This skill allows Hermes to search, analyze, and retrieve information from a user's exported Facebook archive.
The archive is expected to be exported from Facebook Account Center using:
- Format: JSON
- Date Range: All Time
- Media Quality: High
- All available categories selected
The skill must operate entirely on local files and never access Facebook directly.
The skill must support:
- Search posts by Vietnamese keywords.
- Search comments by Vietnamese keywords.
- Search reactions (Like, Love, Care, Haha, Wow, Sad, Angry).
- Find reactions related to posts containing specific keywords.
- Link comments back to their original posts.
- Display Facebook URLs whenever available.
- Provide direct links so the user can manually delete posts or comments.
- Generate timeline and statistics.
- Work correctly with Vietnamese Unicode text.
When a Facebook export archive is uploaded:
-
Extract ZIP files.
-
Parse all JSON files.
-
Detect and normalize:
- Posts
- Comments
- Reactions
- Stories
- Videos
- Photos
- Groups activity
- Pages activity
-
Store normalized data in SQLite.
Suggested tables:
- post_id
- timestamp
- title
- content
- author
- source_file
- facebook_url
- raw_json
- comment_id
- post_id
- timestamp
- content
- facebook_url
- raw_json
- reaction_id
- target_type
- target_id
- reaction_type
- timestamp
- facebook_url
- raw_json
Create FTS5 indexes.
Requirements:
Must support:
- Unicode Vietnamese
- Accent-sensitive search
- Accent-insensitive search
- Exact phrase search
Examples:
Search:
Orange Pi
Must find:
Orange Pi 5
Orange Pi Zero 3
Search:
AI Agent
Must find:
AI Agent
AI agent
AI-Agent
Search:
"Orange Pi Zero"
Must return exact phrase matches first.
Search only literal text.
Example:
Find all posts containing:
Orange Pi
Return:
- Date
- Content snippet
- Post URL
Search comments only.
Example:
Find all comments containing:
Gemma
Return:
- Comment content
- Date
- Parent post title
- Parent post URL
- Comment URL
Use embeddings if available.
Example:
Find posts related to:
Edge AI
Even if the words "Edge AI" do not appear.
Return most relevant results.
The skill must be able to analyze reactions.
Examples:
Find all posts that received Angry reactions.
Find all posts that received Love reactions.
Count total Like reactions during 2025.
Return:
- Post title
- Post URL
- Reaction count
- Reaction breakdown
The skill must support:
Find posts containing:
Orange Pi
and show:
- Total reactions
- Like count
- Love count
- Haha count
- Wow count
- Angry count
Return top results ordered by reaction count.
When a matching comment is found:
The skill must identify:
- Original post
- Post author
- Post date
- Comment date
Output:
Post:
Comment:
This allows the user to understand the context of the comment.
Highest priority.
The skill must always try to recover:
Examples:
Examples:
https://www.facebook.com/?comment_id=
Use URLs found in export data whenever available.
If missing:
Generate URLs from available Facebook IDs.
If a URL cannot be reconstructed:
Display:
URL unavailable in archive.
For every result returned:
Display:
- Post URL
- Comment URL
- Open URL action
The skill must NEVER attempt deletion.
The skill only helps the user locate content.
The user performs deletion manually on Facebook.
Examples:
When did I first mention Orange Pi?
When did I first mention picoERP?
When did I first mention AI Agent?
Return:
- First occurrence date
- Matching content
- Facebook URL
Examples:
Top 100 most-used keywords.
Posts per year.
Comments per year.
Most reacted posts.
Most active months.
Most active years.
For every search result:
Date: YYYY-MM-DD HH:MM
Type: Post | Comment | Reaction
Content:
Facebook URL:
Parent Post URL:
Reaction Summary: Like: X Love: X Care: X Haha: X Wow: X Sad: X Angry: X
- Never modify archive files.
- Never delete Facebook content.
- Never log in to Facebook.
- Never request Facebook credentials.
- Never perform actions on behalf of the user.
- Read-only mode only.
The skill acts as a Facebook Archive Search Engine and Knowledge Retrieval System.