research trace
After URL collection	`./scripts/scout_normalize_sources.py`	Dedupe and sort sources before the final table
Long / multi-session runs	`./scripts/scout_evidence_bundle.py`	Freeze fetch records (JSONL → `evidence_bundle.json` + `content_sha256`)
Before high-stakes delivery	`./scripts/scout_citation_audit.py`	Gate: every `key_findings[]` row has `source_ids` or `unsupported` / `inferential`

Details and exit codes: ./scripts/README.md.

Handoff schema (optional, for subagents)

## Scope
[One sentence]

## Findings
- [Bullet + URL]

## Gaps
- [What could not be verified]

## Sources touched
- [Title](URL) — quality flag

Scout skill — Orchestrated research pipeline · 2026-04-10