Response Speed Test

Measures OpenClaw system response times by testing stages, providing millisecond precision metrics like TTFT, TPS, and total response time.

Security Scan

The code and CLI match the stated purpose (measuring OpenClaw response timings), but the skill reads environment variables and local files/paths that are not declared in the manifest or SKILL.md — an inconsistency you should review before installing.

Purpose & Capability

Name, description, CLI, core measurement engine, probes and reporters are coherent: files implement timing probes for Gateway, Session, Memory, and LLM and generate reports as described.

Instruction Scope

SKILL.md instructs how to run the CLI but does not explicitly mention that probes will read environment variables and local filesystem paths. The code will access memory and state directories (e.g. SOUL_MEMORY_PATH default /root/.openclaw/workspace/memory and OPENCLAW_STATE_DIR default /root/.openclaw/state) and may read index.json; SKILL.md does not document these file reads.

Install Mechanism

No install spec is declared (lower risk), but a pyproject.toml and packaging metadata are present and SKILL.md suggests using 'clawhub install'. There are no external downloads or installer URLs; install risk is low but the packaging vs. registry metadata mismatch is worth noting.

Credentials

The registry lists no required env vars, yet code reads several environment variables at runtime (OPENCLAW_GATEWAY_URL, LLM_API_URL, LLM_MODEL, SOUL_MEMORY_PATH, OPENCLAW_STATE_DIR). These are relevant to the skill's function, but the manifest/README/SKILL.md do not declare them — a transparency gap. Also the skill will read local files (memory .md files and index.json) which can contain sensitive data; the code does not transmit data externally, but prompts and metadata could include snippets.

Persistence & Privilege

Skill does not request always:true, does not attempt to modify other skills or system-wide agent settings, and has no autonomous elevated persistence. It only provides CLI and library code.

What to consider before installing

This skill appears to implement the described timing/benchmark functionality, but review before installing: 1) Note that it reads environment variables (OPENCLAW_GATEWAY_URL, LLM_API_URL, LLM_MODEL, SOUL_MEMORY_PATH, OPENCLAW_STATE_DIR) even though the registry declares none — set them intentionally or leave defaults. 2) It inspects local paths (defaulting to /root/.openclaw/...), including listing .md files and reading index.json; if those directories contain sensitive data, run the skill in a restricted/sandboxed environment or point SOUL_MEMORY_PATH/OPENCLAW_STATE_DIR to a safe location. 3) There are no network calls in the current code, but LLM/Gateway probes are written to be extended to call APIs — verify any modifications that add network I/O. 4) Prefer installing from a trusted source (the pyproject references a GitHub repo); check the repository URL and author before running 'clawhub install' or running code locally. If you want, I can list the exact lines that read environment variables and file paths so you can audit them.

Response Speed Test Skill

精確測量 OpenClaw 系統響應速度的 Skill，提供毫秒級時間測量和詳細報告生成。

安裝

clawhub install response-speed-test

使用

單次測試

response-speed-test run

基準測試

response-speed-test benchmark --iterations 10

生成報告

response-speed-test report --format markdown

測量階段

階段	代碼	說明
T0	MESSAGE_SENT	用戶發送消息
T1	GATEWAY_RECEIVE	Gateway 收到消息
T2	SESSION_DISPATCH	Session 分發
T3	MEMORY_LOAD	Memory 載入
T4	LLM_REQUEST_START	LLM 請求開始
T5	LLM_FIRST_TOKEN	第一個 Token (TTFT)
T6	LLM_RESPONSE_COMPLETE	LLM 響應完成
T7	MESSAGE_SENT	消息發送完成

關鍵指標

• TTFT: Time To First Token
• TPS: Tokens Per Second
• Total Response Time

配置

在 HEARTBEAT.md 或對話中使用：

- response-speed-test: run benchmark every 6 hours