菜单
Guide to writing ai-cli mode files. Use when creating modes for non-interactive Claude Code sessions — preconfigured agent runs with locked-in system prompts, model selection, and prompt wrappers.
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
Structure Claude prompts for clarity and better results using roles, explicit instructions, context, positive framing, and strategic organization. Use when crafting prompts for complex tasks, long documents, tool workflows, or code generation.
Pause agent execution to have the human validate decisions, choose between options, answer freetext, or comment on a document — via the `hl` CLI. Every interaction is a kickoff that returns a job handle immediately; collect the human's answer later with `hl job result`. Use for material design decisions, approval gates, picks between meaningful alternatives, and markdown doc review. Not for trivial yes/no confirmations the agent should decide itself.
Validate technical designs against requirements for coverage, feasibility, and architectural soundness. Use after creating a design.
Validate implementation plans against requirements and design. Use after creating a plan to ensure full coverage and no ambiguities.
Validate requirements documents for EARS compliance, completeness, and consistency. Use after creating requirements.
Design CLIs for humans and LLM agents — subcommand shape, output streams, exit codes, JSON modes, TTY-aware color, structured errors, mutation safety. Use when building or refactoring a CLI, adding machine-readable output, writing --help, deciding stdout vs stderr, or making a tool agent-friendly.
| name | modes-authoring |
| description | Guide to writing ai-cli mode files. Use when creating modes for non-interactive Claude Code sessions — preconfigured agent runs with locked-in system prompts, model selection, and prompt wrappers. |
| user-invocable | false |
| paths | ["**/modes/**/*.md","**/.ai/modes/**/*.md"] |
Modes are markdown files that configure non-interactive Claude Code sessions via the Agent SDK. Each mode defines a system prompt, model, and optional prompt wrapper — producing a preconfigured, fire-and-forget agent.
| Location | Scope | Precedence |
|---|---|---|
.claude/.ai/modes/<name>.md | Project-local | Wins over built-in |
ai-cli/modes/<name>.md | Built-in (bundled) | Fallback |
---
model: opus
system-prompt-mode: append
---
| Field | Required | Values | Default |
|---|---|---|---|
model | Yes | Any valid model ID | — |
system-prompt-mode | No | append, replace | append |
append — mode's system prompt is appended to the default Claude Code system prompt. The agent retains all standard capabilities.replace — mode's system prompt completely replaces the default. Use for fully custom agent personas.Everything after the frontmatter (before ## Prompt Wrapper if present) becomes the system prompt content.
---
model: claude-sonnet-4-5-20250929
system-prompt-mode: append
---
You are a security auditor. Report vulnerabilities only — never edit code.
Focus on: injection, auth bypass, data exposure, SSRF.
A ## Prompt Wrapper heading defines a template applied to the user's -p argument. Use {{prompt}} as the placeholder.
---
model: opus
system-prompt-mode: append
---
You are a systematic debugger. Investigate only — no code changes.
## Prompt Wrapper
Debug the following issue. Diagnose root cause — don't fix.
Issue description:
{{prompt}}
Without a prompt wrapper, the -p argument is passed directly as the user prompt.
# Built-in mode
ai -m review -p "src/auth/"
# Project-local mode
ai -m security-audit -p "check the API layer"
# List all available modes (built-in + local)
ai -l
append for most modes — the agent keeps Claude Code's full toolset; you're just narrowing its focusreplace sparingly — only when you need a completely custom personageneral — sonnet, append, no wrapper. Minimal: "direct, efficient assistant"debug — opus, append, wrapper frames input as "Issue description". Multi-phase methodology in system prompt.review — opus, append, wrapper frames input as review target. Detailed concern taxonomy in system prompt.Modes are callable from hooks, scripts, and other tools:
# In a pre-commit hook
ai -m validate-commit -p "$(git diff --cached)"
# In a CI script
ai -m review -p "$(git diff origin/main...HEAD)"