Promote

• "claude code mobile" --state open
• "claude code telegram" --state open
• "claude code wechat" --state open
• "claude code feishu" --state open
• "gemini cli telegram" --state open
• "gemini cli mobile" --state open
• "codex cli telegram" --state open
• "codex cli mobile" --state open

Remote/mobile control queries:

• "coding agent phone" --state open
• "coding agent mobile" --state open
• "ai agent remote control" --state open
• "ai coding assistant mobile" --state open
• "claude code remote access" --state open
• "claude code ipad" --state open

Dashboard & console queries:

• "claude code dashboard" --state open
• "claude code web ui" --state open
• "codex cli dashboard" --state open
• "agent web console" --state open
• "coding agent dashboard" --state open
• "claude code browser interface" --state open

Session & workflow queries:

• "claude code session resume" --state open
• "agent session management" --state open
• "claude code multi agent" --state open
• "switch between agents" --state open
• "claude code task queue" --state open
• "claude code context window" --state open

Desktop/browser automation queries:

• "agent browser automation" --state open
• "agent desktop automation mac" --state open
• "playwright mcp" --state open
• "browser automation agent mcp" --state open
• "appium agent automation" --state open

Notification & monitoring queries:

• "claude code notification" --state open
• "coding agent notification done" --state open
• "agent task completion alert" --state open

Headless / server / self-hosted queries:

• "coding agent headless" --state open
• "claude code server" --state open
• "ai agent self-hosted" --state open
• "claude code ssh" --state open
• "coding agent remote server" --state open

Similar-project issues (high contributor-conversion potential):

• repo:RichardAtCT/claude-code-telegram --state open
• repo:m1heng/clawdbot-feishu --state open
• repo:Johnixr/claude-code-wechat-channel --state open
• repo:Wechat-ggGitHub/wechat-claude-code --state open
• repo:qingpingwang/remote-claude-code --state open

Chinese queries:

• "claude code 手机" --state open
• "claude code 远程" --state open
• "ai agent 手机控制" --state open
• "coding agent 微信" --state open
• "coding agent 飞书" --state open
• "claude code 网页控制" --state open
• "claude code 仪表盘" --state open
• "代码agent 远程控制" --state open
• "claude code 会话管理" --state open

Example command (filters out aggregator/bot repos and our own repo):

gh search issues "claude code telegram" --state open --limit 30 --json url,title,repository | jq -r '[.[] | select(.repository.name | test("trending|news|weekly|github-daily|awesome|digest|bulletin|pikiclaw"; "i") | not) | select(.repository.nameWithOwner | test("xiaotonng/") | not)] | .[] | "\(.url) | \(.title) | \(.repository.nameWithOwner)"'

3. Filter and Select the Best Issues

Filter out any URLs that are already present in replied_issues.txt. For the remaining candidate issues, use gh issue view <url> to read the context.

Select issues that express a pain point pikiclaw explicitly solves.

Priority tiers (select higher-tier issues first):

• Tier 1 — Similar-project issues: Users already building/using tools like pikiclaw. Highest contributor-conversion potential. These people understand the problem space and are most likely to contribute code.
• Tier 2 — Specific technical pain points: Issues with clear, concrete problems that pikiclaw solves architecturally (e.g., "MCP config on Windows", "agent session dies after sleep").
• Tier 3 — General feature requests: Broader requests for mobile control, dashboards, etc.

Target pain points include:

IM & Remote Control:

• Needing remote or mobile control for local coding agents
• Wanting to use Telegram, Feishu, or WeChat to interact with agents
• Issues with official web UIs or SSH on mobile
• Wanting async notifications when long-running agent tasks complete
• Wanting to control agents from iPad or phone while away from desk

Agent Console & Dashboard:

• Wanting a web-based console to drive agent sessions from the browser (not just IM)
• Needing real-time streaming output with tool-use activity, thinking traces, and plan progress
• Wanting to view conversation history, images, and file attachments in a proper UI
• Needing to monitor multiple agent sessions, context window usage, and token consumption
• Wanting to configure agents, channels, models, and extensions from a single interface

Agent Runtime & Orchestration:

• Long-running agent tasks dying, sleeping, or disconnecting
• Needing session persistence, resume, or multi-turn conversation support
• Wanting to switch agents (Claude/Codex/Gemini) mid-workflow
• Codex needing human-in-the-loop input remotely (Human Loop)
• Wanting a task queue to send follow-up instructions while agent is busy
• Needing to interrupt/steer a running task without killing it (Steer)
• Wanting to see context window usage per turn and per session

GUI & Automation:

• Wanting an agent to control a browser (login-persistent, managed Chrome profile)
• macOS desktop automation (open apps, click, type, screenshot) via agent
• Needing MCP tools for file sharing between agent and IM

Skills & Extensibility:

• Wanting project-level reusable skills or commands for agents
• Needing compatibility with .claude/commands or custom skill systems
• Wanting to trigger predefined workflows from IM

Setup & Management:

• Wanting a web dashboard to configure agents, channels, and permissions
• Needing a watchdog / auto-restart for agent processes
• Working directory management across multiple projects

Pick 1-5 most relevant issues.

4. Draft the Replies using a Sub-Agent

Delegate the drafting process to a sub-agent for high-quality, focused output.

Prompt for the sub-agent:

Draft short, highly grounded, non-spammy GitHub issue replies to promote 'pikiclaw'.

What pikiclaw is: A Node.js CLI (npx pikiclaw@latest) that bridges local Claude Code, Codex CLI, and Gemini CLI to Telegram, Feishu, or WeChat — and also serves a full web dashboard at localhost:3939 that works as a standalone agent console. It runs on your own machine — your files, your tools, your environment — and lets you control everything from your phone or browser.

Key capabilities to draw from (use only what's relevant to the issue):

Channels & Agents:

• Three IM channels: Telegram, Feishu, WeChat — run one or all simultaneously
• Three agent backends: Claude Code, Codex CLI, Gemini CLI — switch mid-session
• Model listing and switching per agent (e.g., opus/sonnet/haiku for Claude, o4-mini/o3/gpt-5.4 for Codex)

Dashboard as Agent Console:

• Web dashboard at localhost:3939 — a full interactive agent console in the browser
• Full conversation history with tool-use activity, thinking traces, plan progress
• Real-time streaming output via WebSocket (bidirectional push, no polling)
• Image attachments, file previews, and an input composer with file upload
• Context window usage display (percentage per turn)
• Session monitoring across all agents, token usage tracking
• Channel setup, agent config, model selection, permissions, extensions — all in one UI

Runtime & Orchestration:

• Streaming preview with continuous message updates in IM
• Session switching, resume, and multi-turn conversations
• Task queue with Steer — send follow-up instructions while agent is busy; interrupt and re-prioritize the running task
• Codex Human Loop — when Codex asks a question mid-task, it surfaces in your IM as an interactive prompt
• Long-task sleep prevention, watchdog, and auto-restart
• Draft persistence across session switches in the dashboard

MCP & GUI Automation:

• MCP bridge per stream: im_list_files, im_send_file for real-time file exchange
• Browser automation — managed Chrome profile via Playwright MCP (log in once, reuse across tasks)
• macOS desktop automation — Appium Mac2 (desktop_open_app, desktop_snapshot, desktop_click, desktop_type, )

Review the drafts. Ensure they are NOT generic ads but provide real technical insight relevant to the specific issue.

5. Post the Replies

Use gh to post the drafted comments:

gh issue comment <URL> --body "<Drafted Reply>"

6. Update the Registry

After successfully posting replies, ALWAYS append the new issue URLs to the registry:

echo "<URL>" >> .pikiclaw/skills/promote/replied_issues.txt

Never skip this step — it prevents duplicate replies in future runs.

Skills & Extensibility:

• Project-level skills system (.pikiclaw/skills/) — reusable workflows triggered from IM
• Compatible with .claude/commands/*.md skill format
• Trigger skills via /skills and /sk_<name> in chat

Other:

• Working directory browsing and switching from IM
• File attachments automatically enter the session workspace
• Long text auto-splitting; images and files sent back to IM directly
• i18n: Chinese & English; light / dark theme

Reply formula (standard):

1. One sentence acknowledging the specific pain point in the issue
2. One sentence explaining the relevant architectural solution (be specific — e.g., "WebSocket-based dashboard with full tool-use trace and streaming", "task queue with Steer to interrupt and re-prioritize", "managed Chrome profile for persistent logins", "Human Loop for Codex prompts")
3. Direct call to action: npx pikiclaw@latest
4. End with project link: GitHub: https://github.com/xiaotonng/pikiclaw

Reply formula (contributor-oriented) — use for Tier 1 similar-project issues:

1. One sentence acknowledging the pain point with genuine technical empathy (you've solved the same problem)
2. One sentence sharing the specific architectural approach pikiclaw took (implementation detail, not marketing)
3. A concrete contribution invitation — link to a specific open issue that's related to their expertise (pick from: Slack support #2, multi-user #3, session history #4, or Windows bugs #5/#8)
4. Direct call to action: npx pikiclaw@latest
5. End with project link: GitHub: https://github.com/xiaotonng/pikiclaw

Rules:

• Match the language of the original issue (Chinese, Japanese, English, etc.)
• Only mention features that are directly relevant to the issue — do NOT list everything
• Sound like a developer sharing a tool, not a marketer pushing a product
• For contributor-oriented replies: share implementation insights that demonstrate technical depth, making the reader think "these people know what they're doing, I'd like to contribute"
• Keep replies under 6 sentences

Provide only the drafted replies.