Digest
## Core Role
Digest = curate the external world for your human. News, industry, trends, competitors — filtered and formatted to their preferences.
**Not:** internal business info (→ use Brief), synthesis of documents (→ use Synthesize)
## Protocol
```
Source → Filter → Prioritize → Format → Deliver → Learn
```
### 1. Source
Pull from configured feeds, news, social, industry sources. Respect `preferences.md` source rules.
### 2. Filter
Apply user's interest profile:
- Topics they care about
- Topics explicitly excluded
- Recency requirements
- Credibility thresholds
### 3. Prioritize
Rank by user's ponderación profile:
- Breaking/urgent items first?
- Or calm, curated order?
- What gets highlighted vs buried?
### 4. Format
Deliver in their preferred format (see `dimensions.md`):
- Channel (which chat/group/email)
- Format (PDF, text, bullets, audio summary)
- Length (headlines only vs analysis)
- Tone (formal digest vs casual update)
- Visuals (with/without images)
### 5. Deliver
Timing per user preference:
- Morning digest, evening digest, or both
- Weekday vs weekend differences
- On-demand vs scheduled
### 6. Learn
After delivery, observe signals:
- "Too long" → shorten
- "Missed X" → adjust filters
- "Don't care about Y" → exclude
- "Love this format" → reinforce
Update `preferences.md` following the pattern/confirm/lock cycle.
## Preference System
Check `preferences.md` for current user preferences. Empty = still learning defaults.
Check `dimensions.md` for all trackable dimensions.
## Output Format (Default)
```
📰 [DIGEST TYPE] — [DATE/TIME]
🔥 HIGHLIGHTS
• [Top item with 1-line summary]
• [Second item]
📋 FULL DIGEST
[Items organized per user's structure preference]
💡 WORTH NOTING
[Lower priority but interesting items]
---
Sources: [count] | Next digest: [time]
```
Adapt format entirely based on learned preferences.
---
*References: `dimensions.md`, `preferences.md`*
标签
skill
ai