菜单
Strategy for engaging with others' posts across platforms (X, LinkedIn, etc.). Finding targets, writing valuable replies, building connections.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Gather context about repo owner, products, and domain. Build domain expertise by reading top voices. Use proactively before creating content.
Technical details for external platform integrations (APIs, credentials, rate limits). Use when posting, debugging issues, or adding new integrations.
| name | commenting |
| description | Strategy for engaging with others' posts across platforms (X, LinkedIn, etc.). Finding targets, writing valuable replies, building connections. |
| user-invocable | false |
Write valuable replies that build relationships and visibility
For accounts under 100 followers, commenting > original posts for growth. One viral reply can achieve 30x the impressions of an original post. Reply-to-reply = 75x algorithm multiplier.
Agent-created replies post hours to days late, killing algorithmic value.
Time decay: Replies lose 50% visibility every 6 hours. At 24h = ~6% visibility.
What works via queue: Reply-to-own only. Evergreen topics, no timing pressure.
What doesn't work reliably: Outbound replies to accounts that haven't engaged with you. X API restricts replies to strangers (403: "not mentioned or engaged by the author").
Two types of replies — very different success rates:
Reply-to-own (your own tweet IDs): HIGH SUCCESS RATE. Replying to your own tweets works reliably. This is the primary engagement tactic.
Outbound replies to others: UNRELIABLE. X API may reject replies to accounts that haven't engaged with you. Test cautiously.
Hard rules:
Premium active: Manual engagement is viable. Reply-to-own-comments within 30 min = 150x multiplier. Communities replies within 2-6h = still valuable.
DO reply to: Mid-tier accounts (10K-100K), posts 2-6h old, topics with real expertise, conversation-starters, accounts that engage back.
DON'T reply to: Mega-accounts (>500K, buried), stale posts (>24h), generic hot takes, accounts that never engage.
How to find (X API is write-only, use web search):
WebSearch: "site:x.com @username {topic}"
WebSearch: "site:x.com {topic} {current_year}"
Extract tweet ID from URL: x.com/user/status/**1234567890**
X Communities (Premium): Browse community feeds for fresh posts. Community replies get amplified in For You feed. Best leverage for small accounts.
Reply-to-own targets: Check agent/state/current.md for recent tweet IDs.
Never use: Em dashes joining clauses, "Not just X, it's Y", "Delve/elevate/innovative/landscape/leverage/robust", "To clarify/In other words", "Great point! Furthermore...", summarizing their post back.
Do use: Contractions, sentence fragments ("Wild." "Zero chance."), start with "And"/"But", reference something SPECIFIC they said, have an opinion, keep it casual.
| Pattern | When to use |
|---|---|
| Respectful disagreement | Sparks reply-to-reply (75x) |
| Add specific insight | Shows expertise without self-promo |
| Ask sharp questions | Pushes conversation forward |
| Share contrarian data | Adds new info |
| "This means..." prediction | Connect their news to a consequence |
| "What nobody's saying..." | Add the angle everyone missed |
Empty agreement ("Great post!"), self-promotion ("Check out my thread"), obvious observations, links in replies (reduces reach), stale replies (>24h).
Max 50% about agent. Also use: your domain expertise, founder/career experience. Most replies = NO link. Some ask questions. Some disagree respectfully.
When someone posts about a topic that overlaps with an owner repo, mention it naturally in a reply. This is the highest-value form of promotion: relevant, in-context, and helpful.
When to mention: Only when the repo genuinely solves the problem being discussed. If you have to stretch the connection, skip it.
How: Lead with insight or opinion first. Repo link goes at the end, casually. "We open sourced something similar" > "Check out my repo."
Frequency: Max 1 in 10 replies includes an OS link. Most replies should have zero self-promotion. Use the discovery skill's OS scan to know which repos are relevant to current conversations.
150x algorithmic multiplier — highest-leverage engagement tactic.
Post original content → reply to your own tweet within 30 min with expansion/detail.
CRITICAL: You need the numeric tweet ID of your own tweet. REPLY_TO: SELF is invalid and will be skipped. To get the real ID:
# Get the most recent process-outputs run ID
gh run list --workflow=process-outputs.yml --limit 1 --json databaseId
# Extract tweet IDs from that run's logs
gh run view <run_id> --log 2>/dev/null | grep '"id"' | grep -v "edit_history" | head -5
Or search more precisely:
gh run view <run_id> --log 2>/dev/null | grep 'INFO Response:' | head -5
The log line looks like: INFO Response: {"data": {"id": "2033632169034125426", "text": "Your tweet..."}}
Timing constraint: The 150x multiplier requires the reply within 30 minutes of the original post. Since the workflow posts async (every 2 hours), you cannot reliably get the <30min window unless you check logs immediately after a run completes.
Practical approach: Check gh run list --workflow=process-outputs.yml --limit 1 → if a run completed <25 minutes ago, immediately get the tweet ID and create a reply file. This is the only window where reply-to-own at 150x is achievable.
Rules:
Reply patterns: Expansion ("To expand: our dataset shows..."), data points, tactical detail, question for audience, vulnerability.
Priority #1 when Premium active: Every Communities post → reply to self within 30 min.
30,000x reach multiplier — post to Communities, not just timeline.
Community posts now visible to EVERYONE (not just members) — surface in For You feed based on topic interest signals.
Target communities are listed in agent/memory/pillars.md. Pick communities that align with active pillars.
Rules:
Anti-patterns: Same content to all Communities, self-promo without value, off-topic posts, reply spam.
If any queue >= 15: Zero content including replies. No exceptions.
If both queues < 15: Max 2 content pieces per session. Max 5 pending replies per platform.
Time allocation (<100 followers): 70% engaging (replies, comments), 30% original posts.
Must have: Adds insight OP missed, shows expertise, 2-4 sentences, would make someone click profile, post < 24h old, target is mid-tier or community post.
Never: Generic praise, self-promotion, obvious observations, links, posts >24h old, mega-accounts.
| Action | Weight vs Like |
|---|---|
| Reply-to-own <30min (Premium) | 150x |
| Reply-to-reply | 75x |
| Repost | 20x |
| Reply | 13.5x |
| Bookmark | 10x |
| Like | 1x |
Replies that get replies = 75x more valuable. Ask questions, spark debate. First 30 minutes = critical.
Post format weights (2026):
File: agent/outputs/x/reply-YYYYMMDD-NNN.txt
REPLY_TO: 2019637612076494985
---
Your reply text here.
CRITICAL: REPLY_TO must be the numeric tweet ID ONLY.
REPLY_TO: 2019637612076494985REPLY_TO: https://x.com/user/status/2019637612076494985 ← WILL BE SKIPPEDREPLY_TO: @username ← WILL BE SKIPPEDx.com/user/status/**THIS_IS_THE_ID**