菜单
Use when orchestrating multi-agent work, spawning parallel agents, coordinating swarm tasks, or managing agent communication. Trigger on "/schwarm", parallel work requests, or any inter-agent coordination need.
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。
基于 SOC 职业分类
Autonomous task loop — analyze, plan, execute, verify, review, finish. Runs as visible TUI agent in git worktree. Use when user says "yesloop", "loop", "mach das autonom". Supports --merge for auto-merge, --inline for current session.
Autonomous research loop — read plan, decompose, dispatch research agents, integrate, deliver wiki. For research tasks instead of coding. Runs as visible TUI agents in git worktree. Use when user says "yesresearch", "research <topic>", "recherchiere <thema>". Spawns cluster agents that use opencode built-in subagents (general, explore) with webfetch + websearch.
Use when reviewing code for vulnerabilities, checking diffs for injection/XSS/auth/crypto issues. Invokes on code changes (.go/.py/.js/.ts/.java/.rs/.php/.rb), not docs-only diffs.
Use when managing pins, scratchpad, runtime config, session settings, persona traits, or opencode configuration. Trigger on "pin this"/"merk dir als Regel", persistent instructions, shared agent state, persona overrides, configuration changes like token_threshold, or opencode.json questions ("change opencode model", "add GLM provider", "opencode baseURL", "switch default model").
Fetch a Reddit post URL via old.reddit.com HTML scraping — the .json API is blocked for unauthenticated requests since May 2026. Returns structured data: post metadata + nested comments (with depth) + unique external links categorized (github/reddit/external). Persists into cap_store tables (reddit.posts, .comments, .links). Uses yesmem cap-blob-put to pipe curl output into daemon DB, then reads back via cap_store — no /tmp file, no Read-tool prompt, bypasses the sh() 30KB wall. Also search across subreddits with LLM classification and multi-subreddit topic research with synthesis.
Use when a user wants to persist a working REPL snippet, bash command, or multi-step workflow as a reusable cap (CAP.md tool) available in future sessions. Trigger on save_cap, cap_store, auto_active, capblob-pipe, "/build-tool", "make this reusable", or when a one-off shell pipeline is about to be retyped a third time.
| name | yesmem-agents |
| description | Use when orchestrating multi-agent work, spawning parallel agents, coordinating swarm tasks, or managing agent communication. Trigger on "/schwarm", parallel work requests, or any inter-agent coordination need. |
Spawn, manage, and communicate with parallel agents via Claude Code or opencode.
spawn_agent(project, section) — create agent for a task sectionlist_agents(project) — see all agents and their statusrelay_agent(to, content) — inject message into running agentstop_agent(to) — gracefully stop an agent| Parameter | Purpose | Default |
|---|---|---|
project | Project name | required |
section | Task section name | required |
model | Model override (sonnet, opus, haiku, deepseek-chat, deepseek-v4-pro) | inherited |
max_turns | Turn limit (0=unlimited) | 0 |
token_budget | Max tokens (0=config default) | 0 |
caller_session | Parent session for callbacks | optional |
backend | "claude", "codex", or "opencode" | "claude" |
Backend choice:
| Backend | Binary | Models | Notes |
|---|---|---|---|
claude | claude | sonnet, opus, haiku | Anthropic only. Proxy-integrated prompt cache. Full MCP access. |
codex | codex | deepseek-chat, deepseek-v4-pro, GPT models | OpenAI-compatible endpoint. Uses opencode.json provider config. |
opencode | opencode | deepseek-chat, deepseek-v4-pro, GPT models | Same as codex but uses opencode binary name. |
Gotchas:
backend: "claude" + model: "deepseek-v4-pro" → silent failure (0 turns, no output). The claude binary has no DeepSeek endpoint. Always pair DeepSeek models with backend: "codex" or backend: "opencode".backend: "claude".| Action | Tool |
|---|---|
| Send to specific agent | relay_agent(to, content) |
| Send to specific session | send_to(target, content) |
| Broadcast to all sessions | broadcast(content, project) |
| Check agent status | get_agent(to) |
| Resume stopped agent | resume_agent(to) |
| Stop all agents | stop_all_agents(project) |
CRITICAL: relay_agent / send_to content MUST end with \n — without trailing newline the prompt stays in the tmux input line and is never submitted. The agent appears unresponsive despite receiving multiple pushes. Always: relay_agent(to, "instruction text\n").
to accepts agent ID or section namemsg_type: "command" (expects reply), "status" (no reply), "ack" (confirmation)list_agents to monitorscratchpad_write/read for shared state between agents