AgentSkillsCN

content-curation

从第1A阶段的URL清单中，按来源提取并整理内容，生成5个临时文件。在运行新闻简报管道的第1B阶段时，按来源依次处理URL，采用通用提取格式提取相关内容，每个来源输出一个临时文件。关键词：内容提取、第1B阶段、网页提取、临时文件。

SKILL.md
--- frontmatter
name: content-curation
description: "Transforms raw discoveries into polished, newsletter-ready content sections. Use when running Phase 3 of the newsletter pipeline. Selects 15-20 most enterprise-relevant items from 30-50 discoveries, applies GA/PREVIEW labels, formats into Lead section, Copilot Latest Releases, and Copilot at Scale. Keywords: content curation, phase 3, selection, formatting, enterprise relevance."
metadata:
  category: domain
  phase: "3"

Content Curation

Transform Phase 1C discoveries into polished, newsletter-ready content sections.

Quick Start

  1. •Read Phase 1C discoveries from workspace/newsletter_phase1a_discoveries_*.md
  2. •Select 15-20 most enterprise-relevant items using selection criteria
  3. •Organize into sections: Lead (optional), Copilot (Latest + IDE Parity), Copilot at Scale
  4. •Apply formatting: bold terms, GA/PREVIEW labels, embedded links, strip metadata
  5. •Write output to workspace/newsletter_phase3_curated_sections_YYYY-MM-DD.md

Inputs

  • •Phase 1C Discoveries: workspace/newsletter_phase1a_discoveries_*.md (required)

Output

  • •File: workspace/newsletter_phase3_curated_sections_YYYY-MM-DD.md
  • •Target: 15-20 curated items (quality over quantity)
  • •Content: Core product sections only (no events, no introduction/closing)

Core Workflow

Step 1: Analyze Discoveries

Read Phase 1C input. Inventory candidates by:

  • •Enterprise relevance and impact
  • •Recency within DATE_RANGE
  • •Thematic clusters that could drive a lead section
  • •Overlapping/duplicate items needing consolidation

Step 2: Select and Bundle Items

Apply selection criteria in priority order. See selection-criteria.md.

Key finding (see reference/source-intelligence/meta-analysis.md): Cross-cycle analysis shows discoveries have ~100% survival rate. The curation job is NOT selection (nearly everything survives). It is:

  1. •Bundling: Models → single bullet. IDE parity → nested bullet. Governance cluster → single enriched bullet.
  2. •Flagship detection: 1-2 items per cycle get expanded dedicated sections.
  3. •Gap-filling: Add Azure, devblogs, enablement content the pipeline missed.
  4. •Competitive framing: Identify competitive positioning signals (CLI vs Claude Code, 3P agents, OpenCode, BYOK = platform openness).

Priority weights:

  1. •Competitive positioning (3.5x) — features that counter rival tools
  2. •Governance/Admin (3.0x) — policies, controls, billing, metrics
  3. •Security/Compliance (2.5x) — scanning, GHAS, supply chain
  4. •GA Status (2.0x) — GA always leads
  5. •Novelty (2.0x) — underappreciated items, new categories, legal changes
  6. •Platform openness (2.0x) — BYOK, 3P integrations, multi-surface
  7. •IDE Parity (1.5x) — cross-IDE rollout
  8. •Copilot Features (1.0x baseline)

Step 3: Organize Into Sections

Lead Section (optional): Only when discoveries show a clear theme (major launch, vision update). Derive title from content cluster, not generic label.

Copilot - Latest Releases: New features, model updates, agent capabilities. VS Code features go here, organized by feature theme (not by version number). Never reference specific VS Code version numbers in bullet text -- version numbers appear only in link URLs. With weekly VS Code releases, features may span multiple versions; present the end-of-period state.

Copilot at Scale: Enterprise governance, billing, training, metrics. Always include standard changelog links.

Additional Sections: Security Updates, Platform Updates, etc. only when warranted by content.

Step 3.5: Build Cross-IDE Feature Alignment Matrix (L66)

Before writing the IDE parity section, build a feature alignment matrix. This is the single most important step for IDE parity quality.

Procedure:

  1. •List ALL features that appeared in ANY non-VS-Code IDE during the period (Visual Studio, JetBrains, Eclipse, Xcode)
  2. •For each feature, check its end-of-period status in EACH IDE:
    • •GA = Generally Available
    • •PREVIEW = Public Preview / Experimental
    • •— = Not available in this IDE
  3. •Cross-reference with VS Code status (from Latest Releases) to identify parity vs. unique features
  4. •Record the matrix in a working table:
code
| Feature | VS Code | Visual Studio | JetBrains | Eclipse | Xcode |
|---------|---------|--------------|-----------|---------|-------|
| Agent Skills | GA | — | PREVIEW | — | — |
| MCP Registry | GA | — | PREVIEW | — | PREVIEW |
  1. •Use this matrix to generate the IDE parity section:
    • •Every IDE that had ANY update MUST appear (all 4 non-VS-Code IDEs if they had updates)
    • •Every feature MUST have a GA/PREVIEW label per IDE (never omit labels)
    • •Feature-centric format (list features per IDE, not versions per IDE)
    • •JetBrains: list features with labels, NOT version-by-version changelogs
    • •Include the standard rollout note at the bottom

Gate: If Eclipse or Xcode had releases in the period but do not appear in the parity section, STOP and add them.

Step 4: Format

See content-format-spec.md for complete formatting rules.

Key rules:

  • •Strip all raw metadata (dates, scores, IDE fields) from output
  • •Merge duplicates into single consolidated entry
  • •Bold key terms, no em dashes, no raw URLs
  • •GA/PREVIEW labels when known, omit when ambiguous
  • •Link priority: [Announcement] > [Docs] > [Release Notes] > [Changelog]
  • •Consolidate model rollouts into single "Model availability updates" bullet
  • •Surface governance/legal under Copilot at Scale

Step 5: Quality Check

Before writing output:

  • • Lead section included only when clear theme exists
  • • Copilot section has Latest Releases + IDE Parity grouping
  • • Copilot at Scale includes enterprise items + changelog links
  • • GA before PREVIEW when both exist for same feature
  • • Labels omitted when ambiguous
  • • No Copilot Free/Individual/Pro/Pro+ mentions
  • • Metadata stripped from final bullets
  • • All links use [Text](URL) format (never [[Text]](URL))
  • • Status labels verified per-IDE (never assume GA because another IDE is GA)
  • • PREVIEW features with DPA coverage have a Note with link
  • • Quantitative metrics are directly from source (no derived calculations)
  • • Sections with 3+ items have a bold framing intro: Theme -- Enterprise context
  • • Feature descriptions cross-checked against docs, not just changelog titles

Reference

Key Signals to Watch For

Before curating, check for these high-weight signals in the discoveries:

  1. •Competitive positioning: CLI features (vs Claude Code), 3P agent support, OpenCode, BYOK (platform openness)
  2. •Governance clustering: >=5 admin/policy/compliance items forming a narrative
  3. •Blog posts from news-insights/: Major strategic announcements (CPO/CEO posts) that may not be in the changelog
  4. •VS Code hidden features: Read the actual release notes page, not just the changelog entry summary

Done When

  • • Curated sections file exists at workspace/newsletter_phase3_curated_sections_*.md
  • • 15-20 curated items total
  • • Proper section structure (Lead if warranted, Copilot, Copilot at Scale)
  • • GA/PREVIEW labels present where known
  • • IDE Parity pattern with rollout note included
  • • Standard changelog links in Copilot at Scale
  • • No raw metadata, no em dashes, no raw URLs
  • • Enterprise focus throughout