Revela
Turns Instagram, TikTok, X and LinkedIn screenshots into a structured markdown library. Detects multi-image carousels, groups slides from the same post, normalizes prompts to a portable XML format, and tags every piece by user-declared sectors, concrete utility, priority and effort. Sectors are configured by the user at runtime — nothing is hardcoded.
Installation
Extracts, translates and catalogs information from screenshots (typically social media posts) into a structured, queryable markdown library.
What it does
Processes folders of screenshots and produces a markdown catalogue with rich YAML metadata, ready to query with Obsidian + Dataview, filter by user-declared sectors, and feed editorial or knowledge-base pipelines.
When to use it
- You collect Instagram/TikTok/X/LinkedIn captures of prompts, post ideas, tools, courses or flyers.
- You want to stop losing them in a chaotic folder.
- You want a queryable library tagged by your own sectors and priority.
How it works
- Drop screenshots into
revela/inbox/. - Trigger the utility (e.g.
/revelain Claude Code, or natural language). - The agent groups carousels (multi-image posts), extracts text via vision, classifies, normalizes prompts and writes one
.mdper post. - Originals move to
revela/ready/, generated markdown lives inrevela/output/. - Indexes and a dashboard regenerate on every run.
Configuration
On first run, the agent asks which sectors apply to your context (e.g. personal, marketing, research, my-org). The answer goes into revela/.config.yaml and tags every catalogued piece. Sectors are never hardcoded — they’re always yours.
If you skip configuration, defaults are [work, personal, marketing, research].
Output schema
Every .md carries a YAML frontmatter with: identity, source, author + metrics, classification (type, sectors, actionable_as), decision (priority, effort, copy_risk, expires), connections (variables, tags, related, derived_idea). See SKILL.md for the full schema.
Compatibility
Tested with: Claude Code (primary), and any agent that supports vision and follows skill-style instruction loading. Mark compatible_with in utility.json per agent verified.
Limitations
- Optimized for the user’s language being any common Latin-alphabet language. Right-to-left or CJK scripts are best-effort.
- Carousel grouping relies on filename timestamps. If your capture tool strips timestamps, grouping falls back to author + visible pagination indicator.
- Vision extraction quality scales with model capability. Use a model with strong OCR-on-screenshots performance.
License
MIT — see repo root.