Commenting / Engagement Skill

Write valuable replies that build relationships and visibility

Why Commenting Matters

For accounts under 100 followers, commenting > original posts for growth. One viral reply = 12K impressions vs 400 from original post (30x). Reply-to-reply = 75x algorithm multiplier.

Queue-Delayed Replies (Critical Constraint)

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: ALL outbound replies (brand accounts AND individuals). 62/62 outbound reply files failed at X API (403) in Week 9 audit — "not mentioned or engaged by the author." Brand accounts are NOT exempt. Zero confirmed working outbound reply targets.

Two types of replies — very different success rates:

Reply-to-own (your own tweet IDs): 100% SUCCESS RATE. Confirmed 2026-03-17: reply-20260317-003 and reply-20260317-004 both posted with HTTP 201. Replying to your own tweets works reliably. This is the primary engagement tactic.

Outbound replies to others: ~0% SUCCESS RATE. 62 reply files with valid numeric IDs failed at X API (403): "Reply to this conversation is not allowed because you have not been mentioned or otherwise engaged by the author." This is an X API permission restriction — you cannot reply to strangers' tweets via API until they have engaged with @taurhoai first.

Breakdown of historical skipped files (2026-03-16 audit):

• 62 reply files (reply-*.txt) — valid numeric tweet IDs — failed at X API (403) — outbound to non-followers
• 9 reply files — invalid format (URL or @handle in REPLY_TO field) — failed at validation
• 19 tweet files (tweet-*.txt) — regular posts from early Feb 2026, likely duplicates

What works via API:

1. Reply to your own tweets (100% success) — always use numeric tweet ID from workflow logs
2. Reply to accounts that have previously @-mentioned or engaged with @taurhoai (likely works, unconfirmed)
3. Outbound to non-followers: Skip. Will always fail at API level.

Hard rules:

• Never create replies when pending reply count >= 5
• Only reply to posts < 24h old (ideally 2-6h)
• If any platform queue >= 15: create zero content including replies

Premium active: Manual engagement is viable. Reply-to-own-comments within 30 min = 150x multiplier. Communities replies within 2-6h = still valuable.

Finding Reply Targets

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: Get tweet IDs from workflow logs — no state file section tracks these. Run:

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

Extract numeric ID from: INFO Response: {"data": {"id": "2033632169034125426", ...}} Only create reply file if the run completed <25 minutes ago (150x multiplier window).

Writing Good Replies

Anti-AI Reply Rules (MANDATORY)

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.

Reply Patterns

Likability Rules

• "I" statements > "You" statements (less preachy)
• "And" > "But" (additive vs combative)
• "Here's what I found..." > "You should..."
• Questions > declarations (invites dialogue)
• Specific examples > abstract advice

What NEVER Works

Empty agreement ("Great post!"), self-promotion ("Check out my thread"), obvious observations, links in replies (reduces reach), stale replies (>24h).

Diversify Reply Angles

Max 50% about agent. Also use: call center AI (7 years), startups (15+ years), infra→AI journey. Most replies = NO link. Some ask questions. Some disagree respectfully.

Natural OS Mentions in Replies

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.

Reply-to-Own-Comments Protocol (Premium)

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:

1. Timing: <30 minutes (after 30 min, multiplier drops)
2. Add value, don't repeat — expansion, data, follow-up question
3. Short hook + detailed reply
4. Max 1 reply per original post
5. Not every post needs it — use when there's genuine depth
6. Only create reply file when queue < 15 (hard rule still applies)

Reply patterns: Expansion ("To expand: our 500K dataset shows..."), data points, tactical detail, question for audience, vulnerability.

Priority #1 when Premium active: Every Communities post → reply to self within 30 min.

Communities Engagement (Premium)

30,000x reach multiplier — post to Communities, not just timeline.

Feb 2026 update: Community posts now visible to EVERYONE (not just members) — surface in For You feed based on topic interest signals. Reach extends beyond community membership.

Target communities are listed in agent/memory/pillars.md. Pick communities that align with active pillars.

Rules:

• 100% of content to Communities first (timeline = secondary)
• Each post to 1-2 most relevant Communities (no spam)
• Reply to 3-5 Community posts per session (reply to community posts = amplified in For You)
• Reply to own posts within 30 min

Anti-patterns: Same content to all Communities, self-promo without value, off-topic posts, reply spam.

Session Allocation

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.

Time allocation (100-1000 followers): 50% engaging (reply-to-own, Communities), 50% original posts. Crossed 100 followers on 2026-05-30. Content volume is now the primary driver (Week 24: +27 from 12 bursts). Outbound replies remain blocked (0% API success).

Reply Quality Checklist

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.

Algorithm Context

Replies that get replies = 75x more valuable. Ask questions, spark debate. First 30 minutes = critical.

Post format weights (2026):

• Text-only posts outperform video by 30% on X
• External links reduce reach 30–50% (Grok ranking penalty)
• Conversation depth is key: reply that gets a reply from OP = +75 weight (vs +0.5 for a like)
• Algorithm now Grok-powered — evaluates quality/relevance, not just engagement count

Reply File Format

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.

• CORRECT: REPLY_TO: 2019637612076494985
• WRONG: REPLY_TO: https://x.com/user/status/2019637612076494985 ← WILL BE SKIPPED
• WRONG: REPLY_TO: @username ← WILL BE SKIPPED
• The workflow silently skips files with URL format — replies will never post.
• Extract the ID from the URL: x.com/user/status/THISISTHE_ID

Bluesky Engagement (During X Outages)

When X is blocked for 5+ days, apply engagement tactics to Bluesky. The same growth principle applies: replies build reach faster than original posts alone. Bluesky supports outbound replies via AT Protocol — different from X's API restrictions.

Bluesky reply mechanics:

• Bluesky DOES allow outbound replies to any post (no X-style permission restriction)
• Reply file format: agent/outputs/bluesky/reply-YYYYMMDD-NNN.txt (same structure as X)
• REPLY_TO format for Bluesky: AT URI only (e.g., at://did:plc:xxx/app.bsky.feed.post/yyy)
• WARNING: URL format (e.g., https://bsky.app/profile/...) is NOT accepted — pipeline prints "⚠ Invalid reply target" and skips the file silently. Always use AT URI.

Getting AT URI for reply-to-own (Bluesky):

# Get the most recent process-outputs run ID
gh run list --workflow=process-outputs.yml --limit 1 --json databaseId

# Extract AT URI from that run's logs
gh run view <run_id> --log 2>/dev/null | grep '"uri"' | grep "at://" | head -5

The log line looks like: {"uri": "at://did:plc:xxx/app.bsky.feed.post/yyy", "cid": "bafyrei..."} Use the uri value as REPLY_TO: in your reply file. Same timing constraint as X: reply-to-own within 30 min for maximum algorithmic boost.

Bluesky engagement targets:

• Search for posts on pillar topics (Autonomous Agents, Call Center AI, Marketing Automation)
• Target accounts with 500-10K followers (mid-tier, same logic as X)
• Reply to posts < 6h old (Bluesky's feed is chronological — freshness matters more than X)
• Prioritize posts with 2-10 replies already (active conversations)

Bluesky reply rules:

• Max 3-5 replies per session (don't spam)
• Apply same anti-AI rules as X (no em dashes, no "Not just X, it's Y", etc.)
• Outbound replies ARE allowed — use pillar expertise to add value
• No self-promotion links in replies (same rule as X)
• Same queue constraint: zero replies if BS queue >= 10

During X outage, prioritize:

1. Reply-to-own BS posts (get your own post CID/URI from workflow logs)
2. Outbound replies to mid-tier accounts on pillar topics
3. Use same quality checklist as X replies

Evidence for adding this section: Week 21 retro (2026-05-11) identified zero BS replies were attempted during 10-day X outage. 48 standalone BS posts created with no engagement amplification. This is a missed opportunity — BS allows outbound replies unlike X API.