Social data

Social Data

Use this skill for two related jobs:

• fetch: collect public social posts from a specified platform and return structured, deduplicated items.
• analyze: calculate and interpret metrics from user-provided social posts, campaign exports, or fetched social samples.

Keep the two steps distinct. Fetching returns evidence samples; analysis turns user-provided or fetched data into metrics and recommendations.

When To Use

• The user asks what people are saying about a brand, product, competitor, topic, or pain point on Xiaohongshu, X/Twitter, Reddit, YouTube, or Bilibili.
• The user provides social post/campaign data and asks for engagement rate, CTR, ROI, ROAS, CPC, CPE, CPM, CPA, top/bottom posts, platform comparison, or content recommendations.
• The user wants social evidence for reputation, buzz, user complaints, market feedback, campaign performance, or next content experiments.

Do not use for:

• Private messages, login-gated content, paid API bypassing, scraping restricted pages, or non-public data.
• Claiming that zero fetched results means no discussion exists.
• Treating benchmark comparisons as facts without user-provided or current benchmark context.
• Replacing platform analytics exports when the user needs complete official reporting.

How To Call

1. Choose the mode:

   • fetch: user needs public posts or discussion samples.
   • analyze: user already has post/campaign data or wants metrics/recommendations from fetched samples.
   • fetch_then_analyze: user asks for a topic analysis and no data is provided.

2. For fetch mode:

   • Follow references/fetching.md.
   • One platform per script call: xhs, twitter, reddit, youtube, or bilibili.
   • Expand the user's topic into 3-8 short divergent keyword groups unless the user explicitly restricts keywords.
   • Report diag.status, failures, empty results, and platform dependency limits.

3. For analysis mode:

   • Follow references/metrics.md.
   • Validate fields before calculating.
   • Label evidence as direct calculation, limited inference, or assumption.
   • Separate organic and paid performance when possible.

4. Use scripts only when useful:

   • Fetch public samples: $ORKAS_NODE $ORKAS_PC_DIR/bin/run-skill.cjs social-data fetch -- <platform> <keywords> [options]
   • Calculate metrics from JSON: $ORKAS_NODE $ORKAS_PC_DIR/bin/run-skill.cjs social-data calculate_metrics -- data.json
   • Analyze performance from JSON: $ORKAS_NODE $ORKAS_PC_DIR/bin/run-skill.cjs social-data analyze_performance -- data.json

5. Return traceable conclusions.

   • Tie claims back to item URLs, post IDs, comments, or user-provided rows.
   • Include data quality warnings and next tests.
   • Do not synthesize sample posts when fetch fails.

Default Output

For fetched discussion:

# Social Data Fetch Result

## Scope
- Platform:
- Topic:
- Keywords:
- Status:

## Sample Summary
- Items:
- Top repeated themes:
- Notable examples:

## Diagnostics
- Raw hits:
- Deduplicated:
- Failed:
- Limits:

## Next Analysis
- ...

For performance analysis:

# Social Performance Analysis

## Bottom Line

## Data Quality
- Sample size:
- Missing fields:
- Abnormal values:
- Confidence:

## Key Metrics
| Metric | Value | Meaning | Evidence |
|---|---:|---|---|

## Top / Bottom Content
| Rank | Post | Platform | Format | Metric | Possible reason |
|---|---|---|---|---:|---|

## ROI And Cost
- Spend:
- Revenue / value:
- ROI:
- ROAS:
- CPC / CPE / CPA:

## Recommendations
1. ...

External Dependencies

• Python 3 for bundled scripts.
• Fetch mode may need requests, browser_cookie3, curl_cffi, xreach, yt-dlp, or a local Xiaohongshu proxy depending on platform.
• Analysis mode needs user-provided JSON, table data, or exported platform/campaign metrics.

Limits And Known Issues

• Fetch coverage varies by platform, login state, rate limits, anti-bot behavior, and local dependencies.
• Public fetches are samples, not complete platform analytics.
• Anonymous Reddit/Bilibili recall can be weak; browser login cookies may improve results.
• Xiaohongshu requires an external local proxy service at http://localhost:18060.
• ROI estimates based on engagement value are assumptions unless the user provides real revenue, conversion value, or lead value.
• Benchmark values can become stale and should be treated as rough context, not a pass/fail truth.