Agent Tools

Reusable Agent Skills, plus runtime capabilities (statusline, provider usage, vision) for Codex, Claude Code, and opencode.

Requires Node.js >= 22.

中文

Skills

Install

# List available skills
npx -y skills@latest add kairyou/agent-tools --list

# Install globally (pass one or more names after --skill)
npx -y skills@latest add kairyou/agent-tools --skill <name...> -g -y

at-commit

Generate a Conventional Commits message from staged changes and wait for user confirmation before committing.

npx -y skills@latest add kairyou/agent-tools --skill at-commit -g -y

Usage:

• /at-commit [<language>] — language for the commit description (Conventional Commits tokens stay in English)

at-review

Review local changes or a hosted PR/MR for correctness bugs, regressions, convention violations, and high-value cleanup findings.

npx -y skills@latest add kairyou/agent-tools --skill at-review -g -y

Usage:

• /at-review [--fix] [<pr-or-mr-url|branch|path>] — reports review findings; private hosted targets require locally available read access, and --fix applies fixes only to a matching working tree

at-simplify

Refactor changes to reduce duplication, lower complexity, and improve code quality.

npx -y skills@latest add kairyou/agent-tools --skill at-simplify -g -y

Usage:

• /at-simplify [<pr|branch|path>]

at-zentao

Work ZenTao (禅道) bugs/tasks end to end, manage task status and hours, or read linked story context; asks before committing and before writing back.

npx -y skills@latest add kairyou/agent-tools --skill at-zentao -g -y

Usage:

• /at-zentao bugs — list bugs assigned to you (the configured account); pick one or several (several = batch mode)
• /at-zentao tasks — same, for tasks
• /at-zentao bug <id> — work a specific bug
• /at-zentao task <id> [request] — work a task, start/pause/resume it, or log hours with a natural-language request
• /at-zentao story <id> — read requirement scope and acceptance context without writing back

Config: ~/.agent-tools/config.jsonc → "zentao": { "url", "account", "password" }. The password can be entered directly or referenced with "password": { "env": "ZENTAO_PASSWORD" }; never paste credentials or tokens into chat.

Capabilities

Runtime capabilities, installed per agent:

npx -y @kairyou/agent-tools@latest <capability> -a <agent...>
# npx -y github:kairyou/agent-tools <capability> -a <agent...>  # Or the latest code from GitHub (needs Git)

--dry-run previews, --uninstall removes the capability from the agent, and re-running the install command updates. The installer only touches config entries it wrote itself, and config.jsonc updates only add missing default keys without touching your edits or comments.

Statusline

npx -y @kairyou/agent-tools@latest statusline -a claude

The installer writes statusLine to ~/.claude/settings.json. Example output:

# Pick and order the fields via statusline.fields in ~/.agent-tools/config.jsonc:
⎇ main | Opus 4.8 | 5h 7% ⟳2h54m | w 41% ⟳3d1h

# With a compatible API relay, quota info is shown too:
⎇ main | Opus 4.8 | balance $362 | today $61.7 | 30d $566

Here 5h and w are Claude's rolling usage windows and ⟳ is the reset countdown; see Provider usage below for relay quota compatibility and configuration.

Provider usage

Shows API relay / gateway balance and quota inside the agent. Supports Sub2API, One API (including OneHub and DoneHub), New API, Claude Code Hub, OpenRouter, and Command Code; compatibility depends on the gateway version and enabled usage endpoints.

npx -y @kairyou/agent-tools@latest usage -a claude codex opencode

• Claude Code — invoke /at-usage to show the current usage.
• Codex — run /hooks once after installation to approve it. The Codex CLI displays usage automatically; clients that hide hook output can use $at-usage.
• OpenCode — usage refreshes when the session goes idle and appears as a toast; /at-usage shows it on demand. Restart opencode after installing or updating.

The relay endpoint is auto-discovered from the existing Codex and Claude Code configuration; official (non-relay) endpoints are skipped. If it reports Provider usage is unavailable., set PROVIDER_USAGE_BASE_URL and PROVIDER_USAGE_API_KEY to override the endpoint and key. Configure providerUsage in ~/.agent-tools/config.jsonc only when needed:

{
  "providerUsage": {
    // auto | sub2api | openai-compatible | one-api | one-hub |
    // done-hub | new-api | claude-code-hub | openrouter | commandcode | <custom-route-id>
    "preset": "auto", // auto-detect, or select one protocol listed above
    "days": 30,       // how many recent days of spend to count
    "debug": false    // true: log probes to ~/.agent-tools/logs/usage-debug.log
  }
}

Keep preset set to auto for automatic detection. Select a specific protocol only when you know which usage endpoint the gateway exposes; a configured custom route id is also accepted. Command Code usage relies on a version-sensitive endpoint that is not part of its documented Provider API.

Output examples:

# Plan limits (sub2api / openai-compatible).
D $0.0/$100 | W $0.0/$300 ⟳3d1h | Exp 07-08

# Multi-window limits (claude-code-hub).
5h $2.1/$10.0 | D $8.0/$20.0 | T $19.0/$100 | Exp 08-31

# Balance and usage (one-api / one-hub / done-hub / new-api / openrouter).
balance $15.0 | used $5.0/$20.0 | ⟳3d1h

# Wallet and recent spend (sub2api).
balance $362 | today $61.7 | 30d $566

Fields: 5h/D/W/M/T are five-hour/daily/weekly/monthly/total spend against limits; ⟳ is the time until a known limit reset; Exp is the plan expiry; balance is wallet credit; used is consumed credit; today and 30d are today's and the last 30 days' API spend. A reset countdown is shown only when the gateway returns enough timing information to determine it reliably.

Custom gateway routes

Gateways not covered by the built-in probes can use a custom route. See Custom gateway routes.

Vision (cross-model image understanding)

Lets a main model that cannot see images use a multimodal model to analyze error screenshots, implement UI from design mockups, and inspect bug-report screenshots.

Install

npx -y @kairyou/agent-tools@latest vision -a claude codex opencode

Uninstalling keeps your vision provider config by default.

Configure

~/.agent-tools/config.jsonc is the only config entry point:

{
  "vision": {
    "provider": "openai-compatible",       // or "anthropic-compatible"
    "baseUrl": "https://gateway.example.com/v1",  // anthropic-compatible: gateway root, /v1/messages is appended
    "model": "internal-vlm",
    "apiKey": { "env": "OPENAI_API_KEY" }  // read from the OPENAI_API_KEY environment variable
    // , "timeoutMs": 30000, "maxImageBytes": 20971520, "maxOutputTokens": 8192
    // , "maxConcurrentRequests": 2, "maxRequestsPerMinute": 30
  }
}

apiKey can be a key string such as "apiKey": "sk-...", or an environment variable reference such as { "env": "OPENAI_API_KEY" }. Omit it if your gateway requires no key. Prefer an environment variable to avoid storing the key in the config file.

Use

Pass a local image path or URL in your message. If the main model cannot accept pasted images, save the image first and pass its file path instead.

To diagnose the provider setup or test recognition quality manually:

npx -y @kairyou/agent-tools@latest inspect-image <path|url> -q "What are the navbar background color and height?"

Extras

See extras for situational content.

FAQ

Why does installing skills report PromptScript as unsupported?

PromptScript does not support global skill installation means that the PromptScript agent does not support global installation. It does not affect other agents and can be ignored. See skills issue #1352.

How do I install skills without GitHub access?

Install from the npm package instead (swap at-commit for the skills you want):

cd "$(mktemp -d)" && tar -xf "$(npm pack @kairyou/agent-tools --silent)" && npx -y skills@latest add ./package --skill at-commit -g -y

Want to contribute?

See the repository structure.

Acknowledgements

• at-commit draws on commit-message generation ideas from OpenCommit and GitLens, reimplemented for an Agent Skill workflow.
• at-review and at-simplify are installable Agent Skills derived and adapted from Claude Code's built-in code-review and simplify workflow prompts. Automated upstream tracking uses versioned prompt data from tweakcc; human-readable prompt history comes from claude-code-system-prompts.