Clawprobe

Monitor OpenClaw agent health, token usage, API cost, and context window in real time. Use when you need to check your own status, inspect context utilizatio...

Security Scan

The skill's instructions and requested capabilities match its stated purpose (a CLI-based observability tool), but it relies on installing an external npm package and running a background daemon which you should verify before installing.

Purpose & Capability

Name/description align with the commands and outputs in SKILL.md: all commands relate to monitoring agent health, token/cost usage, context utilization, suggestions, and compaction events. The skill does not ask for unrelated credentials, binaries, or config paths.

Instruction Scope

Instructions only tell the agent to install and run a CLI (clawprobe) and to parse its JSON output. They do not instruct reading unrelated system files or environment variables, but they do instruct starting a daemon (clawprobe start) which may access agent internals or local config once installed—this access is implicit in the CLI and not described in the SKILL.md.

Install Mechanism

There is no install spec in the skill metadata; the README suggests using npm install -g clawprobe. Installing from npm is a common pattern but can execute arbitrary install scripts and will pull code from the public registry. The SKILL.md does not pin a version or link to a specific, audited release artifact.

Credentials

The skill declares no required environment variables, credentials, or config paths, which is consistent with its documentation. Be aware that the installed CLI/daemon could read local agent config or tokens when running even though the skill metadata doesn't request those.

Persistence & Privilege

always is false and model invocation is allowed (normal). However, the instructions encourage running a long‑lived daemon (clawprobe start), which introduces persistence on the host outside the skill metadata—monitor and audit the daemon if you install it.

Assessment

This skill is an instruction-only wrapper that expects you to install an external npm package (clawprobe) and run its CLI/daemon. That pattern is coherent with its purpose but carries the usual risks of installing third-party packages: npm installs can run arbitrary code and a daemon may access local agent config or tokens. Before installing, verify the npm package and repository (check the GitHub homepage, author, recent commits, and published package contents), prefer a pinned version, inspect postinstall scripts in the package, consider installing in an isolated environment or container, and monitor the daemon's network and file activity. If you maintain sensitive agent credentials, confirm the tool's source and exact permissions before running `clawprobe start`.

Real-time observability for OpenClaw agents. Check health, token usage, API cost, context window, and optimization suggestions — all from the CLI.

Install

npm install -g clawprobe
clawprobe start

Core commands

Command	What it does
clawprobe status --json	Full health snapshot (session, model, context %, cost, alerts)
clawprobe cost --week --json	API cost breakdown for the past week
clawprobe session --json	Current session token timeline, turn by turn
clawprobe context --json	Context window utilization and truncation warnings
clawprobe suggest --json	Active optimization alerts with ruleId and action
clawprobe compacts --json	List recent compaction events and lost content
clawprobe top	Live auto-refreshing dashboard (human use)

Recommended self-check workflow

Run at session start or before any expensive task:

STATUS=$(clawprobe status --json)
UTIL=$(echo $STATUS | node -e "process.stdin.resume();let d='';process.stdin.on('data',c=>d+=c);process.stdin.on('end',()=>console.log(JSON.parse(d).utilizationPct))")
echo "Context: $UTIL%"

Or in a single call:

clawprobe status --json

Key fields to act on:

• utilizationPct ≥ 85 → compact soon or start a fresh session
• suggestions[].ruleId === "tools-truncation" → TOOLS.md is being cut off
• suggestions[].ruleId === "cost-spike" → today's spend is unusually high
• daemonRunning: false → run clawprobe start to enable monitoring

Handling suggestions programmatically

# List active suggestions
clawprobe suggest --json

# Dismiss a noisy rule
clawprobe suggest --dismiss memory-bloat --json
# → { "ok": true, "dismissed": "memory-bloat" }

Schema introspection

When you need to know exact field names for any command:

clawprobe schema status    # full field spec for status --json
clawprobe schema cost      # full field spec for cost --json
clawprobe schema           # list all available schemas

Error format

All errors under --json are structured — never colored text:

{ "ok": false, "error": "no_active_session", "message": "No active session found" }

Exit code is always 1 on error.

Cost note

Cost figures are estimates based on public model pricing. Verify exact amounts with your model provider's billing dashboard.