cascade-summarize-document-content

name	cascade-summarize-document-content
description	Sub-skill: Summarize and categorize fetched content from .temp/cascade/context/. Extracts key information, identifies newly discovered links, and writes analysis summaries. Used as a building block by parent skills like generate-questions and write-jira-story.

Analyze Content

Summarize and categorize raw content previously fetched by the load-linked-resource-content sub-skill. Extract key information and discover any additional links embedded in the content.

When to Use

This is a sub-skill — called by parent skills after load-linked-resource-content has fetched raw content. Use when the parent skill needs summarized, structured content before synthesis or generation steps.

Prerequisites

• Raw content files exist in .temp/cascade/context/ (written by load-linked-resource-content)
• .temp/cascade/context/to-load.md exists with loading manifest

Procedure

1. Read content files

Read all .md files in .temp/cascade/context/ (excluding to-load.md and any files ending in -summary.md).

2. Analyze each content file

For each raw content file, produce a summary that extracts:

For Jira issues (jira-*.md):

• Issue type, status, priority
• Summary and key requirements from description
• Acceptance criteria (if present)
• Linked issues and their relationships (blocks, is blocked by, relates to)
• Figma/Confluence/Google Docs URLs found in description or comments
• Key decisions or answers from comments (look for 💬 markers)
• Open questions (look for ❓ markers)

For Confluence pages (confluence-*.md):

• Page purpose and key sections
• Requirements, specifications, or design decisions documented
• Referenced Figma designs or mockups
• Links to other Confluence pages, Jira issues, or external docs
• Tables of data (preserve structure in summary)

For Google Docs (gdoc-*.md):

• Document purpose
• Key requirements or specifications
• Design decisions or constraints
• Referenced links

3. Write summary files

Save each summary alongside the raw content file with a -summary.md suffix:

.temp/cascade/context/
├── jira-PROJ-123.md              ← raw content (from load-linked-resource-content)
├── jira-PROJ-123-summary.md      ← summary (from summarize-document-content)
├── confluence-design-spec.md
├── confluence-design-spec-summary.md
├── gdoc-requirements.md
├── gdoc-requirements-summary.md
└── to-load.md

4. Extract newly discovered links

While analyzing, collect any URLs found in the content that are NOT already in to-load.md. Append them as [ ] entries in the ## Unloaded section.

5. Return to parent skill

Report back:

• How many files were analyzed
• How many new links were discovered (and added to to-load.md)
• Brief summary of what was found (e.g., "Analyzed 3 files: 1 Jira epic with 5 linked stories, 1 Confluence design spec, 1 Google Doc requirements doc. Found 2 new Confluence links.")

Summary Format

Keep summaries concise but complete. Target 200-500 words per file. Use this structure:

# Summary: {source title}

**Source**: {URL}
**Type**: {Jira Issue | Confluence Page | Google Doc}

## Key Information
- {bullet points of the most important facts}

## Requirements / Specifications
- {requirements extracted from the content}

## Open Questions
- ❓ {any unanswered questions found}
- 💬 {any answered questions found with their answers}

## Discovered Links
- {new URLs found in this content}

Important Notes

• Do NOT re-analyze files that already have a -summary.md — skip them unless the parent skill explicitly requests re-analysis
• Preserve scope markers — if the source content uses ☐/✅/❌/⏬/❓/💬 markers, preserve them in the summary
• Be factual — summaries should extract information, not generate new content or opinions
• Link discovery is critical — the parent skill's load→analyze loop depends on finding new links to continue iterating