Menu
Gather context about repo owner, products, and domain. Build domain expertise by reading top voices. Use proactively before creating content.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | discovery |
| description | Gather context about repo owner, products, and domain. Build domain expertise by reading top voices. Use proactively before creating content. |
| user-invocable | false |
Find voices, read content, build expertise
gh api users/{owner} # Returns: name, bio, blog, twitter_username, company, location
Additional: Check ME.md, GitHub profile README, pinned repos, blog/website.
Products: gh api users/{owner}/repos?sort=updated → read READMEs, check live demos.
Staleness: Owner profile >30 days = refresh. Products = weekly.
Web search for current data:
"X Twitter growth strategies {current_year}""AI developer Twitter accounts successful""{niche} best practices {current_year}"Refresh each session (trends change constantly).
Find via web search ("best {niche} blogs", "top {niche} Twitter accounts"), follow-the-follows, curated lists.
Store in agent/memory/research/top-voices.md:
## @handle / Name
- Platform: X / Blog / Newsletter
- Focus: [niche/angle]
- Why follow: [value]
Refresh monthly.
Each session, pick 2-3 voices. Read with intent — look for: key arguments, data points, emerging trends, contrarian takes, gaps.
Cadence: Top 5 voices every session (skim). Voices 6-20 weekly rotation.
X outbound replies fail 100% of the time via X API (403) — confirmed Week 9 audit: 62/62 failed. Do NOT spend turns looking for X posts to reply to outbound.
For X: Reply-to-own only (100% success rate). Look for tweet IDs from recent workflow runs:
gh run list --workflow=process-outputs.yml --limit 1 --json databaseId,createdAt
gh run view <run_id> --log 2>/dev/null | grep 'INFO Response:' | head -5
For Bluesky (especially during X outages): Outbound replies ARE allowed. While reading top voices and doing research, note any Bluesky posts on pillar topics that are <6h old and worth replying to. Capture the AT URI for the reply file. See "Reply Targets: Platform-Specific Rules" below.
| Reading output | Content use |
|---|---|
| Key takeaway | Authority post (your angle) |
| Disagreement | Contrarian take |
| Data point | Credibility boost |
| Trend spotted | First-mover post |
| Content gap | Fill it — own the topic |
Rules: Never plagiarize. Credit sources. Aim for 1 reading-inspired post per 3-5 articles. Goal = informed originality, not summary.
Research in agent/memory/research/ helps THIS session. Research in .claude/skills/ helps ALL future sessions.
When to graduate: Substantial research (15+ sources), validated and actionable, broadly applicable.
Follow "Skill Development (High Bar)" from CLAUDE.md. NOT every session — reserve for validated, substantial findings.
Periodically scan the owner's public repos for promotable content. This feeds the publishing skill's OS promotion allocation.
When: Once per session (during research phase), or when looking for content ideas.
How:
ME.md → find the owner's GitHub profile URL under "Open Source (Promotable)"What makes something promotable right now:
A. Live outcomes (highest priority) — Real content/services produced by agents. These are proof, not promises.
ME.md for known live outcome URLs, then WebFetch each to check for recent articles, digests, postsB. Repos with live outputs — Code that powers running services. Link both the code AND the output.
C. Star milestones — Crossing 5, 10, 25, 50, 100 stars.
D. Trending topic overlap — Repo solves something people are discussing this week. Strongest combo: trend + live proof.
E. New repos — Launch announcements get one-time boost.
What to capture: For each candidate: what it does (1 line), proof it works (live output links + recent specific content), hook angle (why someone cares today). Store in agent/memory/research/os-promo-candidates.md.
Cross-reference with trends: Trending topic aligns with an owner repo or outcome? Post about the trend, link the repo/outcome as "we built this, and here's it running."
Reference: GitHub accounts and known live outcomes in ME.md under "Open Source (Promotable)".
DO NOT search for outbound X reply targets. All outbound replies fail at X API (403) — confirmed 62/62 failures in Week 9 audit. No exceptions for brand accounts or individuals.
Reply-to-own is the only working X reply strategy. To get targets:
gh run list --workflow=process-outputs.yml --limit 1 --json databaseId,createdAt → gh run view <id> --log | grep 'INFO Response:'Only create X reply files when:
Bluesky DOES allow outbound replies to any post via AT Protocol — no X-style permission restriction. During X outages, Bluesky outbound replies are a valid engagement tactic.
Finding Bluesky reply targets:
"bsky.app {topic} {current_year}" or browse topic feedsBluesky reply format: Use AT URI in REPLY_TO: header (see integrations skill for exact format).
Rules: Max 3-5 replies per session, apply same anti-AI writing rules, no self-promotion links, BS queue < 10 required.
See commenting skill ## Bluesky Engagement (During X Outages) for full detail.
agent/memory/research/top-voices.md — curated voice listagent/memory/research/reading-notes/ — per-article notes (optional)agent/memory/research/expertise/ — synthesized domain knowledgeagent/memory/research/reply-targets.mdContent strategy for external platforms (X, LinkedIn, etc.). Voice, style, and growth strategies.
Technical details for external platform integrations (APIs, credentials, rate limits). Use when posting, debugging issues, or adding new integrations.
Strategy for engaging with others' posts across platforms (X, LinkedIn, etc.). Finding targets, writing valuable replies, building connections.