Discovery Skill

Find voices, read content, build expertise

Gather context to make better decisions and create relevant content.

Discover Repo Owner

bash

gh api users/{owner}
# Returns: name, bio, blog, twitter_username, company, location

Additional sources:

•ME.md for links not in GitHub API
•Owner's GitHub profile README
•Pinned repositories
•Blog/website (from profile)

Discover Owner's Products

•Check pinned repos: gh api users/{owner}/repos?sort=updated
•Read each repo's README for product descriptions
•Look for live demos, documentation sites
•Check owner's blog for product announcements

Discover Domain Trends

For current goal, research:

•What works for similar accounts?
•Current trends in the niche
•Competitor analysis
•Platform algorithm changes

Use web search for fresh data:

code

WebSearch: "X Twitter growth strategies {current_year OR previous_year}"
WebSearch: "AI developer Twitter accounts successful"

Discovery Cadence

Type	Frequency	Purpose
Owner profile	Once, update if stale	Accurate promotion
Products	Weekly	New launches, updates
Domain trends	Per session	Stay current
Competitors	Weekly	Learn from others
Top voices list	Monthly refresh	Know who matters
Reading (top 5)	Every session	Stay sharp
Reading (6-20)	Weekly rotation	Broad awareness
Expertise synthesis	Monthly	Consolidate knowledge

Staleness Check

Research gets stale. Check dates and refresh:

•Owner profile: Refresh if >30 days old
•Products: Refresh weekly
•Domain: Fresh each session
•Top voices list: Refresh monthly
•Expertise docs: Re-synthesize monthly

Become a Domain Expert

The best content comes from deep knowledge. Systematically read and learn from top voices in the field.

1. Build a Top Voices List

Identify ~20 leading voices in the domain (based on current GOALS.md niche).

How to find them:

•Web search: "best {niche} blogs {current_year OR previous_year}", "top {niche} Twitter accounts"
•Look at who top accounts follow and retweet
•Check curated lists (awesome-lists, blog rolls, newsletter recommendations)
•Note authors cited repeatedly across sources

For each voice, capture:

markdown

## @handle / Name
- Platform: X / Blog / Newsletter / YouTube
- URL: [primary URL]
- Focus: [their specific niche/angle]
- Why follow: [what makes them valuable]
- Content style: [threads, essays, hot takes, tutorials...]
- Posting frequency: [daily, weekly, etc.]

Store in: agent/memory/research/top-voices.md

Refresh: Re-evaluate the list monthly. Drop inactive voices, add rising ones.

2. Regular Reading Routine

Each session, read fresh content from top voices to stay current.

Reading process:

•Pick 2-3 voices from the list (rotate through all ~20 over time)
•
Fetch their latest content:
- •Blogs: WebFetch their RSS/blog URL
- •X: Search for their recent posts via web search
- •Newsletters: Check their archive page
•
Read with intent - look for:
- •Key arguments and insights
- •Data points and stats worth citing
- •Emerging trends or shifts in thinking
- •Contrarian takes (agree or disagree?)
- •Gaps - what are they NOT talking about?

Reading cadence:

Source type	Frequency	Depth
Top 5 voices	Every session	Skim latest, deep-read standouts
Voices 6-20	Weekly rotation	Skim latest
New/trending	Per session	Quick scan for relevance

Capture Reply Targets During Reading (Week 2 Learning)

While reading top voices, also note reply-worthy posts. This eliminates the need for a separate "find reply targets" step.

For each reading session:

•Read the source for content notes (normal reading)
•Note 1-2 recent posts (< 48h old) from the author worth replying to
•Add to agent/memory/research/reply-targets.md with reply angle
•Create reply files in the same PR as content

This doubles the output of reading sessions (content + engagement) without extra search time.

Evidence: Week 2 retro — 0 replies in 2 weeks because reply discovery was treated as a separate task and never prioritized.

3. Synthesize Into Expertise

Reading without synthesis is wasted time. After reading, always produce output.

Per-article notes (when something is noteworthy):

markdown

# Reading: [Title]
Source: [URL]
Author: [name]
Date read: [YYYY-MM-DD]

## Key Takeaways
- [insight 1]
- [insight 2]

## Quotable / Citeable
- "[exact quote]" — useful for [context]

## My Take
- Agree/disagree because [reasoning]
- This connects to [other knowledge]

## Content Ideas Sparked
- [post idea inspired by this reading]

Store in: agent/memory/research/reading-notes/YYYY-MM-DD-slug.md

Monthly synthesis - consolidate reading notes into domain expertise:

markdown

# Domain Expertise: [Topic Area]
Last updated: [date]

## Current Consensus
[What most experts agree on]

## Active Debates
[Where experts disagree, with positions]

## Emerging Trends
[New ideas gaining traction]

## Contrarian Opportunities
[Gaps, overlooked angles, things no one is saying]

## Key Stats & Data Points
[Reusable facts with sources]

Store in: agent/memory/research/expertise/

4. Turn Reading Into Content

Reading directly fuels better publishing:

Reading output	Content use
Key takeaway	Authority post (share the insight with your angle)
Disagreement	Contrarian take (builds engagement)
Data point	Credibility boost in threads
Trend spotted	First-mover post (be early on trends)
Content gap	Fill it - own the topic others miss
Inspired idea	Original post with fresh framing

Rules:

•Never plagiarize - add your own perspective, experience, or framing
•Credit sources when directly building on someone's idea
•Aim for 1 reading-inspired post per 3-5 articles consumed
•The goal is informed originality, not summary

Find Reply Targets

Discover recent posts worth replying to (feeds the commenting skill). X API free tier is write-only, so use web search.

Search Process

Each session, find 2-3 reply targets:

By account (prioritize top voices from agent/memory/research/top-voices.md):

code

WebSearch: "site:x.com @handle {topic}"
WebSearch: "site:linkedin.com/posts {name} {topic}"

By topic (trending conversations in your niche):

code

WebSearch: "site:x.com {topic} {current_year}"
WebSearch: "site:x.com autonomous agents building"

By mention (people talking about you/your project):

code

WebSearch: "site:x.com {repo_name}"
WebSearch: "site:x.com @{owner_handle}"

Extracting Post IDs

•X: https://x.com/user/status/2019637612076494985 → ID: 2019637612076494985
•LinkedIn: post URL or URN from the share link

Prioritization

•Large accounts in your niche (more exposure per reply)
•Recent posts (within 24-48h — old replies get buried)
•Active conversations (more eyeballs on your reply)
•Topics where you have genuine insight
•People who engaged with your posts (reciprocity)

Store Targets

Save to agent/memory/research/reply-targets.md:

markdown

# Reply Targets
Last updated: YYYY-MM-DD

## Pending
- ID: 2019637612076494985 | @username | "Post summary..." | Reply angle: [your insight]

## Replied
- ID: ... | @username | Replied YYYY-MM-DD

Storing Discoveries

Store findings in agent/memory/. Structure:

•agent/memory/research/top-voices.md - curated voice list
•agent/memory/research/reading-notes/ - per-article notes
•agent/memory/research/expertise/ - synthesized domain knowledge
•agent/memory/research/reply-targets.md - posts to reply to
•agent/memory/research/ - other research and analysis

Using Discoveries

Discoveries inform:

•Publishing: What to promote, how to frame
•Content: Topics that resonate with audience
•Strategy: What's working in the space
•Expertise: Deep knowledge that makes content authoritative
•Originality: Gaps and angles others aren't covering