OpenClaw Waste Audit

name: waste-audit
description: "Find recurring OpenClaw jobs that may be wasting tokens before the waste compounds. Read-only by default. Gives evidence and a copy-paste agent prompt for safe manual verification."
version: 1.8.12
platforms: [linux, macos, windows]
metadata:
hermes:
tags: [openclaw, tokensave, cron, waste, audit, tokens]
related_skills: []

Features

• 🔍 Recurring Waste Detection: Finds recurring OpenClaw jobs that may be wasting tokens.
• 📊 Token Burn Ranking: Ranks likely waste candidates by recurring usage, errors, and delivery signals.
• 🧾 Evidence Summary: Shows why a job was flagged before suggesting action.
• 🛠 Manual Verification Prompt: Gives a copy-paste prompt for safe agent-side verification.
• 🔒 Read-Only Safety: Does not edit, disable, delete, upload, or auto-fix jobs.

Install

For shared OpenClaw agents, install into the global managed skills directory:

openclaw skills install waste-audit --global

To upgrade an existing shared install:

openclaw skills install waste-audit --global --force

Then test with:

check openclaw waste

Activation

Primary activation phrase:

check openclaw waste

Do not use this for general OpenClaw setup, gateway debugging, provider configuration, or normal job management.

Where This Fits

This skill is the OpenClaw-specific starting point for agent token waste auditing.
Use it when you want to inspect recurring OpenClaw jobs for possible token waste.
For broader model routing, sub-agent routing, fallback, retry, or cross-agent runtime analysis, use a separate Agent Routing Waste Audit workflow instead. This skill should not be treated as a generic routing optimizer.

What You Will Get

1. Fix First

Include:

• Job
• Why it looks wasteful
• Evidence
• Confidence: High / Medium / Low
• Recommended manual action

2. Top Waste Candidates

List up to 5 candidates.

For each candidate, include:

• Job
• Waste signal
• Evidence
• Why it matters

3. Manual Verification Prompt

A ready-to-copy prompt for your agent.

Please inspect this recurring OpenClaw job for possible token waste.

Job: <job name>
Reason it was flagged: <short reason>
Evidence: <schedule, runs checked, tokens used, error rate, delivery/summary signal>

Please verify whether this job is still useful.

Do not edit, disable, delete, or mutate anything yet.

First explain:
1. whether this is real waste,
2. what caused it,
3. the safest manual next step,
4. what evidence I should check before changing anything.

Redact secrets and do not expose private payloads.

If any candidate looks important but you are not sure whether it is real waste, send only the "Top Waste Candidates" section to @BeeGeeEth on X. Do not include secrets, API keys, private logs, wallet data, full config files, or production credentials.

Related Next Step

If this audit finds a job where the main issue appears to be model choice, retry behavior, fallback behavior, or sub-agent routing rather than simple recurring job waste, run a separate routing audit before changing any model policy.

The next workflow should inspect:

• whether the task was overpowered by an unnecessarily strong model
• whether retries or fallbacks created hidden waste
• whether the task should remain on a strong model because of coding, review, security, or production risk
• whether a conservative manual routing policy is safer than immediate changes

Safety

• 🔒 Read-only first: Never edit, disable, delete, or auto-fix a job as the first recommendation.
• 🚫 No auto-apply: Always surface evidence and let the owner decide.
• ✏️ Edit as last resort: Changing a job schedule, prompt, or config should only be suggested after manual verification confirms waste.
• 🔑 No secrets exposed: Never show raw tokens, API keys, bot tokens, Telegram chat IDs, private server paths, or raw private payloads in examples or evidence.
• 📋 Redact before sharing: Use <redacted>, <token>, or similar placeholders for anything that could identify a private resource.

Feedback / Free Second Look

If waste-audit flags recurring token waste and you are not sure whether it is real, you can DM me on X: @BeeGeeEth.

Please send only:

• the top 3 flagged jobs
• the evidence summary
• what confused you or what you want checked

Do not send secrets, API keys, private logs, wallet data, full config files, or production credentials.

I'll manually review a few safe examples and use the feedback to improve this skill.