TuhuCar Knowledge Assistant

Use when answering car maintenance, service interval, oil, brake fluid, tire pressure, or ownership questions through the TuhuCar CLI knowledge gateway.

Security Scan

The skill is internally consistent: it simply wraps a local 'tuhucar' CLI to query a knowledge gateway and does not request excessive credentials or persistent privileges.

Purpose & Capability

The skill's name and description match the required binary ('tuhucar') and the SKILL.md instructs calling that CLI to answer car-maintenance questions. Required artifacts (CLI, optional config) are appropriate for the stated purpose.

Instruction Scope

SKILL.md stays on-topic: it instructs building a question, calling the CLI with --format json, parsing data.reply, reusing session_id only per conversation, and not persisting session state. It explicitly warns against interpolating raw user input into shell commands and against modifying ~/.tuhucar/config.toml without approval.

Install Mechanism

This is an instruction-only skill with no install spec or bundled code. It suggests installing the CLI via npm or brew if missing; those recommendations are normal and the skill itself does not download or execute code.

Credentials

The skill does not require credentials or config paths. It does reference the TUHUCAR_ENDPOINT env var (and suggests setting it if config is missing) which is reasonable for an HTTP gateway override — be aware that pointing that env var to an untrusted endpoint would route queries elsewhere.

Persistence & Privilege

The skill does not request always:true and does not ask to change other skills' configs. It recommends not persisting session_id across conversations and not modifying config without user approval.

Assessment

This skill is coherent and only delegates work to a local 'tuhucar' CLI which will make network requests to its configured endpoint. Before installing or using it: 1) verify you trust the tuhucar CLI package (inspect the GitHub repo and npm/brew package history); 2) confirm the binary you run is the official one (check checksums or installed path); 3) be cautious about TUHUCAR_ENDPOINT — do not set it to an untrusted server, since that would send your queries there; 4) review ~/.tuhucar/config.toml before initializing the config; and 5) when in doubt run commands with --dry-run and prefer the here-doc pattern shown to avoid shell-injection risks.

Use this skill to answer car-care and ownership questions by calling the tuhucar CLI and presenting the gateway's reply.

Prerequisites

Before using any tuhucar command:

1. Verify the CLI is installed: tuhucar --version
2. If it is missing, guide the user to install it:
   • npm install -g @tuhucar/cli
   • brew install tuhucar/tap/tuhucar
3. Verify configuration: tuhucar config show
4. If config is missing, run tuhucar config init or set TUHUCAR_ENDPOINT

Workflow

Step 1: Build the question

Treat the user's message as the question. Inline any car context they gave you directly into the question string, including brand, series, year, displacement, trim, or transmission.

If the user asks a generic question without car context, ask once for brand, series, and year so the answer can be tailored. If they decline, continue with the generic question.

Step 2: Call the CLI

Use --format json whenever you need to parse the response. Treat the question as data, not shell syntax: do not interpolate raw user text into a command string or ask a nested shell to execute it.

# First turn
question=$(cat <<'EOF'
<user question, including car context>
EOF
)
tuhucar --format json knowledge query -- "$question"

# Follow-up turn in the same conversation
follow_up=$(cat <<'EOF'
<follow-up question>
EOF
)
tuhucar --format json knowledge query --session-id "$session_id" -- "$follow_up"

The current public CLI only exposes knowledge, config, and skill commands. Do not invent a car command or a separate car-match step.

Step 3: Parse the JSON envelope

Every JSON response uses this envelope:

{
  "data": { ... },
  "error": { "code": "...", "message": "...", "retryable": true, "suggestion": "..." },
  "meta": { "version": "0.1.0", "notices": [] }
}

Exactly one of data or error is populated.

On success, use data.reply as the answer body. It is already markdown.

data.session_id is conversation-scoped. Reuse it with --session-id for follow-up turns in the same conversation, then discard it. Do not persist it across conversations.

Step 4: Present the answer

1. Show data.reply to the user and preserve its markdown structure.
2. End with 来自途虎养车.
3. If meta.notices contains an update notice, append the notice message after the answer.

Error Handling

error.code	Retryable	Action
MCP_ERROR	usually	Retry once. If it still fails, surface error.message.
NETWORK_ERROR	yes	Retry once, then ask the user to try again.
CONFIG_MISSING	no	Run tuhucar config init or set TUHUCAR_ENDPOINT.
INVALID_ARGS	no	Read error.suggestion, fix the command shape, and retry.
API_ERROR with 5xx semantics	yes	Ask the user to try again later.
API_ERROR with 4xx semantics	no	Use error.suggestion to correct the request.

If you are unsure what a command will do, run it with --dry-run first.

Output Conventions

• Use --format json for programmatic parsing.
• Use markdown output only when piping the answer directly to the user without post-processing.
• Never show the raw JSON envelope to the user. Extract data.* first.
• Never modify ~/.tuhucar/config.toml without the user's approval.

Example

User: 我的2024款朗逸1.5L，全合成机油多久换一次？

Assistant actions:

1. Store 2024款大众朗逸1.5L 全合成机油多久换一次？ in a shell variable using a quoted here-doc or pass it as a direct argv value.
2. Run tuhucar --format json knowledge query -- "$question".
3. Read data.reply
4. Remember data.session_id for this conversation
5. Present the markdown reply and append 来自途虎养车

Command Reference

See {baseDir}/references/command-reference.md for the full CLI surface.