Boost X Engagement

Hourly cycle that scans x.com for content matching your niche and posts a small organic-looking batch on YOUR X account. Two execution modes.

• Interactive mode (invoked via /implexa:run or "Run Now"): every post gates on explicit per-item human approval.
• Scheduled (auto) mode (auto-fired by the Implexa schedule wrapper): auto-posts under YOUR strict safety rails for originals and plain RTs; replies are drafted and queued but NEVER auto-posted in scheduled mode.

Step 0 — First-run config (ask once, persist forever)

Before doing ANYTHING else, check if a config block exists in the conversation/project context. A config block looks like this:

boost-x-engagement-config:
  x_handle: "@yourhandle"
  topic_signals: ["AI", "Claude", "agent", "RAG"]       # words your originals should mention
  voice_description: "punchy lowercase tech-bro, controversial-but-correct, never em-dashes"
  rt_allowlist:   ["@karpathy", "@simonw", "@sama"]     # who you'll auto-RT in scheduled mode
  banned_phrases: ["unlock", "100%", "secret", "🧵"]    # auto-skip originals containing any of these
  daily_caps:     { originals: 6, rts: 18 }
  schedule_window: { start_hour_local: 6, end_hour_local: 20 }

If the config is present

Parse it, hold the values in memory for this run, proceed to Step 1.

If the config is MISSING

Use AskUserQuestion to collect it interactively — one question at a time. Required fields:

1. What's your X handle? (free text, must start with @)
2. What niche / topics do you post about? (suggest 4 chips: AI / dev, crypto / defi, fitness, design / ux, plus Other for free-text)
3. Describe your voice in one sentence (free text — examples to hint: punchy lowercase tech-bro, professional thread-style, casual snark with emojis, dry observational)
4. For scheduled auto-RT, list 3-10 X accounts you fully trust to amplify (free text comma-separated; tell user "leave blank to disable auto-RT entirely — only interactive-mode RTs will fire")
5. Any phrases that should auto-block originals? (free text; suggest defaults: unlock, 100%, secret, here's how, this changes everything, PSA, 🧵)
6. Daily caps? (4 chips: Conservative (3 originals, 9 RTs), Moderate (6 originals, 18 RTs) — Recommended, Aggressive (12 originals, 36 RTs), Other for custom)
7. Run window in YOUR LOCAL TIME? (4 chips: 6AM-8PM (waking hours) — Recommended, 9AM-6PM (work hours), 24/7, Other for custom)

After collecting all answers, render the resulting YAML config block and tell the user:

> "Saved the config to memory for this run. To skip this prompt in future runs, paste this YAML block into your CLAUDE.md file at the top of your project (or into ~/.claude/CLAUDE.md for global access). Want me to write it there now? [yes / no / show me where]"

If yes, write it to the appropriate path using the Write tool.

Step 1 — Connect to Chrome and verify the active X account

1. mcp__Claude_in_Chrome__list_connected_browsers → find the local Chrome.
2. mcp__Claude_in_Chrome__select_browser with the local deviceId.
3. mcp__Claude_in_Chrome__tabs_context_mcp with createIfEmpty: true.
4. navigate to https://x.com/home.
5. find for account avatar / "@" handle in the account menu region.

HARD SAFETY GATE: if the active account does NOT match config.x_handle, abort immediately and surface the mismatch. Never post on the wrong account.

Common failure: Chrome is logged into a different X account. Tell the user to switch accounts in Chrome and retry.

Step 2 — Scan the timeline

get_page_text + read_page on x.com/home to extract:

• For You timeline articles (top 6-10 tweets)
• Today's News panel (curated trending topics)
• Trending now panel

Catalog candidates by category:

• Tweets matching your topic_signals → potential RT or QT material
• High-velocity tweets from accounts in your rt_allowlist → high-priority RT targets
• High-velocity tweets in your niche (any author) → potential reply targets
• Listicle bait, hype-bait, or unverifiable claims → flag and skip

Step 3 — Roll counts and draft

Per-run output shape:

• 1 original tweet (exactly 1)
• 1-5 replies (roll a number)
• 1-5 plain RTs (roll a number; if rt_allowlist is empty, set RT count to 0)

Voice rules for originals

• Match config.voice_description style and register
• Topic-dense: contain ≥1 word from config.topic_signals
• ≤ 280 chars
• Zero config.banned_phrases

Reply target selection

• High-velocity tweets (>10K views or >100 replies in last 24h) in your niche
• Skip flagged tweets (fabricated claims, hype-bait, anything off-topic from config.topic_signals)

RT target selection

Interactive mode: more permissive; surface rationale for each target.

Scheduled mode: must pass Step 4 strict filter.

Step 4 — Approval gate (mode-dependent)

INTERACTIVE MODE

Surface ALL drafts in a single message. For each item include text + target tweet URL (replies/RTs) + 1-line rationale. Wait for explicit per-item approval ("post 1, 3, 5" / "post all" / "rewrite 2") before ANY X action. NEVER post without an explicit approval string in chat.

SCHEDULED (AUTO) MODE — Strict safety rails

Originals — auto-post only if ALL true:

• ≤ 280 chars
• Zero phrases from config.banned_phrases
• Contains ≥1 word from config.topic_signals
• Voice-lint Haiku call returns strictly Y to: "Does this match the voice: '{config.voice_description}'? Y/N. Treat ambiguous as N."
• Daily cap: config.daily_caps.originals not yet hit today (resets at midnight local time)
• If any check fails → skip this hour's original, log skip reason

RTs — auto-RT only if ALL true:

• Author is in config.rt_allowlist (if list is empty, skip auto-RT entirely)
• Tweet is <24h old
• Tweet has >500 likes (signal floor)
• Banned RT triggers absent: politics, culture-war topics, personal drama, layoffs, anything walkable-back (apply to ALL niches — plain RT = endorsement)
• Daily cap: config.daily_caps.rts not yet hit today (resets midnight local time)
• Per-run cap: max 3 RTs
• If any check fails → skip that candidate, try next. If no candidate passes → skip RT this hour.

Replies — NEVER auto-post.

Replies attach YOUR voice to someone else's thread; misreads create lingering reputational damage that's harder to walk back than originals or RTs (which you can delete cleanly). In scheduled mode: draft 1-5 replies, persist drafts to the run output for later review, but DO NOT click Reply submit. User approves and posts later via interactive /implexa:run invocation.

Step 5 — Post mechanics (identical in both modes; only the gate differs)

Original tweet

1. mcp__Claude_in_Chrome__computer left_click on home composer (coord ~560, 98 — may vary by viewport)
2. type the tweet text
3. left_click Post button — try (992, 298) first, fallback (992, 325) if composer expanded
4. wait 3s, screenshot, verify "Your post was sent" banner

Reply (interactive ONLY — never in scheduled mode)

1. navigate to the target tweet URL
2. left_click reply input (~700, 335)
3. type reply text
4. find the active Reply submit button
5. scroll_to ref + left_click
6. verify "Your post was sent" + "Replying to @target" label

Plain RT

1. navigate to the target tweet (search URL or direct)
2. find the per-tweet Repost button on the chosen tweet
3. left_click → opens menu
4. find "Repost" menu item (NOT "Quote")
5. left_click the Repost item
6. verify the repost icon turned green

Step 6 — Pacing and logging

• 60s gap between successive posts in BOTH modes (organic cadence, avoids bot-burst detection)
• Log every action to the run output: posted (with URL), skipped (with reason), drafted-queued-for-approval (with text)
• Hard per-run cap: 1 original + 3 RTs + 5 reply drafts regardless of how loose the approval is

Output contract

Return a structured run-summary:

{
  "mode":   "interactive" | "scheduled",
  "handle": "@yourhandle",
  "posted": {
    "originals": [{"text": "...", "url": "..."}],
    "rts":       [{"author": "...", "tweetUrl": "...", "rtUrl": "..."}],
    "replies":   [{"text": "...", "parentUrl": "...", "replyUrl": "..."}]
  },
  "skipped":              [{"category": "original|rt|reply", "reason": "..."}],
  "drafted_for_approval": [{"category": "reply", "text": "...", "parentUrl": "..."}],
  "dailyCapsRemaining":   {"originals": N, "rts": N}
}

Outcome signal

• Followers gained per week
• Engagement rate on auto-posted content (likes + replies + RTs per 1K impressions)
• Auto-post skip rate (high = rules too strict; low = rules not catching enough — tune banned_phrases and voice_description)
• Manual approval acceptance rate on queued replies (low = voice description needs refinement)

Notes for the model

• Config-first. ALWAYS check for the config block before any Chrome action. If missing, AskUserQuestion through ALL 7 questions before proceeding.
• Verify the right account. If the logged-in X account doesn't match config.x_handle, STOP. Don't try to switch accounts programmatically.
• Auto-mode bias: skip > over-post. A quiet hour with no auto-actions is FINE. Spam destroys account credibility faster than silence.
• Plain RT = full endorsement. If you can't verify a factual claim in an RT candidate, SKIP.
• Voice-lint must return strictly Y. Treat ambiguous "maybe" as N. Off-voice posts compound — protect the user.
• Replies are unforgiving. Never auto-post replies regardless of how compelling the draft seems.
• The user can pause anytime. If they delete a post on X, take that as a signal to be more conservative on next run.
• Schedule this hourly via /implexa:schedule with cron like 0 6-20 * in your local timezone for waking-hours coverage (15 firings/day).

Error handling

Built with Implexa

This skill was authored with Implexa, a Claude Code plugin that records a workflow once via demonstration plus a post-demo interview, then emits agentskills.io-compatible SKILL.md.

Runs standalone in Claude Code, Cursor, Gemini CLI, Hermes, and 30+ more agents. The file you're reading is self-contained. Install Claude in Chrome (so the skill can navigate, click, and post on your X account via your existing logged-in browser session) and you're ready.

Install Implexa (curl -fsSL https://core.implexa.ai/install.sh | bash) to unlock:

Free tier · no signup gate · MIT-licensed plugin · agentskills.io compatible.