Trend Analysis

Skill: trend-analysis

Purpose

Build a comprehensive, multi-platform view of what is trending in a market category. Used in the market-deep-dive workflow and as a standalone research tool. Output is appended to the shared market intelligence file.

Input

• Topic or category to analyze (e.g., "nutrition apps", "personal finance", "AI automation agencies")
• The likely business_model (b2c | prosumer | b2b-smb | b2b2c) from idea.md, the user profile, or the user; it decides which platform set to recommend
• Platform prompt templates from skills/trend-analysis/prompts/ (see Prompts section below)

User Interaction

Before executing, ask the user which platforms to include in the analysis. Recommend a default set by business model and wait for their answer before proceeding:

Prompt to user:

Which sources would you like to include in this analysis? (select one or more)

1. TikTok — hashtag trends, viral content angles, creator gaps
2. Reddit — community pain points, recurring complaints, unmet needs
3. App Store — category rankings, new entrants, top review complaints
4. Web Search (Google) — search volume trends, rising queries, SEO demand
5. X/Twitter — public builder threads, product complaints, creator demand signals
6. B2B communities — G2/Capterra reviews, Hacker News, Indie Hackers, operator subreddits, LinkedIn, vendor pricing pages (recommended for business buyers)
7. All of the above — full multi-platform analysis (recommended for a new niche)

Wait for the user's selection. Accept answers like "1, 3", "TikTok and Reddit", "all", or "just App Store". Map their answer to the corresponding prompt files before continuing.

If the user selects multiple platforms, run each one sequentially and produce a separate output file per platform. Do not combine platforms into a single file unless explicitly asked.

Prompts Directory

Per-platform analyst prompt templates live in skills/trend-analysis/prompts/. Each contains a structured research brief with source tiers, output sections, and a [NICHE] or [CATEGORY] placeholder to fill before invoking:

Completed analyses are stored as individual files in memory/market_insights/ following the naming convention <niche>-<platform>-<YYYY>-<MM>.md. See memory/market_insights/README.md for the full schema.

Platform Coverage

Process

1. Load the platform prompt from prompts/<platform>.md and replace [NICHE] / [CATEGORY] with the target topic.
2. Execute the prompt: research the platform using the source tiers and output structure defined in the prompt.
3. Score trend velocity: rising-fast / rising / stable / declining.
4. Identify the strongest creator/content angle and any monetization evidence.
5. Write findings to a new file: memory/market_insights/<niche>-<platform>-<YYYY>-<MM>.md. The narrative must end with a ## Sources section listing every URL consulted as a markdown link; downstream skills copy these URLs into their own sources arrays, and a file with no URLs fails python tests/validate_memory.py.
6. Append a row for the new file to the index table in memory/market_insights/README.md.

One file per platform run. To cover all platforms, run this skill once per platform and produce one file each.

When a platform returns nothing usable

Some platforms have no signal for some niches. A search of climbing communities for training discussion returns peer-reviewed hangboard studies; a search of TikTok for compliance tooling returns nothing at all. When that happens:

1. Do not write a file. An insight file with no community evidence in it is worse than no file, because downstream skills treat it as researched ground.
2. Do not substitute a different platform's data under the empty platform's name.
3. Record the attempted queries and the outcome in the inputs_missing field of whichever skill consumes the analysis, and in scores.json.missing_inputs.
4. Lower score_confidence in idea-scoring by one level if the missing platform was one of the two primary sources for the business model.
5. Say so plainly when presenting: which platform was attempted, what came back, and what was used instead.

Two platforms attempted and one usable is a normal outcome, not a failure. Fabricating the second is the failure.

Output

Write to memory/market_insights/<niche>-<platform>-<YYYY>-<MM>.md.

The file has two parts:

Part 1 — YAML Frontmatter (machine-readable summary)

---
niche: <topic>
platform: tiktok | reddit | apps | web-search | x-twitter | b2b-communities | multi
analyzed_at: YYYY-MM-DD
status: fresh
stale_after: YYYY-MM-DD   # 6 months after analyzed_at
trend_velocity: rising-fast | rising | stable | declining
overall_verdict: hot | warm | cool | cold
key_insight: "<one-sentence takeaway>"
top_signals: []           # 3–5 bullet strings: top hashtags, subreddits, queries, or app categories
monetization_evidence: [] # 1–3 strings: existing products/revenue that confirm willingness to pay
---

Part 2 — Full Narrative Analysis (long-form Markdown)

Follow the output structure mandated by the platform prompt template exactly

The narrative must be self-contained — a downstream skill reading only this file should have everything it needs to map trends to product opportunities.