菜单
Manage Claude Code --channels sessions (start/stop/status/health) via channel-manager.cjs and terminal-tracker.cjs.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/oimiragieo/agent-studio --skill channel-management命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Agent frontmatter enhancements — disallowedTools, mcpServers scoping, fork_eligible field
Context management — pre-compact persistence, threshold alignment, microcompact/circuit breaker detection
Hook enhancements — updatedInput for bash safety prefixes, suppressOutput for verbose blocks, denial-based routing feedback
正在显示 SKILL.md
基于 SOC 职业分类
| name | channel-management |
| description | Manage Claude Code --channels sessions (start/stop/status/health) via channel-manager.cjs and terminal-tracker.cjs. |
| version | 1.0.0 |
| model | sonnet |
| invoked_by | both |
| user_invocable | true |
| tools | ["Read","Bash","TaskUpdate"] |
| agents | ["developer","devops","heartbeat-orchestrator"] |
| category | External Integrations |
| tags | ["channels","telegram","session","lifecycle","pid","terminal","health"] |
| best_practices | ["Always call isChannelRunning() before startChannel() to avoid duplicate sessions","Track PID via terminal-tracker after every start","Run health check within 10 seconds of start to confirm session is live","Use CHANNEL_AUTO_START=true in .env only for supervised production environments"] |
| error_handling | strict |
| streaming | not-applicable |
| source | builtin |
| trust_score | 100 |
| provenance_sha | 76303e7de92f80e4 |
This skill manages the full lifecycle of a Claude Code --channels session (Telegram bot, Discord bot, or any plugin-based channel integration). It wraps the CLI tools at .claude/tools/cli/channel-manager.cjs and .claude/tools/cli/terminal-tracker.cjs and provides a repeatable, observable workflow for:
Skill({ skill: 'channel-management' });
Invoke when:
CHANNEL_AUTO_START=true is set and session startup must be verified on agent bootNEVER start a channel session without checking isChannelRunning() first.
NEVER assume a tracked PID is alive without calling isPidAlive(pid).
NEVER restart after a crash without calling killOrphaned() to prune stale entries.
Before executing any lifecycle action, read memory to check for known issues:
node -e "const fs=require('fs');const p='.claude/context/memory/learnings.md';if(fs.existsSync(p))console.log(fs.readFileSync(p,'utf8').split(/\r?\n/).slice(0,80).join('\n'));"
Check .claude/context/memory/issues.md for any known channel startup problems:
node -e "const fs=require('fs');const p='.claude/context/memory/issues.md';if(fs.existsSync(p))console.log(fs.readFileSync(p,'utf8').split(/\r?\n/).slice(0,60).join('\n'));"
Command:
node -e "
const { isChannelRunning, getChannelPid } = require('./.claude/tools/cli/channel-manager.cjs');
const running = isChannelRunning();
const pid = getChannelPid();
console.log(JSON.stringify({ running, pid }));
"
Expected output: {"running":false,"pid":null} (not running) or {"running":true,"pid":12345} (running)
Verify: Exit code 0 and valid JSON printed to stdout.
Decision gate:
running: true → Skip Step 2 (already running), proceed to Step 4 (health check)running: false → Continue to Step 2Pre-condition: Confirm TELEGRAM_BOT_TOKEN is set:
node -e "
require('dotenv').config({ path: '.env' });
const token = process.env.TELEGRAM_BOT_TOKEN;
console.log(token ? 'TOKEN_PRESENT' : 'TOKEN_MISSING');
"
If TOKEN_MISSING: Log to issues.md and abort — channel cannot start without credentials.
Start command:
node -e "
const { startChannel } = require('./.claude/tools/cli/channel-manager.cjs');
const result = startChannel();
console.log(JSON.stringify(result));
"
Expected output: {"ok":true,"pid":12345,"reason":"started"} on success, or {"ok":false,"pid":null,"reason":"TELEGRAM_BOT_TOKEN not set — skipped"} when token missing.
Verify: ok: true in JSON output. Exit code 0.
After a successful start, confirm the PID is registered in the tracker:
node -e "
const { listTracked } = require('./.claude/tools/cli/terminal-tracker.cjs');
const sessions = listTracked();
const channel = sessions.filter(s => s.purpose === 'channel-session');
console.log(JSON.stringify(channel, null, 2));
"
Expected output: Array containing at least one entry with "status": "active" and a numeric "pid".
Verify: Entry exists with status: "active".
Confirm the session is alive after startup (or as a standalone health probe):
node -e "
const { isChannelRunning, getChannelPid } = require('./.claude/tools/cli/channel-manager.cjs');
const { listTracked } = require('./.claude/tools/cli/terminal-tracker.cjs');
const running = isChannelRunning();
const pid = getChannelPid();
const tracked = listTracked().find(s => s.purpose === 'channel-session' && s.status === 'active');
console.log(JSON.stringify({
health: running ? 'OK' : 'DEGRADED',
pid,
trackerEntry: tracked || null
}));
"
Expected output: {"health":"OK","pid":12345,"trackerEntry":{...}} when healthy.
Verify: health: "OK" in JSON. If health: "DEGRADED", proceed to Step 5 (restart).
Use when explicitly stopping the channel or before a clean restart:
node -e "
const { stopChannel } = require('./.claude/tools/cli/channel-manager.cjs');
const result = stopChannel();
console.log(JSON.stringify(result));
"
Expected output: {"ok":true,"reason":"stopped"} or {"ok":true,"reason":"not-running"}.
Verify: ok: true.
After a crash or unexpected shutdown, clean up stale PID entries:
node -e "
const { killOrphaned } = require('./.claude/tools/cli/terminal-tracker.cjs');
killOrphaned();
console.log('Orphan prune complete');
"
Expected output: Orphan prune complete
Verify: Exit code 0. Re-run Step 1 to confirm running: false after prune.
After any lifecycle action, append a brief note:
# Append to learnings.md (adjust message as appropriate)
node -e "
const fs = require('fs');
const ts = new Date().toISOString();
const entry = '\n## ' + ts + ' — channel-management\n- Channel session lifecycle action completed. PID: <PID>. Status: <STATUS>.\n';
fs.appendFileSync('.claude/context/memory/learnings.md', entry, 'utf8');
console.log('Memory updated');
"
When CHANNEL_AUTO_START=true is set in .env, agents should execute this skill at startup:
// In agent Step 0 or heartbeat-orchestrator boot sequence
if (process.env.CHANNEL_AUTO_START === 'true') {
Skill({ skill: 'channel-management' });
}
The skill is idempotent — calling it when the session is already running is a safe no-op.
| Variable | Required | Default | Purpose |
|---|---|---|---|
TELEGRAM_BOT_TOKEN | YES | — | Telegram bot credentials |
CHANNEL_AUTO_START | NO | false | Auto-start on agent boot |
CHANNEL_PLUGINS | NO | plugin:telegram@claude-plugins-official | Space-separated plugin list |
CHANNEL_PERMISSIONS | NO | — | Extra claude flags (e.g. --dangerously-skip-permissions) |
Input validated against schemas/input.schema.json before execution.
Output contract defined in schemas/output.schema.json.
Pre-execution hook validates that the action field is a known enum value and that a TELEGRAM_BOT_TOKEN guard is in place for start actions.
startChannel() in a tight loop without checking isChannelRunning() first — each call spawns a new Windows Terminal tabisPidAlive()CHANNEL_AUTO_START=true without monitoring — silent crashes will not be detected without a health-check cronshell: true in any child process spawned from this skill — use array arguments with shell: false (already enforced in channel-manager.cjs).claude/tools/cli/channel-manager.cjs — Core lifecycle API (startChannel, stopChannel, isChannelRunning, getChannelPid).claude/tools/cli/terminal-tracker.cjs — PID registry (registerSpawn, listTracked, killOrphaned, cleanup).claude/context/runtime/terminal-pids.json — Live PID tracker state fileBefore starting: Read .claude/context/memory/learnings.md and .claude/context/memory/issues.md for known channel startup problems.
After completing: Append outcome to .claude/context/memory/learnings.md with timestamp and action taken.
ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.