betterneighborsai/pd-intelligence

Use when asked for a daily briefing, weekly summary, status update, "what happened", "how are we doing", or any period-in-review of PD Intelligence social media data. Produces an evidence-based intelligence briefing from dashboard stats, daily summaries, leaderboards, and comment activity.

Use when a PD Intelligence pull is too big to hold in context — thousands of posts or comments, "pull everything", "all posts for this account/tag/date range", full-dataset exports, bulk extraction, or any request where paging blindly would blow up or lose track. Partitions the pull, caches it to local NDJSON with a manifest, resumes after interruption, and verifies completeness. Also use to resume or verify a cache a previous run started.

Use when asked what people are saying about a post, topic, or person in PD Intelligence data — reaction analysis, sentiment, narratives, controversy, "how did X land", comment analysis, or finding who is driving a conversation. Chains post search, comment summaries, threads, and commenter profiles.

Use when publishing, updating, sharing, or finding documents on PD Docs, PD Intelligence's document surface — "push this report to PD Intelligence", "publish this briefing or deck", "share this document with the team", "share it to a dataset", "send it to someone outside the org", "make it public", "get a shareable link", "who can see this document", "revoke access", "update the published report", "find the document I published", "read that document back". Covers publish_document, share_document, get_document_sharing, list_documents and read_document, the large-document tools (begin_document_upload, append_document_chunk, finish_document_upload, get_document_download_url), plus the sandbox every published HTML document renders in. To build the document itself, use pd-intel-report.

Use whenever working with PD Intelligence data (the PD Intelligence MCP connector) — any question about tracked social media posts, accounts, creators, comments, datasets, tags, or metrics. Explains how the data is organized, which tool to use for what, what a zero or absent metric actually means on each platform, and the pitfalls that cause wrong numbers and wrong caveats in reports.

Use for questions about news coverage and media framing in PD Intelligence — "what's the news today", the daily news briefing, "how did outlets cover X", "what did the left/right miss", media ecosystem comparison, coverage gaps, blindspots, spin, one outlet's front page, "what's trending in the news", how often a term shows up in headlines over time, the weekly news roundup, headline word clouds, which dates have a briefing, or comparing the news cycle with social media activity. Covers get_news_briefing, list_news_dates, get_news_outlet, get_news_roundup, get_news_trend and get_news_word_cloud. News is global, not dataset-scoped, and needs the news entitlement.

Use when the deliverable is an artifact someone else will read, built on PD Intelligence data — an executive report, board update, campaign recap, one-pager, memo, written report document, HTML slides or a presentation deck, an interactive or explorable data story, a scrollytelling piece, or a D3 / three.js visualization. Triggers on "make a report", "build a deck", "turn this into slides", "write this up as a document", "make it interactive", "add a visualization", "an explorable explainer", "something I can send the board". Chooses the format with the user, shapes verified evidence into a story, renders it, and hands it to PD Docs for publishing. For an answer in chat rather than an artifact, use pd-intel-briefing or pd-intel-deep-dive instead.

Use when labelling, organizing, or curating PD Intelligence data — creating a tag, renaming or recoloring one, tagging or untagging posts, accounts, creators, comments or commenters in bulk, building a cohort or campaign label, cleaning up a dataset's tags, or asking "how do I tag these" and "what tags do we have". Covers the write tools create_tag, update_tag, delete_tag and tag_entities, their permission gates, and the inheritance rule that makes bulk tagging cheap.