Index of every page on this site: llms.txt
All skills
Run it yourself

Run a weekly, evidence-tagged scan of what is changing in your niche (AI tool updates, user questions, creator and freelancer problems, platform changes) so your content plan is fed by current reality, not recycled ideas. Every candidate is tagged FACT, SIGNAL, or SPECULATION and ranked by product relevance times confidence. Use weekly before planning, or on demand when a major launch or policy change breaks.

  • 4 files
  • 19.5 KB
  • Updated 5 days ago
  • GitHub
Use this Skill: https://skilld.dev/gh/jokerhtml/needcraft-skills/trend-intelligence

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈108 tokens for metadata: the name and description. ≈2.9k when used: this file. ≈1.7k more on demand in 2 files.

Description uses 5.4% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈433 characters, or 5.4%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Trend Intelligence

What this skill is

A standalone weekly scan. It turns "what is happening in my niche right now?" into a dated file of ranked, evidence-tagged candidates, each with one content angle and one recommended format — and a short top-5 briefing for the person who plans the content.

It is a research routine, not a publishing tool. It never writes, schedules, or publishes content. It feeds the planning step, which has its own quality checks and approval.

Who it is for: a creator, educator, or small team publishing regularly in a fast-moving niche (AI tools, freelancing, content creation, digital products, or similar) who wants the week's plan grounded in what actually changed, not in last month's ideas.

Prerequisites

  • A defined niche and audience. Before your first scan, write down in one paragraph: who you serve, what problem you solve for them, and what you sell or want them to do. Every ranking decision in this skill refers back to that paragraph. Without it, "relevance" has no meaning.
  • Access to a web search tool and at least one social/community search route (for example, searching Reddit, X, or Threads). If a source is blocked or paywalled, record the gap and move on — never invent a finding to fill it.
  • A folder to keep dated scan files in, for example research/trend-scans/. Scans are only useful if you can compare this week with last week.

When to use

  • Weekly — suggested Monday, so the week's content plan can absorb it.
  • On demand before monthly or quarterly planning, or when a major launch or policy change breaks (a new model release, a platform rule change, a payment-platform fee change).

Inputs

None required. Optional: focus areas from the content owner, for example "dig into AI video tools this week" or "check what freelancers complain about on X". A focus area narrows the scan; it never overrides the evidence tags.

Process

1. Scan across the lanes

Cover every lane each week, even briefly. A lane with nothing new is recorded as "nothing new", not skipped silently.

Lane What to look for
AI tool updates New models, new features, price or limit changes, deprecations
User questions What people in your niche are asking right now (communities, comments, forums)
Creator problems What creators in your niche struggle with this week
Freelancer / audience problems What your audience complains about, fails at, or pays to fix
Content-creation problems Formats, hooks, distribution, editing — what is working or breaking
Automation Workflows, tools, and integrations people are adopting or abandoning
Digital products / online business Pricing, platforms, launches, refund and payout changes
Platform changes Rule, algorithm, format, or monetisation changes on the platforms you publish on

2. Record each candidate

For every candidate, record all nine fields. A candidate missing a field is incomplete, not ranked.

  • TOPIC — one line, specific (not "AI video" but "YouTube now counts a Shorts view from the first frame").
  • SOURCE — the exact source, with a link where one exists.
  • DATE — the date of the source, and the date you checked it.
  • WHY IT IS CURRENT — what changed, and when.
  • AUDIENCE PROBLEM — the concrete problem this creates or solves for your audience.
  • ANGLE — the single best content angle for your niche.
  • CONTENT FORMAT — the recommended format (carousel, short video, text thread, long post, newsletter section).
  • PRODUCT RELEVANCE — 1–5 (see the rubric below).
  • CONFIDENCE — 1–5 (see the rubric below).

3. Classify every candidate

The tag is mandatory. Anything untagged is rejected from the briefing.

  • FACT — verified, citable source with a date. The claim as written appears in the source you opened.
  • SIGNAL — indicative but weak: community reports, a single anecdote, a pattern you noticed but cannot verify. A signal is evidence of a reported experience; use it to identify what should be tested, never as a fact in content.
  • SPECULATION — unverified. Never presented as fact, never enters content claims. Speculation may be listed in the scan file under "watched", never in the top-5 briefing.

4. Rank with the rubric

Score = PRODUCT RELEVANCE × CONFIDENCE, each on a 1–5 scale. Maximum 25.

Product relevance (1–5)

Score Meaning
5 Directly feeds your offer: the audience problem is the problem your product solves
4 Strong fit: same audience, adjacent problem, natural bridge to your offer
3 Same audience, weak bridge: interesting, but the product link would feel forced
2 Same broad niche, different audience segment
1 Interesting to you, not to your buyer

Confidence (1–5)

Score Meaning
5 Official source (docs, changelog, announcement) opened and dated
4 Reputable independent test or reporting, with a named source
3 Multiple independent community reports saying the same thing
2 Single community report or anecdote
1 Rumour, prediction, or your own hunch

Tie-breaks, in order: (1) higher confidence wins; (2) more recent source date wins; (3) the candidate whose audience problem is most specific wins. A high-confidence topic with no product relevance still loses to a medium-confidence topic that feeds your buyer pool — that is the point of multiplying, not adding.

Worked ranking example: a verified platform rule change scores relevance 4 × confidence 5 = 20. A widely discussed rumour about a new tool scores relevance 5 × confidence 1 = 5. The rule change ranks above the rumour, even though the rumour is more exciting.

5. Output

  • A dated scan file (see the output template below) with all candidates, tags, and scores.
  • A top-5 briefing: the five highest-scoring FACT or SIGNAL candidates, each with topic, why it matters now, the single best angle, and the format. Concise, no fluff. SPECULATION never enters the briefing.

Search patterns (per lane)

Use text queries, never bare URLs. Open the primary source to verify any claim you intend to tag FACT. These are starting patterns — adapt the niche terms to yours.

  • AI tool updates: "[tool name] changelog", "[tool name] new feature", "[tool name] pricing change", news vertical for model releases.
  • User questions: site or community search for "how do I" + [niche problem], "struggling with" + [niche task]; sort by recent.
  • Creator / freelancer problems: "freelancers" + "not getting replies", "creators" + "reach dropped", recent posts in your niche's communities.
  • Platform changes: "YouTube Shorts" + "change", "Instagram" + "algorithm", "[platform]" + "policy update", plus each platform's official help or blog.
  • Digital products: "Gumroad" + "fees", "[platform]" + "payout", "digital product" + "pricing".

For every lane: prefer official docs first, then official announcements, then primary research, then reputable independent tests, then community signal. Random summaries are leads, not sources.

Quality rules

  • Never turn community complaints into universal facts. A complaint is evidence of a reported experience; use it to identify what should be tested.
  • Verify time-sensitive claims at scan time: last month is not today. Re-check AI tool features, platform rules, pricing, and fees every scan.
  • Every FACT needs a citable source with a date. No source, no FACT.
  • Every trend must pass this gate before entering the plan: Is it real? Is it current? Is it relevant? Can we add original value? Does it fit our niche? A "no" on any question parks it under "watched, no angle yet" — never force a product connection.
  • Date-check every source during the scan. Flag anything older than 30 days unless it is a standing fact (a standing rule, not a news item).

Failure modes

  • Stale sources presented as current. Mitigate: date-check every source; flag anything older than 30 days unless it is a standing fact.
  • Hype without evidence. Mitigate: FACT / SIGNAL / SPECULATION tags are mandatory on every candidate; anything untagged is rejected.
  • Trend with no product angle. Mitigate: park it in the scan file under "watched, no angle yet" — never force a product connection.
  • Thin week. If fewer than 3 verifiable candidates emerge, deliver a thin-but-honest scan rather than padding with speculation.

Output template (scan file)

Save one file per scan, named trend-scan-YYYY-MM-DD.md, in your research folder. Use these sections, in this order:

# Trend scan — YYYY-MM-DD

## Candidates (ranked)

### 1. [Topic] — Score [R×C = total] — [FACT / SIGNAL / SPECULATION]
- Source: [name + link]
- Source date: [date] · Checked: [date]
- Why it is current: [one line]
- Audience problem: [one line]
- Angle: [one line]
- Format: [carousel / short video / thread / post]
- Relevance: [1–5] · Confidence: [1–5]

## Watched, no angle yet
- [Topic] — [one line on why it is parked]

## Top-5 briefing
1. [Topic] — why it matters now — angle — format
   (repeat for each of the top 5)

## Gaps and blocked sources
- [Any source that was paywalled, blocked, or unverifiable]

A filled example scan file is in examples/sample-scan-2026-11-16.md.

Example candidate row

The row below is a real example, verified 2026-11-16. It shows the standard a FACT row must meet.

  • TOPIC: YouTube counts a Shorts view from the first frame of playback.
  • SOURCE: YouTube's official blog announcement of the change, reported in trade coverage dated August 2026; YouTube's Shorts search and discovery help page.
  • DATE: Change effective 24 August 2026; checked 16 November 2026.
  • WHY IT IS CURRENT: The counting rule changed in August 2026, so raw view counts before and after the change are not comparable.
  • AUDIENCE PROBLEM: Creators misread a views jump as a performance jump.
  • ANGLE: "Your Shorts views went up in August. Your retention didn't — here's the metric that actually matters."
  • CONTENT FORMAT: Short video (60 seconds).
  • PRODUCT RELEVANCE: 3 (same audience, weak product bridge).
  • CONFIDENCE: 5 (official announcement).
  • TAG: FACT. SCORE: 3 × 5 = 15.

Verification

  • Spot-check 2–3 sources per scan by re-opening them and confirming the claim still reads the same.
  • A human reviews the angles before any candidate enters the content plan. The scan proposes; the plan disposes. The reviewer may demote any candidate one relevance point with a one-line reason, recorded in the scan file.

Stop conditions

  • Stop if key sources are paywalled or blocked with no accessible alternative — report the gap, do not invent content to fill it.
  • Stop if fewer than 3 verifiable candidates emerge — deliver a thin-but-honest scan rather than padding with speculation.
  • Never publish or schedule content from the scan. It feeds planning, which has its own quality checks and approval gates.

Known limits

  • This skill is a research routine. It does not verify the truth of a platform's own claims beyond reading the primary source, and it does not track whether a trend held up after the scan — re-check before reusing an old candidate.
  • Community signal is directionally useful and frequently wrong in detail. Treat it as a prompt to verify, never as a source for a factual claim.
  • Scores are a ranking aid for one operator's judgement, not an objective measurement. Two operators scoring the same candidate may differ by a point; the tie-breaks exist for that reason.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 90f40b1. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 4 days ago.

Activeupdated 5 days ago
Other metadata
metadata
{
  "version": "1.0.0",
  "date": "2026-11-16",
  "author": "Muhammad Saad Ranjha / NeedCraft"
}

README badge

README badge for jokerhtml/needcraft-skills/trend-intelligence