Agent skill

Snipe

by xiaotonng in xiaotonng/pikiloom

Twitter/X 截流。在 coding/AI agent 工具的高流量推广帖下,自动发现→起草→自我批判→护栏→(按 posture)发布 pikiloom 回复,并记录+度量。内容话术统一来自 promo/pitch.md;可由 promo/orchestrate.md 无人值守驱动。

MITAuto-check: notes

Install Snipe

skills CLI
$ npx skills add xiaotonng/pikiloom --skill snipe -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install xiaotonng/pikiloom snipe --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/xiaotonng/pikiloom.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.pikiloom/skills/snipe .claude/skills/snipe && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
snipe
GitHub stars
295
Token cost
~1.3k tokens
SKILL.md length
265 words
Files
4 (incl. scripts)
Skills in repo
6
Repo updated
First seen
Licence
MIT

At a glance

Twitter/X 截流。在 coding/AI agent 工具的高流量推广帖下,自动发现→起草→自我批判→护栏→(按 posture)发布 pikiloom 回复,并记录+度量。内容话术统一来自 promo/pitch.md;可由 promo/orchestrate.md 无人值守驱动。

  • Works in 7 steps: 预检 → 搜索高流量推广帖(浏览器) → 筛选候选 → …
  • SKILL.md covers Twitter 渠道的两个铁律(来自实测 + 平台研究), 工作流(手动单次运行;无人值守见… and 注意事项
  • Runs JavaScript and Python scripts from its folder; calls python3, npx and claude; reaches x.com

What it does

Snipe is an agent skill from xiaotonng/pikiloom. Twitter/X 截流。在 coding/AI agent 工具的高流量推广帖下,自动发现→起草→自我批判→护栏→(按 posture)发布 pikiloom 回复,并记录+度量。内容话术统一来自 promo/pitch.md;可由 promo/orchestrate.md 无人值守驱动。

Its SKILL.md is about 1.3k tokens, which your agent loads only when the skill is triggered. The skill folder holds 4 other files, including scripts (for example `scripts/extract_tweets.js` and `scripts/push_feishu.py`).

It works with X (Twitter). The repository describes itself as: Put the world's smartest AI agents & plugins in your pocket. The licence is MIT.

Example prompts

  • “/snipe”

Requirements

  • Python 3
  • Node.js
  • Pre-approved tools (allowed-tools): Bash, Read, Write, Edit, Grep, Glob, Agent, WebFetch, mcp__pikiloom-browser__browser_navigate, mcp__pikiloom-browser__browser_take_screenshot, mcp__pikiloom-browser__browser_snapshot, mcp__pikiloom-browser__browser_press_key, mcp__pikiloom-browser__browser_click, mcp__pikiloom-browser__browser_evaluate, mcp__pikiloom-browser__browser_type, mcp__pikiloom-browser__browser_run_code_unsafe, mcp__pikiloom-browser__browser_wait_for, mcp__pikiloom-browser__browser_tabs

Workflow steps

7 steps, taken from the step headings in SKILL.md.

  1. 预检
  2. 搜索高流量推广帖(浏览器)
  3. 筛选候选
  4. 起草(子 agent,话术来自 pitch.md)
  5. 自我批判(替代人工 review)
  6. 护栏 + 按 posture 发布
  7. 报告 + 度量

What it can do on your machine

Read from SKILL.md and the folder at commit ad3bdb8. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves these tools, so the agent can use them without asking each time:

    • Bash
    • Read
    • Write
    • Edit
    • Grep
    • Glob
    • Agent
    • WebFetch
    • mcp__pikiloom-browser__browser_navigate
    • mcp__pikiloom-browser__browser_take_screenshot

    …and 8 more on the same allowed-tools line.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships 2 files in scripts/ (JavaScript and Python), which the agent can run.

    Shell commands in SKILL.md call:

    • python3
    • npx
    • claude

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    Hosts in commands or code, which the agent is likely to contact:

    • x.com

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Snipe loads about 1.3k tokens when it runs. Until then it costs about 39 tokens; SKILL.md has 265 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~39
When it runs · the whole SKILL.md, loaded when a task matches
~1.3k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check: notes

The automated check noted patterns worth knowing about, such as sudo or a known installer.

  • NotePre-approves every shell command (allowed-tools: Bash)SKILL.md
    allowed-tools: Bash, Read, Write, Edit, Grep, Glob, Agent, WebFetch, mcp__pikiloom-browser__browser_navigate, mcp__

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.

SKILL.md

The full file from xiaotonng/pikiloom at commit ad3bdb8, republished under its MIT licence (© xiaotonng). 265 words, ~1,321 tokens.

Download SKILL.mdSave it as .claude/skills/snipe/SKILL.md (or your agent's skills folder). This skill also uses 3 other files; get the full folder from GitHub.
name
snipe
description
Twitter/X 截流。在 coding/AI agent 工具的高流量推广帖下,自动发现→起草→自我批判→护栏→(按 posture)发布 pikiloom 回复,并记录+度量。内容话术统一来自 _promo/pitch.md;可由 _promo/orchestrate.md 无人值守驱动。
allowed-tools
Bash, Read, Write, Edit, Grep, Glob, Agent, WebFetch, mcp__pikiloom-browser__browser_navigate, mcp__pikiloom-browser__browser_take_screenshot, mcp__pikiloom-browser__browser_snapshot, mcp__pikiloom-browser__browser_press_key, mcp__pikiloom-browser__browser_click, mcp__pikiloom-browser__browser_evaluate, mcp__pikiloom-browser__browser_type, mcp__pikiloom-browser__browser_run_code_unsafe, mcp__pikiloom-browser__browser_wait_for, mcp__pikiloom-browser__browser_tabs
user-invocable
true
argument-hint
[keywords] or blank for default sweep

Snipe: Twitter/X 热帖截流

在 coding agent / AI 开发工具领域的高流量推广帖下回复 pikiloom,截取流量拿 star。

本 skill 只负责 Twitter 渠道的「发现 + 抽取 + 发布」机制。 所有产品话术、差异点、诚实边界、 语气/语言规则、回复骨架与反模式,统一来自 ../_promo/pitch.md —— 不在本文件 重复,也不要在本文件改话术(改话术只改 pitch.md)。去重 / 护栏 / 度量 / 发布姿态由共享核心负责:

关注点位置
产品话术 SSOT_promo/pitch.md(§9 表里 twitter 列是本渠道的机制差异)
去重记录(写时强制)_promo/registry.py(channel = twitter)
发布护栏(频次/集中度/变体/熔断)_promo/guard.py
飞书推送_promo/push_feishu.py
无人值守编排 + posture_promo/orchestrate.md
旋钮(caps / posture / kill switch)_promo/config.json → channels.twitter

运行根目录:cd /Users/admin/Desktop/project/pikiloom。脚本前缀:.pikiloom/skills/_promo/。

Twitter 渠道的两个铁律(来自实测 + 平台研究)

  1. 链接绝不放主回复正文。 X 对带外链的帖/回复显著降权。主回复 = 一句差异化 + npx pikiloom@latest (平台内原生文本,不触发降权);GitHub 链接放到对自己回复的自回复(self-reply)里。
  2. 绝不复制同一段文案 / 同一条外链刷多帖 —— 这是 X 明文的封号信号。每条独立起草(guard.py 的 变体检查会拦截过相似的草稿)。

工作流(手动单次运行;无人值守见 _promo/orchestrate.md)

Step 1: 预检
bash
cd /Users/admin/Desktop/project/pikiloom
python3 .pikiloom/skills/_promo/guard.py caps      # 看 twitter 今日剩余配额;为 0 则今天不跑
Step 2: 搜索高流量推广帖(浏览器)

传入关键词参数 → 直接搜该词。否则用场景关键词动态发现(搜 3–5 组即可,不要死记产品名):

coding agent tool / AI coding assistant / claude code tool / remote coding agent
AI agent dashboard / vibe coding 工具 / coding agent mobile / AI dev tool launch
claude code api cost / claude max subscription   (billing 段 = 最高转化,命中用 pitch 角度 0)

对每个关键词:

  1. browser_navigate → https://x.com/search?q={keyword}&src=typed_query&f=top
  2. browser_wait_for 确认推文加载
  3. 读取 scripts/extract_tweets.js,用 browser_evaluate 注入执行,得到 JSON
  4. 结果不够就按 End 滚动一次再抽
  5. 合并所有关键词结果,按 text[:80]+url 去重。无效关键词直接跳过。
Step 3: 筛选候选

硬门槛(全部满足): views > 5000;最近 48h 内;与 pikiloom 有功能交集;正在推广具体产品/工具 (有 has_product_signal);不在 registry:

bash
python3 .pikiloom/skills/_promo/registry.py seen twitter "<tweet_url>" && echo SKIP   # 已记录则跳过

排除: 自己(@sthnavy)的帖;纯新闻/讨论/提问;已是 pikiloom 用户;政治/争议。

优先级: ① 功能高度重叠的产品推广帖(截流最佳)② 同赛道不同切入点 ③ 泛 AI 工具帖。选 Top 3–5。

Step 4: 起草(子 agent,话术来自 pitch.md)

委托子 agent 起草,内容契约 = ../_promo/pitch.md:orienter(§1) + 挑一个最锋利的 差异点(§2,billing 命中优先) + 诚实边界(§3) + 语言跟随原帖(§6)。Twitter 形态(pitch §9):

{一句差异化,直击原帖产品短板或 pikiloom 独特优势 —— 必须 ground 在原帖具体内容上}
npx pikiloom@latest

(正文到此为止,不含链接。1–2 句最佳,绝不超 3 句。)

自回复(second tweet)内容:

Open-source, runs local: https://github.com/xiaotonng/pikiloom

每条起草后立即记录(存草稿文本供变体检查):

bash
python3 .pikiloom/skills/_promo/registry.py add --channel twitter --url "<tweet_url>" \
  --status drafted --type launch --lang <en|zh> --audience <views> --title "<摘要>" \
  --draft-file /tmp/snipe_draft_<id>.txt
Step 5: 自我批判(替代人工 review)

对每条草稿按 pitch.md §10 反模式自检(营销腔 / 贬低原帖 / 堆功能 / 超长 / 未 ground 原帖 / 越诚实边界)。 FAIL → 让子 agent 修一次;再 FAIL → registry.py update --channel twitter --url <u> --status skipped 丢弃。

Step 6: 护栏 + 按 posture 发布
bash
python3 .pikiloom/skills/_promo/guard.py check --channel twitter --url "<tweet_url>" \
  --draft-file /tmp/snipe_draft_<id>.txt        # exit 3 = deny;deny 则 update --status skipped 跳过
POSTURE=$(python3 -c "import json;print(json.load(open('.pikiloom/skills/_promo/config.json'))['posture'])")
  • shadow:不发,推预览卡片即可。
  • batch:update --status approved + 推 veto 卡片(见 orchestrate.md Phase 4)。
  • auto:直接发(浏览器,见下)。

浏览器发推机制(main reply + self-reply):

  1. browser_navigate 到 tweet_url。
  2. 点回复框,browser_type 主回复正文(差异化 + npx pikiloom@latest,无链接),提交。
  3. 提交后定位到刚发出的自己的回复,点它的「回复」,browser_type 自回复(GitHub 链接行),提交。 X 会对链接自动生成卡片。
  4. 发送前后各截图核对:发前确认正文不含 github.com;发后读 DOM 确认主回复无外链、自回复有链接, 且文案只出现一次(防重复粘贴)。异常则不再操作,记 failed 让人工介入。
  5. 成功后记录:
bash
python3 .pikiloom/skills/_promo/registry.py mark-posted --channel twitter \
  --url "<tweet_url>" --post-url "<我们主回复的 url>"
Step 7: 报告 + 度量

把候选 + 草稿整理为 /tmp/snipe_report.md(每条含:作者/链接/views/likes/推广产品/交集/差异点/草稿),推飞书:

bash
python3 .pikiloom/skills/_promo/push_feishu.py --report-file /tmp/snipe_report.md --title "🎯 Snipe 候选"
python3 .pikiloom/skills/_promo/measure.py report      # t.co referrer / star 关联,附在卡片后

输出 OK: 成功 / SKIP: 缺飞书凭证 / ERROR: 失败(兜底把报告贴进对话)。

注意事项

  • 质量 > 数量:每次 3–5 条候选即可;guard.py 会按 config.json 的 daily_cap / per_author / 变体兜底。
  • 不要固定搜某个产品名 —— 用场景关键词动态发现,这个领域每天有新工具。
  • kill_switch / abort.txt 随时生效;账号尽量用 Premium(链接容忍度 + 触达约 10×)。
  • 历史案例规律:原帖与 pikiloom 功能越接近,回复转化越高。

© xiaotonng, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 3 other files (scripts) in .pikiloom/skills/snipe of xiaotonng/pikiloom.

  • SKILL.md
  • scripts/extract_tweets.js
  • scripts/push_feishu.py
  • sniped_posts.txt

Open the folder on GitHubat commit ad3bdb8

Compare with similar skills

Snipe next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Snipe compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Snipe this skillxiaotonng/pikiloom295—~1.3kAutomated safety check: NotesMIT
Agent ReachPanniantong/Agent-Reach93k1 repos~1.4kAutomated safety check: PassMIT
Socialcoreyhaines31/marketingskills54k4 repos~4.5kAutomated safety check: PassMIT
Banner Design Systemnextlevelbuilder/ui-ux-pro-max-skill134k1 repos~1.8kAutomated safety check: PassMIT
Social Contentfreekmurze/dotfiles1k23 repos~2.1kAutomated safety check: PassNone
X to Markdown ConverterJimLiu/baoyu-skills26k4 repos~1.8kAutomated safety check: WarnMIT

Similar skills

  • Agent Reach

    Panniantong/Agent-Reach

    Routes web research and platform lookups across 16 sites, including Twitter, Reddit, YouTube, Bilibili, Xiaohongshu and GitHub, through one command-line tool.

    93k GitHub starsUsed in 1 repo~1.4k tokens
    Productivity & AutomationAuto-check passed
  • Social

    coreyhaines31/marketingskills

    When the user wants help creating, scheduling, or optimizing social media content for LinkedIn, Twitter/X, Instagram, TikTok, or Facebook, or wants to do social listening and engagement triage.

    54k GitHub starsUsed in 4 repos~4.5k tokens
    Writing & ContentAuto-check passed
  • Banner Design System

    nextlevelbuilder/ui-ux-pro-max-skill

    Walks through designing a banner for social media, ads, a website hero or print, from gathering requirements to building 2 or 3 art-direction options in HTML and CSS.

    134k GitHub starsUsed in 1 repo~1.8k tokens
    Frontend & DesignAuto-check passed
  • Social Content

    freekmurze/dotfiles

    When the user wants help creating, scheduling, or optimizing social media content for LinkedIn, Twitter/X, Instagram, TikTok, Facebook, or other platforms.

    1k GitHub starsUsed in 23 repos~2.1k tokens
    Writing & ContentAuto-check passed
  • X to Markdown Converter

    JimLiu/baoyu-skills

    Saves tweets, threads and X Articles as Markdown files with YAML front matter, using an unofficial API that asks for your consent first.

    26k GitHub starsUsed in 4 repos~1.8k tokens
    Knowledge ManagementAuto-check: warnings
  • URL to Markdown Fetcher

    JimLiu/baoyu-skills

    Fetches a web page, X post, YouTube transcript or Hacker News thread through a Chrome-driven CLI and saves it as clean markdown.

    26k GitHub starsUsed in 2 repos~2.1k tokens
    Knowledge ManagementAuto-check passed

More from xiaotonng/pikiloom

  • Install

    xiaotonng/pikiloom

    This skill should be used when the user asks to "install pikiclaw", "build and install", "deploy locally", "update local binary", "release", or "publish".

    295 GitHub stars~464 tokensUpdated 2 days ago
    Auto-check passed
  • Reddit Snipe

    xiaotonng/pikiloom

    Reddit 截流。在 coding agent / Claude Code / AI dev tool 的高互动帖下,以 pikiloom builder 身份自动发现→起草→自我批判→护栏→(按 posture)发布英文 peer 评论,并记录+度量。English-first。话术统一来自 promo/pitch.md;可由 promo/orchestrate.md 无人值守驱动。

    295 GitHub stars~3.3k tokensUpdated 2 days ago
    Auto-check: notes
  • Promote

    xiaotonng/pikiloom

    GitHub 截流。在同领域项目(local coding agent ↔ IM / mobile / remote console)的 feature-request issue 下,自动发现→起草→自我批判→护栏→(按 posture)发布 pikiloom 回复,并记录+度量。话术统一来自 promo/pitch.md;可由 promo/orchestrate.md 无人值守驱动。

    295 GitHub stars~2k tokensUpdated 2 days ago
    Auto-check passed
  • Auto Promote

    xiaotonng/pikiloom

    无人值守推广总入口。按 promo/orchestrate.md 跨所有启用渠道(GitHub / Twitter / Reddit)跑一轮:发现→去重→起草→自我批判→护栏→按 posture 发布→记录→度量。供每日 cron(schedule / loop)调用,也可手动运行。传 "post-approved" 只执行 batch 待发队列。

    295 GitHub stars~629 tokensUpdated 2 days ago
    Auto-check: notes
  • Dev

    xiaotonng/pikiloom

    This skill should be used when the user asks to start, restart, keep alive, or inspect the local pikiclaw development service, including npm run dev, local debug bot startup, dashboard verification…

    295 GitHub stars~174 tokensUpdated 2 days ago
    Auto-check passed

Works with

Questions about Snipe

What does Snipe do?

Twitter/X 截流。在 coding/AI agent 工具的高流量推广帖下,自动发现→起草→自我批判→护栏→(按 posture)发布 pikiloom 回复,并记录+度量。内容话术统一来自 promo/pitch.md;可由 promo/orchestrate.md 无人值守驱动。. Snipe is an agent skill from xiaotonng/pikiloom.

How do I install Snipe in Claude Code?

Run `npx skills add xiaotonng/pikiloom --skill snipe -a claude-code`. Or copy the skill folder (.pikiloom/skills/snipe in xiaotonng/pikiloom) into .claude/skills/snipe in your project. Claude Code loads it when a task matches its description.

How do I install Snipe in Codex?

Run `npx skills add xiaotonng/pikiloom --skill snipe -a codex`. Or copy the skill folder (.pikiloom/skills/snipe in xiaotonng/pikiloom) into .agents/skills/snipe in your project. Codex loads it when a task matches its description.

Can I use Snipe in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add xiaotonng/pikiloom --skill snipe -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/snipe, .gemini/skills/snipe, .github/skills/snipe and .opencode/skills/snipe in your project.

What does Snipe need to run?

Going by SKILL.md and its folder, Snipe needs JavaScript and Python for the scripts in its folder and the command-line tools its instructions call (python3, npx and claude). Our summary lists: Python 3; Node.js. Its frontmatter pre-approves these tools: Bash, Read, Write, Edit, Grep, Glob, Agent, WebFetch, mcp__pikiloom-browser__browser_navigate, mcp__pikiloom-browser__browser_take_screenshot, mcp__pikiloom-browser__browser_snapshot, mcp__pikiloom-browser__browser_press_key, mcp__pikiloom-browser__browser_click, mcp__pikiloom-browser__browser_evaluate, mcp__pikiloom-browser__browser_type, mcp__pikiloom-browser__browser_run_code_unsafe, mcp__pikiloom-browser__browser_wait_for, mcp__pikiloom-browser__browser_tabs.

Does Snipe access the network?

SKILL.md names 1 domain. In commands or code: x.com; the agent is likely to contact it when it follows the instructions. This is read from the text; nothing was executed.

Is Snipe safe to install?

Our automated static check of SKILL.md found notes only (pre-approves every shell command (allowed-tools: bash)), nothing it rates as a warning. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.

What licence does Snipe use?

Snipe is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Snipe use?

About 1.3k tokens (SKILL.md is roughly 5.3k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Snipe?

Skills that share tags, products or a category with Snipe: Agent Reach (Panniantong/Agent-Reach, 93k stars), Social (coreyhaines31/marketingskills, 54k stars), Banner Design System (nextlevelbuilder/ui-ux-pro-max-skill, 134k stars) and Social Content (freekmurze/dotfiles, 1k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Snipe?

xiaotonng (a GitHub user) maintains it in xiaotonng/pikiloom, which has 295 GitHub stars. The repository holds 6 skills in this directory. The repository was last updated on October 5, 2026.

Source: xiaotonng/pikiloom on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.