Source Reader

Source Reader

Detects what kind of source the user referenced and extracts its text content using the appropriate tool. This skill is a building block — other skills (like notetaking or notes-to-slides) use it as step 0.

Source detection

Match the user's input against these patterns to determine the source type:

If the input doesn't match any pattern, it's likely pasted text — pass it through to the downstream skill directly.

Extraction instructions

Google Docs

Extract the document ID from the URL (the segment after /d/ and before /).

gws docs documents get --params '{"documentId":"<id>"}' --format json 2>/dev/null

The response contains the document's structured content in body.content. Walk through the paragraph elements and concatenate each textRun.content value to build the full text. Preserve paragraph breaks.

Google Slides

Extract the presentation ID from the URL.

gws slides presentations get --params '{"presentationId":"<id>"}' --format json 2>/dev/null

Walk through each slide in slides[]. For each slide, extract text from pageElements[].shape.text.textElements[].textRun.content. Prefix each slide's content with --- Slide N --- to preserve structure.

Slack

Use the Slack MCP tools or call the slack-mcp server directly:

• Channel history: get_channel_history with the channel name and a message limit appropriate to the time range
• Search: search_messages with from:<person> or in:<channel>
• DMs: search_messages with from:<person> to:<user> and the reverse

Capture both sides of a conversation. Include timestamps and sender names.

Jira via Dataverse

Use the Dataverse MCP 4-step pipeline:

1. identify_dataproducts — confirms jira data product
2. shortlist_tables with data_product: "jira"
3. get_sql with the shortlisted tables and a query for the issues
4. execute_sql with the generated SQL

For a single issue key, query for that issue's summary, description, status, assignee, and comments. For a project key, query for recent or in-progress issues with key, summary, status, and assignee.

Google Drive (unknown type)

When the user provides a Drive link without a clear type:

gws drive files get --params '{"fileId":"<id>","fields":"id,name,mimeType"}' --format json 2>/dev/null

Check the mimeType field and delegate:

• application/vnd.google-apps.document -> Google Docs extraction
• application/vnd.google-apps.presentation -> Google Slides extraction
• application/vnd.google-apps.spreadsheet -> Google Sheets (use gws sheets spreadsheets values get to pull cell data)

Output format

Return the extracted text to the calling context. Include a source metadata block at the top:

Source: <type> — <name or URL>
Extracted: <timestamp>

<content>

The downstream skill decides how to structure and format the content.

Accuracy rules

• Extract only what the source contains. Do not summarize, interpret, or add context during extraction.
• Preserve the original structure (paragraphs, slide boundaries, message order) so the downstream skill can make informed formatting decisions.
• If extraction fails (auth error, file not found, API error), report the error clearly and suggest what the user can try (check the URL, verify access, re-authenticate).