Menü
Delegate coding to Claude Code CLI (features, PRs).
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
Set up and use 1Password CLI (op). Use when installing the CLI, enabling desktop app integration, signing in, and reading/injecting secrets for commands.
Give the agent its own dedicated email inbox via AgentMail. Send, receive, and manage email autonomously using agent-owned email addresses (e.g. hermes-agent@agentmail.to).
Airtable REST API via curl. Records CRUD, filters, upserts.
Professional anime/2D art style generation skill. Covers 14 sub-styles (modern Japanese anime/moe, retro cel-shading, shonen, shojo, Ghibli, Makoto Shinkai, Chinese xianxia/ink wash, modern Chinese anime, Chinese 3D fantasy, Korean webtoon, Korean impasto, Western cartoon, chibi/moe, 2D cyberpunk) + 5 anti-failure iron laws + cross-style shared rules (character lock / facial proportion spec / stroke consistency / universal negative). Core capabilities: precise style targeting, consistent character identity, cross-style conversion. Trigger: "anime", "2D art", "manga", "illustration", "webtoon", "ghibli", "shinkai", "ufotable", "cel-shading", "impasto", "chibi", "moe", "catgirl", "xianxia", "ink wash", "hanfu character", "cyberpunk anime", "anime character/avatar/style". NOT for: photorealistic (use image agent default) / static posters (use poster-design)
Specialized in anime/2D/character stylization for image generation and conversion. Covers Japanese, Chinese, Korean, and Western art style families. Uses provenance analysis to trace reference images' style DNA, performs a 10-dimension analysis → 3-dimension collapse to precisely lock the style's essence, then matches the optimal tool and prompt approach for generation. Trigger on: "anime-ify", "2D style", "convert to anime", "cel-shading", "ghibli style", "Korean watercolor", "fantasy 3D", "chibi", "Japanese anime style", "style conversion", "manga style", "character illustration", "anime style", "webtoon style", "daily gallery", "daily image series", "daily image in same style", or any request involving converting content into a specific anime/2D art style. Key distinction: User requests generation or conversion to a specific anime/2D art style. Do NOT trigger for: photorealistic photography style, pure logo design, general image editing (crop/background removal etc.).
Operate the Antigravity CLI (agy): plugins, auth, sandbox.
| name | claude-code |
| description | Delegate coding to Claude Code CLI (features, PRs). |
| version | 2.2.0 |
| author | Hermes Agent + Teknium |
| license | MIT |
| platforms | ["linux","macos","windows"] |
| metadata | {"hermes":{"tags":["Coding-Agent","Claude","Anthropic","Code-Review","Refactoring","PTY","Automation"],"related_skills":["codex","hermes-agent","opencode"],"user_owner":"the user ([redacted])","billing_mode":"Pro-Subscription via OAuth (claude.ai) — NOT API-Token-Plan","role_in_swarm":"Worker-Heavy (Backup)","primary_models":{"sonnet-4-5":"default for Refactor/Code-Review/Big-Edit","haiku-4-5":"lightweight tasks, mini-edits, lookups","opus-4":"architecture reviews >10k LOC (use sparingly, on the user's request only)"}}} |
| lane | koenigin |
| reasoning_effort | xhigh |
Delegate coding tasks to Claude Code (Anthropic's autonomous coding agent CLI) via the Hermes terminal. Claude Code v2.x can read files, write code, run shell commands, spawn subagents, and manage git workflows autonomously.
claude auth status → subscriptionType: "pro", authMethod: "claude.ai")[redacted] (Org-ID 4b24e94b-9e73-4abb-b3d0-0a8f2280ec01)ANTHROPIC_API_KEY set — do NOT reference it; Pro-OAuth is the only active auth path~/.local/bin/claude (v2.1.201)Delegiere NUR wenn MiniMax M3/StepFun NICHT reicht. Konkret:
| ✅ Delegieren an Claude | ❌ NICHT delegieren |
|---|---|
| Big Refactor (>1 file, >200 LOC) | Mini-Edit (1-2 lines) |
| Architektur-Review / Design-Feedback | Bug-Search in kleinem Code-Snippet |
| Multi-File Refactor / Test-Writing | Doku-Recherche / Lookups |
| Komplexe Code-Analyse (race conditions, security) | Trivial-Syntax-Korrektur |
Hard rules:
--max-turns setzen (Default 10, Max 30 für Big-Refactor)--max-budget-usd setzen (Default 0.50, Max 2.00 für Big-Refactor)claude -p ...) — keine unnötigen Multi-Turn-Sessions# Default: Refactor, Code-Review, Big-Edit
claude -p "<task>" --model claude-sonnet-4-5 --max-turns 10 --max-budget-usd 0.50
# Lightweight: Mini-Edits, Bug-Fixes, Lookups
claude -p "<task>" --model claude-haiku-4-5 --max-turns 5 --max-budget-usd 0.20
# Heavy: nur auf the user's Anweisung
claude -p "<task>" --model claude-opus-4 --max-turns 30 --max-budget-usd 2.00
For other autonomous coding agents, see: codex, opencode, hermes-agent skills.
npm install -g @anthropic-ai/claude-codeclaude once to log in (browser OAuth for Pro/Max, or set ANTHROPIC_API_KEY)claude auth login --console for API key billingclaude auth login --sso for Enterpriseclaude auth status (JSON) or claude auth status --text (human-readable)claude doctor — checks auto-updater and installation healthclaude --version (requires v2.x+)claude update or claude upgradeterminal(command="claude -p 'Add error handling to all API calls in src/' --allowedTools 'Read,Edit' --max-turns 10", workdir="/path/to/project", timeout=120)
Print mode runs a one-shot task, returns the result, and exits. No PTY, no dialogs, structured output. Use this for 80% of delegated work.
# Start a tmux session
terminal(command="tmux new-session -d -s claude-work -x 140 -y 40")
# Launch Claude Code
terminal(command="tmux send-keys -t claude-work 'cd /path/to/project && claude' Enter")
# Wait for startup, then send your task (~3-5s)
terminal(command="sleep 5 && tmux send-keys -t claude-work 'Refactor the auth module to use JWT tokens' Enter")
# Monitor progress
terminal(command="sleep 15 && tmux capture-pane -t claude-work -p -S -50")
# Exit when done
terminal(command="tmux send-keys -t claude-work '/exit' Enter")
Interactive mode requires tmux orchestration — it gives you capture-pane for monitoring and send-keys for input. terminal(pty=true) works but tmux is preferred.
| Scenario | Mode | Why |
|---|---|---|
| Fix a bug, add a feature, refactor | Print (-p) | No dialogs, structured JSON output, scriptable |
| CI/CD automation, batch tasks | Print (-p) | Non-interactive, no TTY needed |
| Structured data extraction | Print (-p) | Use --json-schema for typed output |
| Piped input analysis | Print (-p) | cat file | claude -p "..." |
| Multi-turn review/fix cycle | Interactive (tmux) | Follow-up prompts, slash commands |
| Human-in-the-loop decisions | Interactive (tmux) | Need conversational flow |
Using /compact, /review, /model | Interactive (tmux) | Slash commands only work here |
| Parallel multi-task work | Interactive (tmux) | One tmux session per instance |
-p)--output-format json — includes session_id, num_turns, total_cost_usd, subtype--output-format stream-json --verbose for real-time token streaming--input-format stream-json --replay-user-messages--json-schema '{...}' — Claude validates before returning--resume <id> or --continue (same directory)--bare for CI (skips hooks, plugins, MCP, CLAUDE.md, OAuth)--fallback-model haiku for graceful overload handling--max-budget-usd and --max-turns (print mode only)→ Details, code examples, JSON schemas: references/print-mode.md
/compact, /review, /security-review, /plan, /model, /effort, /clear, /cost, /context, /resume, /rewind, /voice, and more.claude/commands/*.md (project) or ~/.claude/commands/*.md (user).claude/skills/Shift+Tab (permission modes), Alt+P (model), Ctrl+O (transcript), Esc Esc (rewind)/context (70% threshold for /compact)→ Dialog handling, slash commands, shortcuts: references/interactive-pty.md
.claude/settings.local.json → .claude/settings.json → ~/.claude/settings.json.claude/agents/ (project) or ~/.claude/agents/ (user)-w <name> --tmux for safe parallel work--from-pr <number> or piped git diff→ Settings, hooks, MCP, agents: references/configuration.md
| Category | Key Flags |
|---|---|
| Session | -p, -c, -r, --fork-session, --session-id, --add-dir, -w, --tmux, --from-pr |
| Model | --model, --effort, --max-turns, --max-budget-usd, --fallback-model |
| Permissions | --dangerously-skip-permissions, --permission-mode, --allowedTools, --disallowedTools |
| Output | --output-format, --json-schema, --verbose, --include-partial-messages |
| Context | --append-system-prompt, --bare, --agents, --mcp-config, --strict-mcp-config |
| Debug | -d [filter], --debug-file |
→ Complete flag reference: references/cli-flags.md
terminal(command="cd /path/to/repo && git diff main...feature-branch | claude -p 'Review this diff for bugs, security issues, and style problems.' --max-turns 1", timeout=60)
terminal(command="tmux new-session -d -s review -x 140 -y 40")
terminal(command="tmux send-keys -t review 'cd /path/to/repo && claude -w pr-review' Enter")
terminal(command="sleep 5 && tmux send-keys -t review Enter") # Trust dialog
terminal(command="tmux send-keys -t review 'Review all changes vs main. Check for bugs, security, race conditions, missing tests.' Enter")
terminal(command="tmux new-session -d -s task1 && tmux send-keys -t task1 'cd ~/project && claude -p \"Fix auth bug\" --allowedTools \"Read,Edit\" --max-turns 10' Enter")
terminal(command="tmux new-session -d -s task2 && tmux send-keys -t task2 'cd ~/project && claude -p \"Write API tests\" --allowedTools \"Read,Write,Bash\" --max-turns 15' Enter")
terminal(command="sleep 30 && for s in task1 task2; do echo \"=== $s ===\"; tmux capture-pane -t $s -p -S -5; done")
→ Full PR review patterns, worktree, parallel workflows: references/pr-workflow.md
# Check if Claude is working or waiting
terminal(command="tmux capture-pane -t dev -p -S -10")
TUI status indicators:
❯ at bottom = waiting for input (done or asking a question)● lines = actively using tools⏵⏵ bypass permissions on = permissions mode in status barctrl+o to expand = tool output was truncatedContext health: Use /context in interactive mode. < 70% is normal, 70-85% consider /compact, > 85% hallucination risk spikes.
| Subcommand | Purpose |
|---|---|
claude | Start interactive REPL |
claude "query" | Start REPL with initial prompt |
claude -p "query" | Print mode (non-interactive, exits when done) |
claude -c | Continue the most recent conversation in this directory |
claude -r "id" | Resume a specific session by ID or name |
claude auth login | Sign in (add --console for API billing, --sso for Enterprise) |
claude auth status | Check login status (returns JSON; --text for human-readable) |
claude mcp add <name> -- <cmd> | Add an MCP server |
claude mcp list | List configured MCP servers |
claude mcp remove <name> | Remove an MCP server |
claude agents | List configured agents |
claude doctor | Run health checks on installation and auto-updater |
claude update / claude upgrade | Update Claude Code to latest version |
claude remote-control | Start server to control Claude from claude.ai or mobile app |
claude setup-token | Set up long-lived auth token (requires subscription) |
claude plugin / claude plugins | Manage Claude Code plugins |
claude auto-mode | Inspect auto mode classifier configuration |
--dangerously-skip-permissions dialog defaults to "No, exit" — send Down then Enter. Print mode skips this.--max-turns is print-mode only — ignored in interactive sessions.--max-budget-usd minimum is ~$0.05 — system prompt cache creation costs this much.--continue finds the most recent for the current working directory.--json-schema needs enough --max-turns — Claude must read files before producing structured output.tmux kill-session -t <name> when done.-p mode, describe the task in natural language./context and /compact.→ Full troubleshooting, cost tips, common errors table: references/troubleshooting.md
-p) for single tasks — cleaner, no dialog handling, structured outputworkdir — keep Claude focused on the right project directory--max-turns in print mode — prevents infinite loops and runaway coststmux capture-pane -t <session> -p -S -50❯ prompt — indicates Claude is waiting for inputtmux kill-session -t <name> to avoid resource leaks--allowedTools — restrict capabilities to what the task actually needs