SKILLEMALL.ai
GitHub ↗ RU Sign in

BD browser-harness

用 LLM 友好的方式控制用户已登录的真实 Chrome（CDP）。一行命令在当前标签页跑 JS、点击、滚动、截图、读 DOM、填表、上传文件——共享 cookie/session/登录态，跨 Python 与 TypeScript Agent 操作同一个浏览器。基于 browser-use/browser-harness（Python 守护进程）+ browser-harness-ts（TS 客户端 + bhts CLI）。HIGH-RISK 能力：默认 sensitive-deny（银行/邮箱/内网/admin 模式拒绝写操作）、可选 BH_PUBLIC_ONLY 硬隔离、metadata-only 审计日志、subprocess 隔离不做 in-process import、上游版本精确钉死。

ClawHub Agent Skills author: PING SI v1.0.5 MIT-0 9 files · 1 script body ≈ 2 410 tokens Open the sourceclawhub.ai↗ analyzed 3 d ago

As a process D 48/100 · Unfinished process — weak spots: result and completion, when it triggers, inputs and preconditions

ProcedureSoftware developmentAI and agentstype and topics are labelled automatically from the skill text
Runs in: OpenClaw Claude / Claude Code ChatGPT / Codex Gemini CLI Cursor GitHub Copilot / VS Code Qwen Code Kimi Code CLI Z.ai (GLM) DeepSeek (Deep Code) Mistral Vibe LangChain (deepagents) Goose OpenCode
JSON
Technical rating
B
82/100
safety, quality, tests
Safety 60%
79
Quality 40%
86
Run on models
none yet
Process rating
D
48/100
Unfinished process
Result and completion w 14
0
Inputs and preconditions w 11
0
When it triggers w 12
20
the three weakest of ten parameters · all ten

What is at stake

Medium-severity findings: the skill is probably honest, but read what alarmed the scanner.

Dangerous commands medium severity

Below is the worst case for this category. The finding here is medium: the guard saw a sign, not a proof.

If you install

The skill contains commands that delete files, rewrite disks or execute code fetched from the network. The agent may run them without asking if it believes the instructions require it.

For the author

Replace destructive commands with safe equivalents that ask for confirmation, scope them to one folder, and stop piping curl into a shell: pin a version and a checksum.

How to improve

    For the model run — optional
    • Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
    • A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.

    Guard findings · 5

    ✓ No critical or high findings

    Medium and low: 5
    • medium Dangerous commands cmd-shell-rc scripts/run.sh:140
      Writes to a shell startup file
      printf '    echo '\''export PATH="$(npm prefix -g)/bin:$PATH"'\'' >> ~/.zshrc && source ~/.zshrc\n'
    • medium Dangerous commands cmd-shell-rc scripts/run.sh:164
      Writes to a shell startup file
      printf '    echo '\''export PATH="$HOME/.local/bin:$PATH"'\'' >> ~/.zshrc && source ~/.zshrc\n'
    • medium Dangerous commands cmd-pipe-to-shell-known-host setup.md:25
      Pipe-to-shell installer from a well-known host (still executes remote code)
      curl -LsSf https://astral.sh/uv/install.sh | sh    # 官方
    • medium Dangerous commands cmd-shell-rc setup.md:139
      Writes to a shell startup file
      echo 'export PATH="$(npm prefix -g)/bin:$PATH"' >> ~/.zshrc && source ~/.zshrc
    • low Dangerous commands cmd-pipe-to-shell-known-host scripts/run.sh:120
      Pipe-to-shell installer from a well-known host (still executes remote code) (string literal in code, not executed)
      printf "    curl -LsSf https://astral.sh/uv/install.sh | sh\n"
      code literal

    Files scanned: 9. Evidence is masked. Grey chips explain why severity was lowered.

    Against the Agent Skills spec

    • note frontmatter-key unknown frontmatter key "requiredEnvVars"

    Process rating: all ten parameters 48/100

    • 0Result and completion. Does not say what the result is
    • 0Inputs and preconditions. Does not say what the process needs to start
    • 20When it triggers. No condition that starts the skill
    • 40Consistency. Frontmatter name (browser-harness) differs from the folder (browser-harness-skill)
    • 50Failures and branches. 0 branches, has a failure section
    • 60Tools and files. Uses tools (python, node) that frontmatter does not declare
    • 100Steps. 42 steps
    • 100Execution cost. Instruction body is 2410 tokens
    • 100Running it twice. No mutating operations
    • 100Progress reporting. Reports progress
    • low The response is described with custom markup (19 tags): a typed call is more reliable

    Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.

    Quality signals

    • +5Description has no quoted example phrases that should trigger the skill
    • +4Description does not say when NOT to use the skill (false activations)
    • +3Output format is not stated: the model decides each time
    • +1No license
    • +2Single-language instructions
    • +3Description length 354: enough signal without eating the budget
    • +4Structure: 23 headings
    • +3Step-by-step instructions: 42 items
    • +4Has examples (8 code blocks)
    • +3All 1 scripts are documented

    Quality base 70; lint remarks subtract, signals add up to 100. Result: 86.

    External checks

    ClawHub: suspicious
    This skill is openly designed to let an agent control a real logged-in Chrome browser, but that high-impact access needs careful review before installation.
    LLM: suspicious (high) · VirusTotal: · 29 May 2026