Claw Runaway Loop Detector

Detects potential infinite loops and runaway token usage risks in Claw AI workflows by analyzing retry, recursion, and termination conditions.

Security Scan

The skill is an instruction-only analyzer that is internally consistent with its stated purpose and does not request credentials, install code, or access unrelated resources.

Purpose & Capability

Name and description claim to detect runaway loops in Claw workflows; the SKILL.md contains only analysis instructions for that exact task and does not request unrelated capabilities or credentials.

Instruction Scope

Runtime instructions are limited to analyzing provided workflows/prompts for loop triggers, termination conditions, retries, and estimating token amplification. The SKILL.md does not instruct the agent to read system files, environment variables, or transmit data to external endpoints beyond normal model invocation.

Install Mechanism

No install spec and no code files — instruction-only skill. Nothing is written to disk or downloaded as part of the skill.

Credentials

No required environment variables, credentials, or config paths are declared or referenced in the instructions; requested access is proportionate to an analysis-only tool.

Persistence & Privilege

Flags indicate default behavior (not always:true). The skill does not request persistent system-wide changes or access to other skills' configurations.

Assessment

This is an instruction-only analyzer and appears coherent with its purpose. Before using it, avoid pasting secrets or private credentials into the workflow you submit for analysis because the content will be processed by the agent/model. Treat its findings as advisory — it will suggest safeguards but will not automatically enforce them. If you need automated enforcement (e.g., runtime token caps or circuit breakers), use platform-level controls or instrumented code in addition to this analyzer. If you want stronger assurance about what the skill does with your data, ask the publisher for source code or a hosted implementation detail; absence of install code here means the analysis logic is entirely driven by the SKILL.md and the model invocation context.

Detect infinite loops and runaway token usage risks in Claw AI workflows.

AI agents built with Claw can sometimes enter unintended loops when tools retry repeatedly, tasks recurse, or prompts encourage repeated exploration. These loops can rapidly generate massive token usage and unexpectedly high API costs.

This skill analyzes a Claw workflow or prompt and identifies patterns that could lead to runaway loops.

---

Example input

Agent searches the web for competitor pricing and keeps retrying the search tool until it finds reliable data.

or

Agent reviews documents, summarizes them, and retries the summarization tool until the result improves.

---

How to use

Paste your Claw workflow, task description, or agent prompt.

The detector will analyze the workflow and identify potential loop risks that could lead to runaway token usage.

---

Analysis instructions

You are an AI agent reliability and cost analysis expert.

Analyze the provided Claw workflow or prompt and detect patterns that could cause an agent to repeatedly execute tasks, retry tools, or recursively call itself.

Focus on identifying conditions that could create runaway loops and excessive token usage.

---

Analysis process

When analyzing the workflow:

1. Identify potential loop triggers.
2. Check whether termination conditions exist.
3. Evaluate retry or recursion patterns.
4. Estimate possible token amplification if the loop occurs.
5. Assign an overall runaway loop risk level.

---

Common runaway loop patterns

The detector looks for common patterns including:

• Recursive agent calls
• Unbounded retry logic
• Prompts encouraging repeated exploration
• Tool retry chains
• Missing termination conditions
• Long reasoning chains without limits

---

Risk level guidelines

Low The workflow has clear termination conditions and limited retries.

Medium The workflow contains retry logic or open-ended exploration but appears somewhat bounded.

High The workflow may repeatedly call tools, recurse tasks, or lacks clear stopping conditions.

---

Output format

Output must follow this structure:

Runaway Loop Risk Analysis

Risk level: (Low / Medium / High)

Loop triggers detected: (list)

Estimated token amplification: (short explanation)

Estimated token impact: (range)

Recommended safeguards: (list)

---

Example Output

Runaway Loop Risk Analysis

Risk level: High

Loop triggers detected:

• Tool retry pattern without limit
• Prompt encourages repeated searching

Estimated token amplification: The agent may repeatedly call the search tool until success, causing exponential token growth.

Estimated token impact: 10k – 50k tokens per run if the loop occurs.

Recommended safeguards:

• Limit tool retries to 3
• Add explicit termination conditions
• Restrict recursive agent calls
• Set maximum reasoning steps

---

Why this matters

Runaway loops are one of the most common causes of unexpected AI API costs.

A single workflow mistake can cause an agent to repeatedly call tools or itself, generating thousands of tokens per minute.

Detecting these patterns early helps prevent runaway token usage before deploying agents to production.

---

Tip

In production environments, it is useful to combine loop detection with automated safeguards such as token limits, retry limits, and behavior monitoring.

Gateways such as ClawFirewall can automatically stop abnormal loops and protect your API budget.

---

License

MIT-0