wow-digest

Original：🇺🇸 English

Translated

5 scriptsChecked / no sensitive code detected

Daily digest of 3-7 genuinely surprising items from newsletters and Telegram channels. Scores content for epistemic friction, not just relevance. Appends to daily note. Use when the user says "/wow-digest", "run the wow digest", "what's surprising today", "morning reading", or "digest my newsletters".

4installs

Sourceglebis/claude-skills

Added on2026-05-24

NPX Install

npx skill4agent add glebis/claude-skills wow-digest

SKILL.md Content

View Translation Comparison →

wow-digest

Purpose

Pull last 24h of newsletters (email) and Telegram channel posts, filter noise, score survivors for genuine surprise against the user's focus and recent research, and append 3-7 WOW items to today's daily note.

Workflow

Run
```
scripts/ingest.py
```
to pull and normalize candidates from all sources
Run
```
scripts/enrich.py
```
to fetch full content for link-only newsletters (LinkedIn, beehiiv, Substack)
Run
```
scripts/salience_filter.py
```
to drop obvious noise (marketing, payments, greetings)
Run
```
scripts/wow_score.py
```
on filtered candidates to score and select WOW items
Append selected items to today's daily note under
```
## Reading
```
Save raw candidates to
```
.wow-eval/candidates/YYYYMMDD.jsonl
```
for replay
Archive processed newsletter emails via GWS
During eval phase: run
```
scripts/feedback.py
```
to collect human verdicts

Manual run

bash

python3 scripts/ingest.py --days 1 --output /tmp/wow-candidates.jsonl
python3 scripts/enrich.py --input /tmp/wow-candidates.jsonl --output /tmp/wow-enriched.jsonl
python3 scripts/salience_filter.py --input /tmp/wow-enriched.jsonl --output /tmp/wow-filtered.jsonl
python3 scripts/wow_score.py --input /tmp/wow-filtered.jsonl --output /tmp/wow-selected.json
# Then the skill appends to daily note and archives emails

Dry-Run Mode

When the user says

/wow-digest --dry-run

or "preview the digest", run the full pipeline but:

Do NOT append to daily note
Do NOT archive emails
Instead, print the selected items with scores and hooks directly in the conversation

This lets the user preview what would be appended without side effects.

Context Sourcing

The scoring prompt uses three context signals from the vault (

~/Brains/brain/

):

{focus}
— From
```
My Focus.md
```
, sections
```
## Current
```
,
```
## Base
```
,
```
## Primary
```
(stops at
```
## Nice to have
```
). This tells the scorer what the user cares about right now.
{research}
— From
```
ai-research/*.md
```
files (last 30 days), parsed from filenames (
```
YYYYMMDD-topic.md
```
) and
```
research_topic:
```
frontmatter. Shows what the user has already investigated.
{recent_topics}
— From
```
Daily/YYYYMMDD.md
```
headings (last 7 days), excluding
```
## do
```
and
```
## log
```
. Shows recent daily note themes.

If these files don't exist, scoring still works but with degraded personalization.

Dedup

Ingestion deduplicates against the last 7 days of

.wow-eval/candidates/*.jsonl

using SHA-256 hashes of

title|source_name

(case-insensitive). Same article shared to multiple channels or re-sent in a newsletter won't appear twice. Pass

--no-dedup

to

ingest.py

to skip.

Config

Edit

config/sources.yaml

to add/remove email patterns or Telegram channels. Edit

config/wow_prompt.txt

to tune the scoring prompt.

Output Format

After scoring, append to today's daily note (

Daily/YYYYMMDD.md

) ABOVE the

- - -

separator, below any existing content:

markdown

## Reading

- **[Title]** (Source) — hook explaining WHY it's surprising
- **[Title]** (Source) — hook
...

_WOW digest · N candidates → M selected · YYYY-MM-DD_

CRITICAL: Always run

date +"%Y%m%d"

to get today's date. Never assume.

If

## Reading

already exists in the daily note, append items to it rather than creating a duplicate section.

Eval Mode (first 2 weeks)

During eval phase, do NOT auto-archive. Instead:

Run ingest + scoring as normal
Present the selected items to the user with FULL CONTENT, not just titles. For each item show:
- Title + source
- The snippet (first 300-500 chars of actual content)
- The LLM's hook and challenged_assumption
- WOW score breakdown (relevance, surprise, bridge_value, predictability)
Show all items in a single text block first so the user can read the content
Then ask via AskUserQuestion: "Was this actually WOW?" with options: wow / meh / noise / already_knew
Record feedback via
```
scripts/feedback.py
```
Show current feedback stats
Only archive after user confirms

CRITICAL: The user CANNOT judge WOW from titles alone. Always show the snippet content. If the snippet is empty or too short, fetch the full email body via GWS before presenting.

To check if eval mode is active:

If
```
.wow-eval/feedback.jsonl
```
has fewer than 50 entries → eval mode
If 50+ entries → auto mode (archive without asking)

wow-digest

NPX Install

Tags

SKILL.md Content

wow-digest

Purpose

Workflow

Manual run

Dry-Run Mode

Context Sourcing

Dedup

Config

Output Format

Archive

Eval Mode (first 2 weeks)