• Reply under promotional posts of same-field products (the higher the traffic, the better)
• One-sentence differentiation +

  npx pikiclaw@latest

  + GitHub link
• Core differentiators of pikiclaw (in order of importance): multi-agent session parallel management (control Claude/Codex/Gemini simultaneously in the dashboard), smooth local experience (launch with one line of npx, zero configuration), practical skill/MCP plugin ecosystem (community-built skills and MCPs ready to use), IM integration (take over the same session anytime via Telegram/Feishu/WeChat), fully open-source + local runtime
• IM is not the headline selling point, but part of the "experience" — the headline is always multi-session management + smoothness + plugin ecosystem

• Reply under a remote agent management tool's promotional post → 7,792 views, 31 likes (original post 140,000 views)
• Reply under a macOS agent management tool's promotional post → 1,258 views (original post 54,000 views)
• Rule: The closer the original post's product functions are to pikiclaw, the higher the reply conversion rate

coding agent tool
AI coding assistant 推荐
claude code 工具
coding agent 开源
remote coding agent
AI agent dashboard
vibe coding 工具
coding agent mobile
AI dev tool launch

1. Navigate to

   https://x.com/search?q={keyword}&src=typed_query&f=top

2. Wait for the page to load (confirm tweet content is visible)
3. Read the content of

   .pikiclaw/skills/snipe/scripts/extract_tweets.js

4. Inject and execute the JS via

   browser_evaluate

   to get the returned JSON string
5. Parse the JSON to get the tweet array
6. If there are not enough results, scroll once by pressing the End key and execute the JS again to extract
7. Merge results from all keywords, deduplicate by

   text[:80] + url

• The post mentions specific product names, feature introductions, or installation commands
• has_product_signal is true (contains GitHub links, npm/pip installation commands, or product domains)
• External links point to product official websites or GitHub
• Tone is introductory/recommendatory/announcement-based (not pure discussion or question)

• views > 5,000 (too small a traffic pool is not worth it)
• Published within the last 48 hours
• The promoted product/tool has functional overlap with pikiclaw (coding agent management, remote control, multi-agent switching, IM integration, etc.)
• Not in

  sniped_posts.txt

1. Promotional posts of products with highly overlapping functions with pikiclaw (best traffic capture effect)
2. Promotional posts of products in the same track but with different entry points (can play up differentiation)
3. General AI tool promotional posts (exposure exists but conversion is low)

• Own posts (@sthnavy)
• Pure news/information/discussion posts (no specific product promotion)
• Posts from existing pikiclaw users/retweeters
• Political/controversial/irrelevant topics

{One-sentence differentiation that directly targets the original product's weakness or pikiclaw's unique advantage}
npx pikiclaw@latest

GitHub: https://github.com/xiaotonng/pikiclaw

• The other party only manages single agent / single session → "Manage Claude/Codex/Gemini multi-sessions in parallel in the dashboard, switch anytime"
• The other party has rough experience / complex configuration → "Launch with one line of npx, dashboard ready to use, zero configuration"
• The other party has a closed ecosystem / no plugins → "Open skill/MCP plugin system, practical community-built tools ready to use"
• The other party is closed-source / SaaS → "Fully open-source, all sessions and code stay local"
• The other party only has CLI → "Built-in web dashboard, full control of multi-sessions in the browser"
• The other party can only be used at the desk → "Direct connection via Telegram/Feishu/WeChat, take over the same session anytime on mobile"
• The other party only supports English / single platform → "Bilingual (Chinese/English), macOS desktop automation + Playwright browser control"

• Follow the language of the original post (use Chinese for Chinese posts, English for English posts)
• Keep it minimal, 1-2 sentences are best, never more than 3
• Must include

  npx pikiclaw@latest

  and the GitHub link
• Use a builder/developer tone, avoid promotional words like "recommend" or "endorse"
• Do not belittle the original post's product, only emphasize pikiclaw's differences

• Author: @{handle} ({name})
• Link: {tweet_url}
• Data: {views} views / {likes} likes / {retweets} RT
• Promoted Product: {product_name} — {One-sentence description of what this product does}
• Overlap with pikiclaw: {Functional overlap points}
• Differentiated Advantage of pikiclaw: {The sharpest differentiator}

{draft}

1. Prioritize replying to Candidate 1 (highest traffic / closest functions), then proceed in order
2. Paste the GitHub link when posting the reply on Twitter, Twitter will automatically generate a card
3. After posting, append the tweet URL to

   .pikiclaw/skills/snipe/sniped_posts.txt

undefined

1. Write the Markdown report generated in Step 5 to

   /tmp/snipe_report.md

2. Execute the Feishu push script:

cd /Users/admin/Desktop/project/pikiclaw && python3 .pikiclaw/skills/snipe/scripts/push_feishu.py --report-file /tmp/snipe_report.md

1. Check the script output:
   • Starts with

     OK:

     → Success, both the report document URL and notification have been sent
   • Starts with

     PARTIAL:

     → Document created successfully but notification not sent (missing FEISHU_RECEIVE_ID)
   • Starts with

     ERROR:

     → Failed, directly display the report content in the conversation as a fallback

• Never auto-post to Twitter — All reply drafts are only pushed to Feishu for manual review
• Quality > Quantity — 3-5 candidates per round are sufficient
• Do not fixate on searching for a specific product name — Use scenario keywords to discover dynamically; new tools emerge in this field every day
• Feishu credentials are read from the project root directory

  .env

  (requires

  FEISHU_APP_ID

  ,

  FEISHU_APP_SECRET

  ,

  FEISHU_CHAT_ID

  )