TED Trends Watcher

Purpose

Monitor recent official TED content as a bounded research task, identify defensible directional signals, and translate selected ideas into practical work outputs. Treat TED as a source of ideas and public discourse, not as proof of a market fact, product claim, scientific consensus, or business outcome.

When to use

Use this skill for requests to:

• Track what TED has recently published or discussed.
• Summarize TED content on a specified theme or audience need.
• Compare viewpoints across TED talks or related official content.
• Convert TED-derived ideas into customer conversations, proposals, slide storylines, speeches, experiments, or knowledge cards.
• Set up a recurring monitoring brief, provided the host supports scheduling.

Do not use this skill as a general news monitor, a substitute for academic literature review, or a source of unsupported industry forecasts. State the boundary and recommend a more suitable research scope when the request depends on those things.

Operating principles

1. Define the research question before searching. Distinguish "what TED published" from "what the topic means" and from "what to do next".
2. Bound every result by monitoring window, information cutoff, timezone, themes, audience, and output mode.
3. Prefer first-party TED sources and preserve the exact source URL, publication date, content type, and access level.
4. Treat search snippets, social posts, aggregators, and reuploads as discovery leads, never as evidence by themselves.
5. Separate five layers in the final output: verified source fact, speaker or author claim, TED editorial framing, analyst interpretation, and recommendation.
6. Report uncertainty explicitly. Do not fill missing dates, quotes, statistics, biographies, or talk details from memory or plausible inference.
7. Call a trend only when the threshold in references/source-and-scoring.md is met. Otherwise use "notable idea", "emerging signal", or "single-speaker viewpoint".
8. Adapt ideas to the user's context; do not borrow TED prestige as evidence for a commercial claim or imitate a living speaker's distinctive style.

Input contract and defaults

Extract or infer the following fields before research:

• Question: the decision, topic, or change to investigate.
• Window: publication window, not access window. Interpret relative dates against the current date and state the resolved dates.
• Cutoff: the latest date and time included; default to the current date at Asia/Shanghai unless the user specifies another timezone.
• Themes: requested topics; default to AI and agents, emerging technology and science, business and organizational innovation, learning and personal growth, and communication and public speaking.
• Audience and context: decision makers, sales, product, teaching, personal use, or another stated scenario.
• Output mode: trend brief, deep dive, sales enablement, proposal/PPT input, speech input, knowledge cards, or a concise Chinese briefing.
• Evidence depth: quick scan, standard review, or deep review. Default to standard review when unspecified.

When a parameter is missing, proceed with the defaults and show the assumption. Ask a clarifying question only when the missing choice would materially change the result, such as an ambiguous date range, audience, or deliverable.

Research protocol

Follow the sequence below and keep an internal evidence ledger. Use references/research-protocol.md for the field schema and retrieval checklist.

1. Frame the brief

Write a one-sentence research question and define inclusion and exclusion criteria. For example, distinguish official TED publication from an older talk that merely remains popular, and distinguish "AI agents" from generic technology coverage.

2. Search in layers

Search in this order:

1. TED.com talk, transcript, article, playlist, event, podcast, and newsletter pages.
2. TED's verified YouTube or podcast channels when the TED page is unavailable or incomplete.
3. Official speaker, institution, paper, company, or regulator pages for corroboration or context.
4. Reputable reporting only for context, controversy, or independent corroboration.
5. Community posts and social discussion only for discovery.

Use focused queries that include the theme, the date window, and site restrictions where supported. Search several formulations rather than repeatedly opening one result. Deduplicate syndicated copies, clips, event listings, and the same talk appearing in multiple formats.

3. Capture before interpreting

For every retained item, capture the candidate record before writing a summary:

• stable title and speaker or author;
• publication or event date, including whether it is an upload, update, or recording date;
• official URL and any corroborating URL;
• content type and theme tags;
• what was actually accessed: full transcript, full video, official page, abstract, excerpt, or snippet;
• one or more source-grounded claims;
• evidence limitations and unresolved fields;
• preliminary relevance and score.

Do not promote an item to the synthesis set solely because its title or snippet sounds relevant.

4. Verify source and access level

Open the source when possible. Confirm that the page belongs to the claimed publisher, that the date is visible or reliably stated, and that the content supports the extracted claim. Record access as one of the following:

• A - Full content: full transcript or full video successfully accessed.
• B - Official synopsis: official TED page, abstract, description, or show notes accessed, but not the full content.
• C - Partial lead: search snippet, excerpt, metadata, or third-party mention only.

Use A-level evidence for detailed summaries and quotations. Use B-level evidence for high-level topic and explicitly qualified claims. Use C-level items only as leads or watchlist entries; do not use them to support a firm conclusion.

5. Classify claims

Label each material statement as one of:

• Source fact: title, speaker, date, content type, or wording visible in the source.
• Speaker/author claim: an argument or assertion made by the speaker or author; attribute it.
• TED editorial framing: wording or categorization supplied by TED.
• Interpretation: analysis derived from one or more verified items.
• Recommendation: an action proposed for the user's context.

Make the label visible whenever a reader could confuse interpretation or advocacy with fact.

6. Score and classify signals

Use references/source-and-scoring.md for a transparent triage score, evidence grades, and trend thresholds. Apply hard evidence gates before numeric scores. A high score cannot rescue an untraceable or weakly accessed item.

7. Synthesize across items

Group items by a clearly stated construct, not by vague topic similarity. For every proposed signal, show:

• the directional change in one sentence;
• the independent items that support it;
• what those items actually establish;
• what remains interpretation or conjecture;
• counterevidence, disagreement, or sampling limitations;
• relevance to the user's scenario.

Do not generalize from a single talk to "TED's position" or from a cluster of TED items to an industry-wide trend without outside evidence.

8. Adapt to the requested output

Use references/application-playbook.md to translate only the strongest, relevant ideas. Preserve evidence boundaries in sales, proposal, deck, speech, and knowledge outputs. Include a concrete next action and, where appropriate, an observable success measure.

9. Run the quality gate

Use references/output-templates.md and verify before delivery:

• monitoring window, cutoff, timezone, and assumptions are visible;
• every important factual claim has a traceable source;
• access level is stated when it limits the conclusion;
• facts, attributed claims, interpretation, and recommendations are separated;
• trend language meets the formal threshold;
• conflicting or missing evidence is not hidden;
• quotations are exact, short, and verified against accessible primary content;
• adapted material fits the audience, decision, and requested format;
• no source is described as fully read or watched unless it was actually accessed.

Output modes

Choose the lightest format that satisfies the request:

• Trend brief: executive answer, signals, evidence ledger, watchlist, implications, actions, and sources.
• Deep dive: core question, short answer, evidence for and against, uncertainty, implications, experiments, and sources.
• Sales enablement: customer scenario, bounded opener, discovery questions, value narrative, objections, next action, and source notes.
• Proposal/PPT input: deck thesis, audience tension, storyline, slide-ready claims, visual suggestions, implementation path, and verification notes.
• Speech input: original hook, tension, central idea, attributed example, practical takeaway, and original closing line.
• Knowledge cards: one idea per card with evidence, interpretation, application, reflection question, source, and confidence reason.

Use the templates in references/output-templates.md rather than inventing a looser structure for recurring briefs.

Failure handling and transparency

When a page is inaccessible, transcript is unavailable, date is ambiguous, or sources conflict:

1. Keep the item in the ledger with the limitation recorded.
2. Lower the evidence grade and narrow the claim.
3. Replace detailed summary with a high-level qualified description, or move the item to the watchlist.
4. State what could not be verified and what additional source would resolve it.
5. Never imply that an unavailable source was read, watched, or checked in full.

If no item meets the evidence threshold, deliver a transparent "insufficient evidence" result with the search scope, strongest leads, and a suggested follow-up—not a fabricated trend.

Safety and editorial boundaries

• Do not fabricate or silently repair source metadata.
• Do not present a speaker's opinion as TED's official position.
• Do not use TED as authority for medical, legal, financial, safety-critical, or product-performance claims.
• Do not reproduce long copyrighted passages; summarize and cite instead.
• Do not imitate the distinctive style of a living speaker.
• Do not infer a speaker's identity, affiliation, motive, or biography beyond accessible evidence.